","msg_from_op":true,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On 04/05/2017 07:23 AM, Michael Paquier wrote:\n> fore\n>\n> On Wed, Apr 5, 2017 at 7:05 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n>> I will continue tomorrow, but I wanted to report on what I've done so far.\n>> Attached is a new patch version, quite heavily modified. Notable changes so\n>> far:\n>\n> Great, thanks!\n>\n>> * Use Unicode codepoints, rather than UTF-8 bytes packed in a 32-bit ints.\n>> IMHO this makes the tables easier to read (to a human), and they are also\n>> packed slightly more tightly (see next two points), as you can fit more\n>> codepoints in a 16-bit integer.\n>\n> Using directly codepoints is not much consistent with the existing\n> backend, but for the sake of packing things more, OK.\n\nOh, I see, we already have similar functions in wchar.c. \nunicode_to_utf8() and utf8_to_unicode(). We should probably move those \nto src/common, rather than re-invent the wheel.\n\n> pg_utf8_islegal() and pg_utf_mblen() should as well be moved in their\n> own file I think, and wchar.c can use that.\n\nYeah..\n\n>> * The list of characters excluded from recomposition is currently hard-coded\n>> in utf_norm_generate.pl. However, that list is available in machine-readable\n>> format, in file CompositionExclusions.txt. Since we're reading most of the\n>> data from UnicodeData.txt, would be good to read the exclusion table from a\n>> file, too.\n>\n> Ouch. Those are present here...\n> http://www.unicode.org/reports/tr41/tr41-19.html#Exclusions\n> Definitely it makes more sense to read them from a file.\n\nDid that.\n\n>> * SASLPrep specifies normalization form KC, but it also specifies that some\n>> characters are mapped to space or nothing. Should do those mappings, too.\n>\n> Ah, right. Those ones are here:\n> https://tools.ietf.org/html/rfc3454#appendix-B.1\n\nYep.\n\n\nAttached is a new version. Notable changes since yesterday:\n\n* Implemented the rest of the SASLPrep, mapping some characters to \nspaces, leaving out others, and checking for prohibited characters and \nbidirectional strings.\n\n* Moved things around. There's now a separate directory, \nsrc/common/unicode, which contains the perl scripts and the test code. \nThose are not needed to build from source, as the pre-generated tables \nare put in src/include/common. Similar to the scripts in \nsrc/backend/utils/mb/Unicode, really.\n\n* Renamed many things from utf_* to unicode_*, since they don't deal \nwith utf-8 input anymore.\n\n\nThis is starting to shape up, but still some cleanup work to do. I will \ncontinue tomorrow..\n\n- Heikki\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 5 Apr 2017 19:33:13 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On Thu, Apr 6, 2017 at 1:33 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n> Attached is a new version. Notable changes since yesterday:\n>\n> * Implemented the rest of the SASLPrep, mapping some characters to spaces,\n> leaving out others, and checking for prohibited characters and bidirectional\n> strings.\n>\n> * Moved things around. There's now a separate directory, src/common/unicode,\n> which contains the perl scripts and the test code. Those are not needed to\n> build from source, as the pre-generated tables are put in\n> src/include/common. Similar to the scripts in src/backend/utils/mb/Unicode,\n> really.\n>\n> * Renamed many things from utf_* to unicode_*, since they don't deal with\n> utf-8 input anymore.\n>\n> This is starting to shape up, but still some cleanup work to do. I will\n> continue tomorrow..\n\nThanks for the new patch, that's looking nice. Now I was not able to\ncompile it as saslprep.h is missing from what you have sent...\n\nThere is for example this portion in the new tables:\n+static const Codepoint prohibited_output_chars[] =\n+{\n+ 0xD800, 0xF8FF, /* C.3, C.5 */\n\n ----- Start Table C.5 -----\n D800-DFFF; [SURROGATE CODES]\n ----- End Table C.5 -----\nThis indicates a range of values. Wouldn't it be better to split this\ntable in two, one for the range of codepoints and another one with the\nsingle entries?\n\n+ 0x1D173, 0x1D17A, /* C.2.2 */\nThis is for musical symbols. It seems to me that checking for a range\nis what is intended.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 6 Apr 2017 16:32:08 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":true,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"Another version attached.\n\nI think this is now in committable state, but there's been a lot of \nsmall changes here and there, so please have one more look over it if \nyou have a chance. I'm planning to push this tomorrow, after sleeping on it.\n\nThe code-organization issue with the utf8 functions, pg_utf_mblen, \npg_utf8_islegal, unicode_to_utf8, and utf8_to_unicode, is still kind of \nunresolved. The way I have it in this version is that the functions are \nin wchar.c, like they've always been, and saslprep.c uses them from \nthere. This is ugly from a code organization point of view. as we now \nhave functions in src/common depending on functions in \nsrc/backend/utils/mb/wchar.c. I think wchar.c deserves some \nre-organization, as most of the functions there are also used in \nfrontend code. Perhaps it should be moved to src/common as whole.\n\nLibpq already compiles with wchar.c, so all those functions that \nsrc/common/saslprep.c requires are already available in libpq. So I \ndon't think that's urgent, but something we nevertheless ought to clean \nup at some point.\n\n\nAnother thing I'd like some more eyes on, is how this will work with \nencodings other than UTF-8. We will now try to normalize the password as \nif it was in UTF-8, even if it isn't. That's OK as long as we're \nconsistent about it, but there is one worrisome scenario: what if the \nuser's password consists mostly of characters, that when interpreted as \nUTF-8, are in the list of ignored characters. IOW, is it realistic that \na user might have a password in a non-UTF-8 encoding, that gets silently \nmangled into something much shorter? I think that's highly unlikely, but \ncan anyone come up with a plausible example of that?\n\n- Heikki\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 6 Apr 2017 19:59:16 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"(sorry, I didn't notice your email until after I just sent version 4!)\n\nOn 04/06/2017 10:32 AM, Michael Paquier wrote:\n> On Thu, Apr 6, 2017 at 1:33 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n>> Attached is a new version. Notable changes since yesterday:\n>>\n>> * Implemented the rest of the SASLPrep, mapping some characters to spaces,\n>> leaving out others, and checking for prohibited characters and bidirectional\n>> strings.\n>>\n>> * Moved things around. There's now a separate directory, src/common/unicode,\n>> which contains the perl scripts and the test code. Those are not needed to\n>> build from source, as the pre-generated tables are put in\n>> src/include/common. Similar to the scripts in src/backend/utils/mb/Unicode,\n>> really.\n>>\n>> * Renamed many things from utf_* to unicode_*, since they don't deal with\n>> utf-8 input anymore.\n>>\n>> This is starting to shape up, but still some cleanup work to do. I will\n>> continue tomorrow..\n>\n> Thanks for the new patch, that's looking nice. Now I was not able to\n> compile it as saslprep.h is missing from what you have sent...\n\nD'oh. Here's a new version, with saslprep.h included.\n\n> There is for example this portion in the new tables:\n> +static const Codepoint prohibited_output_chars[] =\n> +{\n> + 0xD800, 0xF8FF, /* C.3, C.5 */\n>\n> ----- Start Table C.5 -----\n> D800-DFFF; [SURROGATE CODES]\n> ----- End Table C.5 -----\n> This indicates a range of values. Wouldn't it be better to split this\n> table in two, one for the range of codepoints and another one with the\n> single entries?\n\nI considered that, but there are relatively few singular codepoints in \nthe tables, so it wouldn't save much space. In this patch, singular \ncodepoints are represented by a range like \"0x3000, 0x3000\".\n\n> + 0x1D173, 0x1D17A, /* C.2.2 */\n> This is for musical symbols. It seems to me that checking for a range\n> is what is intended.\n\nCan you elaborate?\n\n- Heikki\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 6 Apr 2017 20:42:58 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\n> D'oh. Here's a new version, with saslprep.h included.\n\nAnd here it is for real. Sigh.\n\n>> There is for example this portion in the new tables:\n>> +static const Codepoint prohibited_output_chars[] =\n>> +{\n>> + 0xD800, 0xF8FF, /* C.3, C.5 */\n>>\n>> ----- Start Table C.5 -----\n>> D800-DFFF; [SURROGATE CODES]\n>> ----- End Table C.5 -----\n>> This indicates a range of values. Wouldn't it be better to split this\n>> table in two, one for the range of codepoints and another one with the\n>> single entries?\n>\n> I considered that, but there are relatively few singular codepoints in\n> the tables, so it wouldn't save much space. In this patch, singular\n> codepoints are represented by a range like \"0x3000, 0x3000\".\n>\n>> + 0x1D173, 0x1D17A, /* C.2.2 */\n>> This is for musical symbols. It seems to me that checking for a range\n>> is what is intended.\n>\n> Can you elaborate?\n\nOh, I think I understand the confusion now. All the arrays represent \ncodepoint ranges, not singular codepoints. I renamed them to \"*_ranges\", \nto make that more clear.\n\n- Heikki\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 6 Apr 2017 20:47:12 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On Fri, Apr 7, 2017 at 2:47 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n> On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\n>>> There is for example this portion in the new tables:\n>>> +static const Codepoint prohibited_output_chars[] =\n>>> +{\n>>> + 0xD800, 0xF8FF, /* C.3, C.5 */\n>>>\n>>> ----- Start Table C.5 -----\n>>> D800-DFFF; [SURROGATE CODES]\n>>> ----- End Table C.5 -----\n>>> This indicates a range of values. Wouldn't it be better to split this\n>>> table in two, one for the range of codepoints and another one with the\n>>> single entries?\n>>\n>> I considered that, but there are relatively few singular codepoints in\n>> the tables, so it wouldn't save much space. In this patch, singular\n>> codepoints are represented by a range like \"0x3000, 0x3000\".\n\nI am really wondering if this should not reflect the real range\nreported by the RFC. I understand that you have grouped things to save\na couple of bytes, but that would protect from any updates of the\ncodepoints within those ranges (unlikely to happen I agree).\n\n>>> + 0x1D173, 0x1D17A, /* C.2.2 */\n>>> This is for musical symbols. It seems to me that checking for a range\n>>> is what is intended.\n>>\n>> Can you elaborate?\n>\n> Oh, I think I understand the confusion now. All the arrays represent\n> codepoint ranges, not singular codepoints. I renamed them to \"*_ranges\", to\n> make that more clear.\n\nThanks! Yes I got confused by the name.\n\n+/* Is the given Unicode codepoint in the given table? */\n+#define IS_CODE_IN_TABLE(code, map) is_code_in_table(code, map, lengthof(map))\n[...]\n+static bool\n+is_code_in_table(pg_wchar code, const pg_wchar *map, int mapsize)\n+{\n+ Assert(mapsize % 2 == 0);\n+\n+ if (code < map[0] || code > map[mapsize - 1])\n+ return false;\n+\n+ if (bsearch(&code, map, mapsize / 2, sizeof(pg_wchar) * 2,\n+ codepoint_range_cmp))\n+ return true;\n+ else\n+ return false;\n+}\nThose could be renamed to range_table as well to avoid more confusion.\n\n> The SASLprep implementation depends on the UTF-8 functions from\n> src/backend/utils/mb/wchar.c. So to use it, you must also compile and link\n> that. That doesn't change anything for the current users of these\n> functions, the backend and libpq, as they both already link with wchar.o.\n> It would be good to move those functions into a separate file in\n> src/commmon, but I'll leave that for another day.\n\nFine for me.\n\nYou may want to add a .gitignore in src/common/unicode for norm_test\nand norm_test_table.h.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 7 Apr 2017 11:30:36 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":true,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On 04/07/2017 05:30 AM, Michael Paquier wrote:\n> On Fri, Apr 7, 2017 at 2:47 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n>> On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\n>>>> There is for example this portion in the new tables:\n>>>> +static const Codepoint prohibited_output_chars[] =\n>>>> +{\n>>>> + 0xD800, 0xF8FF, /* C.3, C.5 */\n>>>>\n>>>> ----- Start Table C.5 -----\n>>>> D800-DFFF; [SURROGATE CODES]\n>>>> ----- End Table C.5 -----\n>>>> This indicates a range of values. Wouldn't it be better to split this\n>>>> table in two, one for the range of codepoints and another one with the\n>>>> single entries?\n>>>\n>>> I considered that, but there are relatively few singular codepoints in\n>>> the tables, so it wouldn't save much space. In this patch, singular\n>>> codepoints are represented by a range like \"0x3000, 0x3000\".\n>\n> I am really wondering if this should not reflect the real range\n> reported by the RFC. I understand that you have grouped things to save\n> a couple of bytes, but that would protect from any updates of the\n> codepoints within those ranges (unlikely to happen I agree).\n\nIt just means that there will be some more work required to apply the \nchanges to the current lists. I constructed the lists manually to begin \nwith, copy-pasting the lists from the RFC, and moving and merging \nentries by hand. I wouldn't mind doing that by hand again, if the lists \nchange. But as you said, it seems unlikely that they would change any \ntime soon.\n\n> You may want to add a .gitignore in src/common/unicode for norm_test\n> and norm_test_table.h.\n\nAdded, and pushed, with some more comment fixes.\n\nMany thanks, Michael!\n\n- Heikki\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 7 Apr 2017 14:58:17 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On Fri, Apr 7, 2017 at 8:58 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n> On 04/07/2017 05:30 AM, Michael Paquier wrote:\n>> I am really wondering if this should not reflect the real range\n>> reported by the RFC. I understand that you have grouped things to save\n>> a couple of bytes, but that would protect from any updates of the\n>> codepoints within those ranges (unlikely to happen I agree).\n>\n> It just means that there will be some more work required to apply the\n> changes to the current lists. I constructed the lists manually to begin\n> with, copy-pasting the lists from the RFC, and moving and merging entries by\n> hand. I wouldn't mind doing that by hand again, if the lists change. But as\n> you said, it seems unlikely that they would change any time soon.\n\nYeah, I don't mind either. That's simple enough to change should that happen.\n\n>> You may want to add a .gitignore in src/common/unicode for norm_test\n>> and norm_test_table.h.\n>\n> Added, and pushed, with some more comment fixes.\n\nNice. There are still a couple of important items pending for SCRAM,\nso I would think that it is better to not do the refactoring now (but\nrework it in PG11), but polish a bit more the documentation. Your\nthoughts on that?\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 7 Apr 2017 21:31:51 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":true,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On 04/06/2017 07:59 PM, Heikki Linnakangas wrote:\n> Another thing I'd like some more eyes on, is how this will work with\n> encodings other than UTF-8. We will now try to normalize the password as\n> if it was in UTF-8, even if it isn't. That's OK as long as we're\n> consistent about it, but there is one worrisome scenario: what if the\n> user's password consists mostly of characters, that when interpreted as\n> UTF-8, are in the list of ignored characters. IOW, is it realistic that\n> a user might have a password in a non-UTF-8 encoding, that gets silently\n> mangled into something much shorter? I think that's highly unlikely, but\n> can anyone come up with a plausible example of that?\n\nI did some testing on what the byte sequences for the Unicode characters \nthat SASLprep ignores mean in other encodings. I created a text file \ncontaining every ignored character, in UTF-8, and ran \"iconv -f <other \nencoding> -t UTF-8//TRANSLIT\" on the file, using all supported server \nencodings. The idea is to take each of the ignored byte sequences, and \npretend that they are in some other encoding. If converting them to \nUTF-8 results in a legit character, then that character means something \nin that encoding, and could be misinterpreted if it's used in a password.\n\nHere are some characters that seem plausible to be misinterpreted and \nignored by SASLprep:\n\n-------\nEUC-JP and EUC-JISX0213:\n\nU+00AD (C2 AD): 足 (meaning \"foot\", per Unihan database)\nU+FE00-FE0F (EF B8 8X): 鏝 (meaning \"trowel\", per Unihan database)\n\nEUC-CN:\n\nU+00AD (C2 AD): 颅 (meaning \"skull\", per Unihan database)\nU+FE00-FE0FF (EF B8 8X): 锔 (meaning \"curium\", per Unihan database)\nU+FEFF (EF BB BF): 锘 (meaning \"nobelium\", per Wikipedia)\n\nEUC-KR:\n\nU+FE00-FE0F (EF BB BF): 截 (meanings \"cut off, stop, obstruct, \nintersect\", per Unihan database\nU+FEFF (EF BB BF): 癤 (meanings \"pimple, sore, boil\", per Unihan database)\n\nEUC-TW:\nU+FE00-FE0F: 踫 (meanings \"collide, bump into\", per Unihan database)\nU+FEFF: 踢 (meaning \"kick\", per Unihan database)\n\nCP866:\nU+1806: саЖ\nU+180B: саЛ\nU+180C: саМ\nU+180D: саН\nU+200B: тАЛ\nU+200C: тАМ\nU+200D: тАН\n-------\n\nThe CP866 cases seem most likely to cause confusion. Those are all \ncommon words in Russian. I don't know how common those Chinese/Japanese \ncharacters are.\n\nOverall, I think this is OK. Even though there are those characters that \ncan be misinterpreted, for it to be problem all of the following have to \nbe true:\n\n1. The client is using one of those encodings.\n2. The password string as whole has to look like valid UTF-8.\n3. Ignoring those characters/words from the password would lead to a \nsignificantly weaker password, i.e. it was not very long to begin with, \nor it consisted almost entirely of those characters/words.\n\nThoughts? Attached is the full results of running iconv with each \nencoding, from which I picked the above cases.\n\n- Heikki\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Mon, 10 Apr 2017 11:11:51 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"},{"msg_contents":"On Mon, Apr 10, 2017 at 5:11 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n> Here are some characters that seem plausible to be misinterpreted and\n> ignored by SASLprep:\n> EUC-JP and EUC-JISX0213:\n>\n> U+00AD (C2 AD): 足 (meaning \"foot\", per Unihan database)\n> U+FE00-FE0F (EF B8 8X): 鏝 (meaning \"trowel\", per Unihan database)\n\nThose are common words, but I still have to see those characters used\nin passwords. For user names, I have seen games or apps using Kanjis,\nso they could be used in such cases. But any application can be\ncareful in controlling the encoding used by the client, so I would not\nworry much about them.\n\n> EUC-CN:\n>\n> U+00AD (C2 AD): 颅 (meaning \"skull\", per Unihan database)\n> U+FE00-FE0FF (EF B8 8X): 锔 (meaning \"curium\", per Unihan database)\n> U+FEFF (EF BB BF): 锘 (meaning \"nobelium\", per Wikipedia)\n>\n> EUC-KR:\n>\n> U+FE00-FE0F (EF BB BF): 截 (meanings \"cut off, stop, obstruct, intersect\",\n> per Unihan database\n> U+FEFF (EF BB BF): 癤 (meanings \"pimple, sore, boil\", per Unihan database)\n>\n> EUC-TW:\n> U+FE00-FE0F: 踫 (meanings \"collide, bump into\", per Unihan database)\n> U+FEFF: 踢 (meaning \"kick\", per Unihan database)\n\nNot completely sure about those ones. At least I think that the two\nset of characters of Chinese Taipei are pretty common there.\n\n> CP866:\n> U+1806: саЖ\n> U+180B: саЛ\n> U+180C: саМ\n> U+180D: саН\n> U+200B: тАЛ\n> U+200C: тАМ\n> U+200D: тАН\n> -------\n>\n> The CP866 cases seem most likely to cause confusion.\n\nNo idea about those.\n\n> Those are all common\n> words in Russian. I don't know how common those Chinese/Japanese characters are.\n>\n> Overall, I think this is OK. Even though there are those characters that can\n> be misinterpreted, for it to be problem all of the following have to be\n> true:\n>\n> 1. The client is using one of those encodings.\n> 2. The password string as whole has to look like valid UTF-8.\n> 3. Ignoring those characters/words from the password would lead to a\n> significantly weaker password, i.e. it was not very long to begin with, or\n> it consisted almost entirely of those characters/words.\n>\n> Thoughts? Attached is the full results of running iconv with each encoding,\n> from which I picked the above cases.\n\nI am not much worrying about such things either. Still I am wondering\nthough if it would be useful to give users the possibility to block\nconnection attempts from clients that do not use UTF-8 in this case.\nSome people use open instances.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Tue, 11 Apr 2017 14:25:08 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":true,"msg_subject":"Re: Implementation of SASLprep for SCRAM-SHA-256"}],"string":"[\n {\n \"msg_contents\": \"Hi all,\\n\\nHere is another thread for an issue related to SCRAM\\n(https://www.postgresql.org/message-id/243d8c11-6149-a4bb-0909-136992f74b23@iki.fi),\\nthat can be discussed independently: SASLprep.\\n\\nRFC5802 (https://tools.ietf.org/html/rfc5802) regarding the\\nimplementation of SCRAM, needs to have passwords normalized as per\\nRFC4013 (https://tools.ietf.org/html/rfc4013) using SASLprep, which is\\nactually NFKC. I have previously described what happens in this case\\nhere:\\nhttps://www.postgresql.org/message-id/CAB7nPqScgwh6Eu4=c-0L7-cufZrU5rULTSdpMOOWiz1YFvGE6w@mail.gmail.com\\nThis way, we can be sure that two UTf-8 strings are considered as\\nequivalent in a SASL exchange, in our case we care about the password\\nstring (we should care about the username as well). Without SASLprep,\\nour implementation of SCRAM may fail with other third-part tools if\\npasswords are not made only of ASCII characters. And that sucks.\\n\\nTo put in short words, NFKC is made of two steps: a canonical\\ndecomposition of the UTF-8 string (decomposition of the string\\nfollowing by a reordering using the class of each character), followed\\nby its recomposition.\\n\\nPostgres does not have any logic on the frontend-side related to\\nencoding, still it is possible with some of the APIs of wchar.c to\\ncheck if a string provided is valid UTF-8 or not. If the string is not\\nvalid UTF-8, we should just use the string as-is in the exchange,\\nwhich leads to undefined behaviors anyway as SASL needs to do things\\nwith UTF-8.\\n\\nI have also implementation a Postgres extension able to do this\\noperation, which has been useful for testing.\\nhttps://github.com/michaelpq/pg_plugins/tree/master/pg_sasl_prepare\\n\\nAttached is a patch implementing that, the conversion tables are built\\nfrom UnicodeData.txt to be minimal in size. I have added an Open Item\\non the wiki for this problem as well.\\n\\nThanks,\\n-- \\nMichael\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 12:01:49 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Tue, Mar 7, 2017 at 10:01 PM, Michael Paquier\\n<michael.paquier@gmail.com> wrote:\\n> RFC5802 (https://tools.ietf.org/html/rfc5802) regarding the\\n> implementation of SCRAM, needs to have passwords normalized as per\\n> RFC4013 (https://tools.ietf.org/html/rfc4013) using SASLprep, which is\\n> actually NFKC. I have previously described what happens in this case\\n> here:\\n> https://www.postgresql.org/message-id/CAB7nPqScgwh6Eu4=c-0L7-cufZrU5rULTSdpMOOWiz1YFvGE6w@mail.gmail.com\\n> This way, we can be sure that two UTf-8 strings are considered as\\n> equivalent in a SASL exchange, in our case we care about the password\\n> string (we should care about the username as well). Without SASLprep,\\n> our implementation of SCRAM may fail with other third-part tools if\\n> passwords are not made only of ASCII characters. And that sucks.\\n\\nAgreed. I am not sure this quite rises to the level of a stop-ship\\nissue; we could document the restriction. However, that's pretty\\nugly; it would be a whole lot better to get this fixed.\\n\\nI kinda hope Heikki is going to step up to the plate here, because I\\nthink he understands most of it already.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 8 Mar 2017 08:39:02 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 10:39 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Tue, Mar 7, 2017 at 10:01 PM, Michael Paquier\\n> <michael.paquier@gmail.com> wrote:\\n>> This way, we can be sure that two UTf-8 strings are considered as\\n>> equivalent in a SASL exchange, in our case we care about the password\\n>> string (we should care about the username as well). Without SASLprep,\\n>> our implementation of SCRAM may fail with other third-part tools if\\n>> passwords are not made only of ASCII characters. And that sucks.\\n>\\n> Agreed. I am not sure this quite rises to the level of a stop-ship\\n> issue; we could document the restriction.\\n\\nI am not sure about that, what we have now on HEAD is still useful and\\nbetter than MD5.\\n\\n> However, that's pretty ugly; it would be a whole lot better to get this fixed.\\n\\nAgreed.\\n\\n> I kinda hope Heikki is going to step up to the plate here, because I\\n> think he understands most of it already.\\n\\nThe second person who knows a good deal about NFKC is Ishii-san.\\n\\nAttached is a rebased patch.\\n-- \\nMichael\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 31 Mar 2017 16:10:03 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 03/31/2017 10:10 AM, Michael Paquier wrote:\\n> On Wed, Mar 8, 2017 at 10:39 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>> On Tue, Mar 7, 2017 at 10:01 PM, Michael Paquier\\n>> <michael.paquier@gmail.com> wrote:\\n>> I kinda hope Heikki is going to step up to the plate here, because I\\n>> think he understands most of it already.\\n>\\n> The second person who knows a good deal about NFKC is Ishii-san.\\n>\\n> Attached is a rebased patch.\\n\\nThanks, I'm looking at this now.\\n\\n- Heikki\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Tue, 4 Apr 2017 13:52:41 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 04/04/2017 01:52 PM, Heikki Linnakangas wrote:\\n> On 03/31/2017 10:10 AM, Michael Paquier wrote:\\n>> On Wed, Mar 8, 2017 at 10:39 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>>> On Tue, Mar 7, 2017 at 10:01 PM, Michael Paquier\\n>>> <michael.paquier@gmail.com> wrote:\\n>>> I kinda hope Heikki is going to step up to the plate here, because I\\n>>> think he understands most of it already.\\n>>\\n>> The second person who knows a good deal about NFKC is Ishii-san.\\n>>\\n>> Attached is a rebased patch.\\n>\\n> Thanks, I'm looking at this now.\\n\\nI will continue tomorrow, but I wanted to report on what I've done so \\nfar. Attached is a new patch version, quite heavily modified. Notable \\nchanges so far:\\n\\n* Use Unicode codepoints, rather than UTF-8 bytes packed in a 32-bit \\nints. IMHO this makes the tables easier to read (to a human), and they \\nare also packed slightly more tightly (see next two points), as you can \\nfit more codepoints in a 16-bit integer.\\n\\n* Reorganize the decomposition table. Instead of a separate \\nbinary-searched table for each \\\"size\\\", have one table that's ordered by \\ncodepoint, which contains a length and offset into another array, where \\nall the decomposed sequences are. This makes the recomposition step \\nsomewhat slower, as it now has to grovel through an array that contains \\nall decompositions, rather than just the array that contains \\ndecompositions of two characters. But this isn't performance-critical, \\nreadability and tight packing of the tables matter more.\\n\\n* In the lookup table, store singleton decompositions that decompose to \\na single character, and the character fits in a uint16, directly in the \\nmain lookup table, instead of storing an index into the other table. \\nThis makes the tables somewhat smaller, as there are a lot of such \\ndecompositions.\\n\\n* Use uint32 instead of pg_wchar to represent unicode codepoints. \\npg_wchar suggests something that holds multi-byte characters in the OS \\nlocale, used by functions like wcscmp, so avoid that. I added a \\\"typedef \\nuint32 Codepoint\\\" alias, though, to make it more clear which variables \\nhold Unicode codepoints rather than e.g. indexes into arrays.\\n\\n* Unicode consortium publishes a comprehensive normalization test suite, \\nas a text file, NormalizationTest.txt. I wrote a small perl and C \\nprogram to run all the tests from that test suite, with the \\nnormalization function. That uncovered a number of bugs in the \\nrecomposition algorithm, which I then fixed:\\n\\n * decompositions that map to ascii characters cannot be excluded.\\n * non-canonical and non-starter decompositions need to be excluded. \\nThey are now flagged in the lookup table, so that we only use them \\nduring the decomposition phase, not when recomposing.\\n\\n\\nTODOs / Discussion points:\\n\\n* I'm not sure I like the way the code is organized, I feel that we're \\nlittering src/common with these. Perhaps we should add a \\nsrc/common/unicode_normalization directory for all this.\\n\\n* The list of characters excluded from recomposition is currently \\nhard-coded in utf_norm_generate.pl. However, that list is available in \\nmachine-readable format, in file CompositionExclusions.txt. Since we're \\nreading most of the data from UnicodeData.txt, would be good to read the \\nexclusion table from a file, too.\\n\\n* SASLPrep specifies normalization form KC, but it also specifies that \\nsome characters are mapped to space or nothing. Should do those \\nmappings, too.\\n\\n- Heikki\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 5 Apr 2017 01:05:12 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"fore\\n\\nOn Wed, Apr 5, 2017 at 7:05 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> I will continue tomorrow, but I wanted to report on what I've done so far.\\n> Attached is a new patch version, quite heavily modified. Notable changes so\\n> far:\\n\\nGreat, thanks!\\n\\n> * Use Unicode codepoints, rather than UTF-8 bytes packed in a 32-bit ints.\\n> IMHO this makes the tables easier to read (to a human), and they are also\\n> packed slightly more tightly (see next two points), as you can fit more\\n> codepoints in a 16-bit integer.\\n\\nUsing directly codepoints is not much consistent with the existing\\nbackend, but for the sake of packing things more, OK.\\n\\n> * Reorganize the decomposition table. Instead of a separate binary-searched\\n> table for each \\\"size\\\", have one table that's ordered by codepoint, which\\n> contains a length and offset into another array, where all the decomposed\\n> sequences are. This makes the recomposition step somewhat slower, as it now\\n> has to grovel through an array that contains all decompositions, rather than\\n> just the array that contains decompositions of two characters. But this\\n> isn't performance-critical, readability and tight packing of the tables\\n> matter more.\\n\\nOkay, no objection to that.\\n\\n> * In the lookup table, store singleton decompositions that decompose to a\\n> single character, and the character fits in a uint16, directly in the main\\n> lookup table, instead of storing an index into the other table. This makes\\n> the tables somewhat smaller, as there are a lot of such decompositions.\\n\\nIndeed.\\n\\n> * Use uint32 instead of pg_wchar to represent unicode codepoints. pg_wchar\\n> suggests something that holds multi-byte characters in the OS locale, used\\n> by functions like wcscmp, so avoid that. I added a \\\"typedef uint32\\n> Codepoint\\\" alias, though, to make it more clear which variables hold Unicode\\n> codepoints rather than e.g. indexes into arrays.\\n>\\n> * Unicode consortium publishes a comprehensive normalization test suite, as\\n> a text file, NormalizationTest.txt. I wrote a small perl and C program to\\n> run all the tests from that test suite, with the normalization function.\\n> That uncovered a number of bugs in the recomposition algorithm, which I then\\n> fixed:\\n\\nI was looking for something like that for some time, thanks! That's\\nhere actually:\\nhttp://www.unicode.org/Public/8.0.0/ucd/NormalizationTest.txt\\n\\n> * decompositions that map to ascii characters cannot be excluded.\\n> * non-canonical and non-starter decompositions need to be excluded. They\\n> are now flagged in the lookup table, so that we only use them during the\\n> decomposition phase, not when recomposing.\\n\\nOkay on that.\\n\\n> TODOs / Discussion points:\\n>\\n> * I'm not sure I like the way the code is organized, I feel that we're\\n> littering src/common with these. Perhaps we should add a\\n> src/common/unicode_normalization directory for all this.\\n\\npg_utf8_islegal() and pg_utf_mblen() should as well be moved in their\\nown file I think, and wchar.c can use that.\\n\\n> * The list of characters excluded from recomposition is currently hard-coded\\n> in utf_norm_generate.pl. However, that list is available in machine-readable\\n> format, in file CompositionExclusions.txt. Since we're reading most of the\\n> data from UnicodeData.txt, would be good to read the exclusion table from a\\n> file, too.\\n\\nOuch. Those are present here...\\nhttp://www.unicode.org/reports/tr41/tr41-19.html#Exclusions\\nDefinitely it makes more sense to read them from a file.\\n\\n> * SASLPrep specifies normalization form KC, but it also specifies that some\\n> characters are mapped to space or nothing. Should do those mappings, too.\\n\\nAh, right. Those ones are here:\\nhttps://tools.ietf.org/html/rfc3454#appendix-B.1\\nThe conversion tables need some extra tweaking...\\n\\n+ else if ((*utf & 0xf0) == 0xe0)\\n+ {\\n+ if (utf[1] == 0 || utf[2] == 0)\\n+ goto invalid;\\n+ cp = (*utf++) & 0x0F;\\n+ cp = (cp << 6) | ((*utf++) & 0x3F);\\n+ cp = (cp << 6) | ((*utf++) & 0x3F);\\n+ }\\n+ else if ((*utf & 0xf8) == 0xf0)\\n+ {\\n+ if (utf[1] == 0 || utf[2] == 0|| utf[3] == 0)\\n+ goto invalid;\\n+ cp = (*utf++) & 0x07;\\n+ cp = (cp << 6) | ((*utf++) & 0x3F);\\n+ cp = (cp << 6) | ((*utf++) & 0x3F);\\n+ cp = (cp << 6) | ((*utf++) & 0x3F);\\n+ }\\nI find more readable something like pg_utf2wchar_with_len(), but well...\\n\\nSome typos:\\ns/fore/for/\\ns/reprsented/represented/\\n\\nYou seem to be fully on it... I can help out with any of those items as needed.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 13:23:12 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 04/05/2017 07:23 AM, Michael Paquier wrote:\\n> fore\\n>\\n> On Wed, Apr 5, 2017 at 7:05 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n>> I will continue tomorrow, but I wanted to report on what I've done so far.\\n>> Attached is a new patch version, quite heavily modified. Notable changes so\\n>> far:\\n>\\n> Great, thanks!\\n>\\n>> * Use Unicode codepoints, rather than UTF-8 bytes packed in a 32-bit ints.\\n>> IMHO this makes the tables easier to read (to a human), and they are also\\n>> packed slightly more tightly (see next two points), as you can fit more\\n>> codepoints in a 16-bit integer.\\n>\\n> Using directly codepoints is not much consistent with the existing\\n> backend, but for the sake of packing things more, OK.\\n\\nOh, I see, we already have similar functions in wchar.c. \\nunicode_to_utf8() and utf8_to_unicode(). We should probably move those \\nto src/common, rather than re-invent the wheel.\\n\\n> pg_utf8_islegal() and pg_utf_mblen() should as well be moved in their\\n> own file I think, and wchar.c can use that.\\n\\nYeah..\\n\\n>> * The list of characters excluded from recomposition is currently hard-coded\\n>> in utf_norm_generate.pl. However, that list is available in machine-readable\\n>> format, in file CompositionExclusions.txt. Since we're reading most of the\\n>> data from UnicodeData.txt, would be good to read the exclusion table from a\\n>> file, too.\\n>\\n> Ouch. Those are present here...\\n> http://www.unicode.org/reports/tr41/tr41-19.html#Exclusions\\n> Definitely it makes more sense to read them from a file.\\n\\nDid that.\\n\\n>> * SASLPrep specifies normalization form KC, but it also specifies that some\\n>> characters are mapped to space or nothing. Should do those mappings, too.\\n>\\n> Ah, right. Those ones are here:\\n> https://tools.ietf.org/html/rfc3454#appendix-B.1\\n\\nYep.\\n\\n\\nAttached is a new version. Notable changes since yesterday:\\n\\n* Implemented the rest of the SASLPrep, mapping some characters to \\nspaces, leaving out others, and checking for prohibited characters and \\nbidirectional strings.\\n\\n* Moved things around. There's now a separate directory, \\nsrc/common/unicode, which contains the perl scripts and the test code. \\nThose are not needed to build from source, as the pre-generated tables \\nare put in src/include/common. Similar to the scripts in \\nsrc/backend/utils/mb/Unicode, really.\\n\\n* Renamed many things from utf_* to unicode_*, since they don't deal \\nwith utf-8 input anymore.\\n\\n\\nThis is starting to shape up, but still some cleanup work to do. I will \\ncontinue tomorrow..\\n\\n- Heikki\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 5 Apr 2017 19:33:13 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Thu, Apr 6, 2017 at 1:33 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> Attached is a new version. Notable changes since yesterday:\\n>\\n> * Implemented the rest of the SASLPrep, mapping some characters to spaces,\\n> leaving out others, and checking for prohibited characters and bidirectional\\n> strings.\\n>\\n> * Moved things around. There's now a separate directory, src/common/unicode,\\n> which contains the perl scripts and the test code. Those are not needed to\\n> build from source, as the pre-generated tables are put in\\n> src/include/common. Similar to the scripts in src/backend/utils/mb/Unicode,\\n> really.\\n>\\n> * Renamed many things from utf_* to unicode_*, since they don't deal with\\n> utf-8 input anymore.\\n>\\n> This is starting to shape up, but still some cleanup work to do. I will\\n> continue tomorrow..\\n\\nThanks for the new patch, that's looking nice. Now I was not able to\\ncompile it as saslprep.h is missing from what you have sent...\\n\\nThere is for example this portion in the new tables:\\n+static const Codepoint prohibited_output_chars[] =\\n+{\\n+ 0xD800, 0xF8FF, /* C.3, C.5 */\\n\\n ----- Start Table C.5 -----\\n D800-DFFF; [SURROGATE CODES]\\n ----- End Table C.5 -----\\nThis indicates a range of values. Wouldn't it be better to split this\\ntable in two, one for the range of codepoints and another one with the\\nsingle entries?\\n\\n+ 0x1D173, 0x1D17A, /* C.2.2 */\\nThis is for musical symbols. It seems to me that checking for a range\\nis what is intended.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 6 Apr 2017 16:32:08 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"Another version attached.\\n\\nI think this is now in committable state, but there's been a lot of \\nsmall changes here and there, so please have one more look over it if \\nyou have a chance. I'm planning to push this tomorrow, after sleeping on it.\\n\\nThe code-organization issue with the utf8 functions, pg_utf_mblen, \\npg_utf8_islegal, unicode_to_utf8, and utf8_to_unicode, is still kind of \\nunresolved. The way I have it in this version is that the functions are \\nin wchar.c, like they've always been, and saslprep.c uses them from \\nthere. This is ugly from a code organization point of view. as we now \\nhave functions in src/common depending on functions in \\nsrc/backend/utils/mb/wchar.c. I think wchar.c deserves some \\nre-organization, as most of the functions there are also used in \\nfrontend code. Perhaps it should be moved to src/common as whole.\\n\\nLibpq already compiles with wchar.c, so all those functions that \\nsrc/common/saslprep.c requires are already available in libpq. So I \\ndon't think that's urgent, but something we nevertheless ought to clean \\nup at some point.\\n\\n\\nAnother thing I'd like some more eyes on, is how this will work with \\nencodings other than UTF-8. We will now try to normalize the password as \\nif it was in UTF-8, even if it isn't. That's OK as long as we're \\nconsistent about it, but there is one worrisome scenario: what if the \\nuser's password consists mostly of characters, that when interpreted as \\nUTF-8, are in the list of ignored characters. IOW, is it realistic that \\na user might have a password in a non-UTF-8 encoding, that gets silently \\nmangled into something much shorter? I think that's highly unlikely, but \\ncan anyone come up with a plausible example of that?\\n\\n- Heikki\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 6 Apr 2017 19:59:16 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"(sorry, I didn't notice your email until after I just sent version 4!)\\n\\nOn 04/06/2017 10:32 AM, Michael Paquier wrote:\\n> On Thu, Apr 6, 2017 at 1:33 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n>> Attached is a new version. Notable changes since yesterday:\\n>>\\n>> * Implemented the rest of the SASLPrep, mapping some characters to spaces,\\n>> leaving out others, and checking for prohibited characters and bidirectional\\n>> strings.\\n>>\\n>> * Moved things around. There's now a separate directory, src/common/unicode,\\n>> which contains the perl scripts and the test code. Those are not needed to\\n>> build from source, as the pre-generated tables are put in\\n>> src/include/common. Similar to the scripts in src/backend/utils/mb/Unicode,\\n>> really.\\n>>\\n>> * Renamed many things from utf_* to unicode_*, since they don't deal with\\n>> utf-8 input anymore.\\n>>\\n>> This is starting to shape up, but still some cleanup work to do. I will\\n>> continue tomorrow..\\n>\\n> Thanks for the new patch, that's looking nice. Now I was not able to\\n> compile it as saslprep.h is missing from what you have sent...\\n\\nD'oh. Here's a new version, with saslprep.h included.\\n\\n> There is for example this portion in the new tables:\\n> +static const Codepoint prohibited_output_chars[] =\\n> +{\\n> + 0xD800, 0xF8FF, /* C.3, C.5 */\\n>\\n> ----- Start Table C.5 -----\\n> D800-DFFF; [SURROGATE CODES]\\n> ----- End Table C.5 -----\\n> This indicates a range of values. Wouldn't it be better to split this\\n> table in two, one for the range of codepoints and another one with the\\n> single entries?\\n\\nI considered that, but there are relatively few singular codepoints in \\nthe tables, so it wouldn't save much space. In this patch, singular \\ncodepoints are represented by a range like \\\"0x3000, 0x3000\\\".\\n\\n> + 0x1D173, 0x1D17A, /* C.2.2 */\\n> This is for musical symbols. It seems to me that checking for a range\\n> is what is intended.\\n\\nCan you elaborate?\\n\\n- Heikki\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 6 Apr 2017 20:42:58 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\\n> D'oh. Here's a new version, with saslprep.h included.\\n\\nAnd here it is for real. Sigh.\\n\\n>> There is for example this portion in the new tables:\\n>> +static const Codepoint prohibited_output_chars[] =\\n>> +{\\n>> + 0xD800, 0xF8FF, /* C.3, C.5 */\\n>>\\n>> ----- Start Table C.5 -----\\n>> D800-DFFF; [SURROGATE CODES]\\n>> ----- End Table C.5 -----\\n>> This indicates a range of values. Wouldn't it be better to split this\\n>> table in two, one for the range of codepoints and another one with the\\n>> single entries?\\n>\\n> I considered that, but there are relatively few singular codepoints in\\n> the tables, so it wouldn't save much space. In this patch, singular\\n> codepoints are represented by a range like \\\"0x3000, 0x3000\\\".\\n>\\n>> + 0x1D173, 0x1D17A, /* C.2.2 */\\n>> This is for musical symbols. It seems to me that checking for a range\\n>> is what is intended.\\n>\\n> Can you elaborate?\\n\\nOh, I think I understand the confusion now. All the arrays represent \\ncodepoint ranges, not singular codepoints. I renamed them to \\\"*_ranges\\\", \\nto make that more clear.\\n\\n- Heikki\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 6 Apr 2017 20:47:12 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Fri, Apr 7, 2017 at 2:47 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\\n>>> There is for example this portion in the new tables:\\n>>> +static const Codepoint prohibited_output_chars[] =\\n>>> +{\\n>>> + 0xD800, 0xF8FF, /* C.3, C.5 */\\n>>>\\n>>> ----- Start Table C.5 -----\\n>>> D800-DFFF; [SURROGATE CODES]\\n>>> ----- End Table C.5 -----\\n>>> This indicates a range of values. Wouldn't it be better to split this\\n>>> table in two, one for the range of codepoints and another one with the\\n>>> single entries?\\n>>\\n>> I considered that, but there are relatively few singular codepoints in\\n>> the tables, so it wouldn't save much space. In this patch, singular\\n>> codepoints are represented by a range like \\\"0x3000, 0x3000\\\".\\n\\nI am really wondering if this should not reflect the real range\\nreported by the RFC. I understand that you have grouped things to save\\na couple of bytes, but that would protect from any updates of the\\ncodepoints within those ranges (unlikely to happen I agree).\\n\\n>>> + 0x1D173, 0x1D17A, /* C.2.2 */\\n>>> This is for musical symbols. It seems to me that checking for a range\\n>>> is what is intended.\\n>>\\n>> Can you elaborate?\\n>\\n> Oh, I think I understand the confusion now. All the arrays represent\\n> codepoint ranges, not singular codepoints. I renamed them to \\\"*_ranges\\\", to\\n> make that more clear.\\n\\nThanks! Yes I got confused by the name.\\n\\n+/* Is the given Unicode codepoint in the given table? */\\n+#define IS_CODE_IN_TABLE(code, map) is_code_in_table(code, map, lengthof(map))\\n[...]\\n+static bool\\n+is_code_in_table(pg_wchar code, const pg_wchar *map, int mapsize)\\n+{\\n+ Assert(mapsize % 2 == 0);\\n+\\n+ if (code < map[0] || code > map[mapsize - 1])\\n+ return false;\\n+\\n+ if (bsearch(&code, map, mapsize / 2, sizeof(pg_wchar) * 2,\\n+ codepoint_range_cmp))\\n+ return true;\\n+ else\\n+ return false;\\n+}\\nThose could be renamed to range_table as well to avoid more confusion.\\n\\n> The SASLprep implementation depends on the UTF-8 functions from\\n> src/backend/utils/mb/wchar.c. So to use it, you must also compile and link\\n> that. That doesn't change anything for the current users of these\\n> functions, the backend and libpq, as they both already link with wchar.o.\\n> It would be good to move those functions into a separate file in\\n> src/commmon, but I'll leave that for another day.\\n\\nFine for me.\\n\\nYou may want to add a .gitignore in src/common/unicode for norm_test\\nand norm_test_table.h.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 7 Apr 2017 11:30:36 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 04/07/2017 05:30 AM, Michael Paquier wrote:\\n> On Fri, Apr 7, 2017 at 2:47 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n>> On 04/06/2017 08:42 PM, Heikki Linnakangas wrote:\\n>>>> There is for example this portion in the new tables:\\n>>>> +static const Codepoint prohibited_output_chars[] =\\n>>>> +{\\n>>>> + 0xD800, 0xF8FF, /* C.3, C.5 */\\n>>>>\\n>>>> ----- Start Table C.5 -----\\n>>>> D800-DFFF; [SURROGATE CODES]\\n>>>> ----- End Table C.5 -----\\n>>>> This indicates a range of values. Wouldn't it be better to split this\\n>>>> table in two, one for the range of codepoints and another one with the\\n>>>> single entries?\\n>>>\\n>>> I considered that, but there are relatively few singular codepoints in\\n>>> the tables, so it wouldn't save much space. In this patch, singular\\n>>> codepoints are represented by a range like \\\"0x3000, 0x3000\\\".\\n>\\n> I am really wondering if this should not reflect the real range\\n> reported by the RFC. I understand that you have grouped things to save\\n> a couple of bytes, but that would protect from any updates of the\\n> codepoints within those ranges (unlikely to happen I agree).\\n\\nIt just means that there will be some more work required to apply the \\nchanges to the current lists. I constructed the lists manually to begin \\nwith, copy-pasting the lists from the RFC, and moving and merging \\nentries by hand. I wouldn't mind doing that by hand again, if the lists \\nchange. But as you said, it seems unlikely that they would change any \\ntime soon.\\n\\n> You may want to add a .gitignore in src/common/unicode for norm_test\\n> and norm_test_table.h.\\n\\nAdded, and pushed, with some more comment fixes.\\n\\nMany thanks, Michael!\\n\\n- Heikki\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 7 Apr 2017 14:58:17 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Fri, Apr 7, 2017 at 8:58 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> On 04/07/2017 05:30 AM, Michael Paquier wrote:\\n>> I am really wondering if this should not reflect the real range\\n>> reported by the RFC. I understand that you have grouped things to save\\n>> a couple of bytes, but that would protect from any updates of the\\n>> codepoints within those ranges (unlikely to happen I agree).\\n>\\n> It just means that there will be some more work required to apply the\\n> changes to the current lists. I constructed the lists manually to begin\\n> with, copy-pasting the lists from the RFC, and moving and merging entries by\\n> hand. I wouldn't mind doing that by hand again, if the lists change. But as\\n> you said, it seems unlikely that they would change any time soon.\\n\\nYeah, I don't mind either. That's simple enough to change should that happen.\\n\\n>> You may want to add a .gitignore in src/common/unicode for norm_test\\n>> and norm_test_table.h.\\n>\\n> Added, and pushed, with some more comment fixes.\\n\\nNice. There are still a couple of important items pending for SCRAM,\\nso I would think that it is better to not do the refactoring now (but\\nrework it in PG11), but polish a bit more the documentation. Your\\nthoughts on that?\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 7 Apr 2017 21:31:51 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On 04/06/2017 07:59 PM, Heikki Linnakangas wrote:\\n> Another thing I'd like some more eyes on, is how this will work with\\n> encodings other than UTF-8. We will now try to normalize the password as\\n> if it was in UTF-8, even if it isn't. That's OK as long as we're\\n> consistent about it, but there is one worrisome scenario: what if the\\n> user's password consists mostly of characters, that when interpreted as\\n> UTF-8, are in the list of ignored characters. IOW, is it realistic that\\n> a user might have a password in a non-UTF-8 encoding, that gets silently\\n> mangled into something much shorter? I think that's highly unlikely, but\\n> can anyone come up with a plausible example of that?\\n\\nI did some testing on what the byte sequences for the Unicode characters \\nthat SASLprep ignores mean in other encodings. I created a text file \\ncontaining every ignored character, in UTF-8, and ran \\\"iconv -f <other \\nencoding> -t UTF-8//TRANSLIT\\\" on the file, using all supported server \\nencodings. The idea is to take each of the ignored byte sequences, and \\npretend that they are in some other encoding. If converting them to \\nUTF-8 results in a legit character, then that character means something \\nin that encoding, and could be misinterpreted if it's used in a password.\\n\\nHere are some characters that seem plausible to be misinterpreted and \\nignored by SASLprep:\\n\\n-------\\nEUC-JP and EUC-JISX0213:\\n\\nU+00AD (C2 AD): 足 (meaning \\\"foot\\\", per Unihan database)\\nU+FE00-FE0F (EF B8 8X): 鏝 (meaning \\\"trowel\\\", per Unihan database)\\n\\nEUC-CN:\\n\\nU+00AD (C2 AD): 颅 (meaning \\\"skull\\\", per Unihan database)\\nU+FE00-FE0FF (EF B8 8X): 锔 (meaning \\\"curium\\\", per Unihan database)\\nU+FEFF (EF BB BF): 锘 (meaning \\\"nobelium\\\", per Wikipedia)\\n\\nEUC-KR:\\n\\nU+FE00-FE0F (EF BB BF): 截 (meanings \\\"cut off, stop, obstruct, \\nintersect\\\", per Unihan database\\nU+FEFF (EF BB BF): 癤 (meanings \\\"pimple, sore, boil\\\", per Unihan database)\\n\\nEUC-TW:\\nU+FE00-FE0F: 踫 (meanings \\\"collide, bump into\\\", per Unihan database)\\nU+FEFF: 踢 (meaning \\\"kick\\\", per Unihan database)\\n\\nCP866:\\nU+1806: саЖ\\nU+180B: саЛ\\nU+180C: саМ\\nU+180D: саН\\nU+200B: тАЛ\\nU+200C: тАМ\\nU+200D: тАН\\n-------\\n\\nThe CP866 cases seem most likely to cause confusion. Those are all \\ncommon words in Russian. I don't know how common those Chinese/Japanese \\ncharacters are.\\n\\nOverall, I think this is OK. Even though there are those characters that \\ncan be misinterpreted, for it to be problem all of the following have to \\nbe true:\\n\\n1. The client is using one of those encodings.\\n2. The password string as whole has to look like valid UTF-8.\\n3. Ignoring those characters/words from the password would lead to a \\nsignificantly weaker password, i.e. it was not very long to begin with, \\nor it consisted almost entirely of those characters/words.\\n\\nThoughts? Attached is the full results of running iconv with each \\nencoding, from which I picked the above cases.\\n\\n- Heikki\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Mon, 10 Apr 2017 11:11:51 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n },\n {\n \"msg_contents\": \"On Mon, Apr 10, 2017 at 5:11 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> Here are some characters that seem plausible to be misinterpreted and\\n> ignored by SASLprep:\\n> EUC-JP and EUC-JISX0213:\\n>\\n> U+00AD (C2 AD): 足 (meaning \\\"foot\\\", per Unihan database)\\n> U+FE00-FE0F (EF B8 8X): 鏝 (meaning \\\"trowel\\\", per Unihan database)\\n\\nThose are common words, but I still have to see those characters used\\nin passwords. For user names, I have seen games or apps using Kanjis,\\nso they could be used in such cases. But any application can be\\ncareful in controlling the encoding used by the client, so I would not\\nworry much about them.\\n\\n> EUC-CN:\\n>\\n> U+00AD (C2 AD): 颅 (meaning \\\"skull\\\", per Unihan database)\\n> U+FE00-FE0FF (EF B8 8X): 锔 (meaning \\\"curium\\\", per Unihan database)\\n> U+FEFF (EF BB BF): 锘 (meaning \\\"nobelium\\\", per Wikipedia)\\n>\\n> EUC-KR:\\n>\\n> U+FE00-FE0F (EF BB BF): 截 (meanings \\\"cut off, stop, obstruct, intersect\\\",\\n> per Unihan database\\n> U+FEFF (EF BB BF): 癤 (meanings \\\"pimple, sore, boil\\\", per Unihan database)\\n>\\n> EUC-TW:\\n> U+FE00-FE0F: 踫 (meanings \\\"collide, bump into\\\", per Unihan database)\\n> U+FEFF: 踢 (meaning \\\"kick\\\", per Unihan database)\\n\\nNot completely sure about those ones. At least I think that the two\\nset of characters of Chinese Taipei are pretty common there.\\n\\n> CP866:\\n> U+1806: саЖ\\n> U+180B: саЛ\\n> U+180C: саМ\\n> U+180D: саН\\n> U+200B: тАЛ\\n> U+200C: тАМ\\n> U+200D: тАН\\n> -------\\n>\\n> The CP866 cases seem most likely to cause confusion.\\n\\nNo idea about those.\\n\\n> Those are all common\\n> words in Russian. I don't know how common those Chinese/Japanese characters are.\\n>\\n> Overall, I think this is OK. Even though there are those characters that can\\n> be misinterpreted, for it to be problem all of the following have to be\\n> true:\\n>\\n> 1. The client is using one of those encodings.\\n> 2. The password string as whole has to look like valid UTF-8.\\n> 3. Ignoring those characters/words from the password would lead to a\\n> significantly weaker password, i.e. it was not very long to begin with, or\\n> it consisted almost entirely of those characters/words.\\n>\\n> Thoughts? Attached is the full results of running iconv with each encoding,\\n> from which I picked the above cases.\\n\\nI am not much worrying about such things either. Still I am wondering\\nthough if it would be useful to give users the possibility to block\\nconnection attempts from clients that do not use UTF-8 in this case.\\nSome people use open instances.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Tue, 11 Apr 2017 14:25:08 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Implementation of SASLprep for SCRAM-SHA-256\"\n }\n]","isConversational":false}}},{"rowIdx":44086,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hello,\n\ndblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n\n\n\npostgres=# select count(*) from pg_stat_activity;\n count\n-------\n 1\n(1 row)\n\npostgres=# select dblink_connect('dbname=postgres');\n dblink_connect\n----------------\n OK\n(1 row)\n\npostgres=# select count(*) from pg_stat_activity;\n count\n-------\n 2\n(1 row)\n\npostgres=# select dblink_connect('dbname=postgres');\n dblink_connect\n----------------\n OK\n(1 row)\n\npostgres=# select count(*) from pg_stat_activity;\n count\n-------\n 3\n(1 row)\n\n\nRegards\nTakayuki Tsunakawa\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 8 Mar 2017 06:52:26 +0000","msg_from":"\"Tsunakawa, Takayuki\" <tsunakawa.takay@jp.fujitsu.com>","msg_from_op":true,"msg_subject":"[bug fix] dblink leaks unnamed connections"},{"msg_contents":"On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\n<tsunakawa.takay@jp.fujitsu.com> wrote:\n> Hello,\n>\n> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n\nThis issue was reported about ten years ago and added as TODO item.\nhttp://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\n\nI agree that this is a bug, and am tempted to back-patch to all the supported\nversions. But it had not been fixed in many years since the first report of\nthe issue. So I'm not sure if it's ok to just treat this as a bug right now and\nback-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\nmore opinions about this.\n\nRegards,\n\n-- \nFujii Masao\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 00:41:48 +0900","msg_from":"Fujii Masao <masao.fujii@gmail.com>","msg_from_op":false,"msg_subject":"Re: [bug fix] dblink leaks unnamed connections"},{"msg_contents":"Fujii Masao <masao.fujii@gmail.com> writes:\n> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\n> <tsunakawa.takay@jp.fujitsu.com> wrote:\n>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n\n> This issue was reported about ten years ago and added as TODO item.\n> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\n\n> I agree that this is a bug, and am tempted to back-patch to all the supported\n> versions. But it had not been fixed in many years since the first report of\n> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\n> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\n> more opinions about this.\n\nIt looks to me like the issue simply fell through the cracks because Joe\nwasn't excited about fixing it. Now that we have a second complaint,\nI think it's worth treating as a bug and back-patching.\n\n(I've not read this particular patch and am not expressing an opinion\nwhether it's correct.)\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 09 Mar 2017 10:54:11 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: [bug fix] dblink leaks unnamed connections"},{"msg_contents":"On 03/09/2017 07:54 AM, Tom Lane wrote:\n> Fujii Masao <masao.fujii@gmail.com> writes:\n>> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\n>> <tsunakawa.takay@jp.fujitsu.com> wrote:\n>>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n> \n>> This issue was reported about ten years ago and added as TODO item.\n>> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\n> \n>> I agree that this is a bug, and am tempted to back-patch to all the supported\n>> versions. But it had not been fixed in many years since the first report of\n>> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\n>> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\n>> more opinions about this.\n> \n> It looks to me like the issue simply fell through the cracks because Joe\n> wasn't excited about fixing it. Now that we have a second complaint,\n> I think it's worth treating as a bug and back-patching.\n> \n> (I've not read this particular patch and am not expressing an opinion\n> whether it's correct.)\n\nOk, will take another look.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Thu, 9 Mar 2017 08:31:42 -0800","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: [bug fix] dblink leaks unnamed connections"},{"msg_contents":"On 03/09/2017 08:31 AM, Joe Conway wrote:\n> On 03/09/2017 07:54 AM, Tom Lane wrote:\n>> Fujii Masao <masao.fujii@gmail.com> writes:\n>>> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\n>>> <tsunakawa.takay@jp.fujitsu.com> wrote:\n>>>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n>> \n>>> This issue was reported about ten years ago and added as TODO item.\n>>> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\n>> \n>>> I agree that this is a bug, and am tempted to back-patch to all the supported\n>>> versions. But it had not been fixed in many years since the first report of\n>>> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\n>>> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\n>>> more opinions about this.\n>> \n>> It looks to me like the issue simply fell through the cracks because Joe\n>> wasn't excited about fixing it. Now that we have a second complaint,\n>> I think it's worth treating as a bug and back-patching.\n>> \n>> (I've not read this particular patch and am not expressing an opinion\n>> whether it's correct.)\n> \n> Ok, will take another look.\n\nI pushed a fix to all supported branches.\n\nThanks,\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sat, 11 Mar 2017 13:37:36 -0800","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: [bug fix] dblink leaks unnamed connections"}],"string":"[\n {\n \"msg_contents\": \"Hello,\\n\\ndblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\\n\\n\\n\\npostgres=# select count(*) from pg_stat_activity;\\n count\\n-------\\n 1\\n(1 row)\\n\\npostgres=# select dblink_connect('dbname=postgres');\\n dblink_connect\\n----------------\\n OK\\n(1 row)\\n\\npostgres=# select count(*) from pg_stat_activity;\\n count\\n-------\\n 2\\n(1 row)\\n\\npostgres=# select dblink_connect('dbname=postgres');\\n dblink_connect\\n----------------\\n OK\\n(1 row)\\n\\npostgres=# select count(*) from pg_stat_activity;\\n count\\n-------\\n 3\\n(1 row)\\n\\n\\nRegards\\nTakayuki Tsunakawa\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 06:52:26 +0000\",\n \"msg_from\": \"\\\"Tsunakawa, Takayuki\\\" <tsunakawa.takay@jp.fujitsu.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"[bug fix] dblink leaks unnamed connections\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\\n<tsunakawa.takay@jp.fujitsu.com> wrote:\\n> Hello,\\n>\\n> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\\n\\nThis issue was reported about ten years ago and added as TODO item.\\nhttp://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\\n\\nI agree that this is a bug, and am tempted to back-patch to all the supported\\nversions. But it had not been fixed in many years since the first report of\\nthe issue. So I'm not sure if it's ok to just treat this as a bug right now and\\nback-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\\nmore opinions about this.\\n\\nRegards,\\n\\n-- \\nFujii Masao\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 00:41:48 +0900\",\n \"msg_from\": \"Fujii Masao <masao.fujii@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [bug fix] dblink leaks unnamed connections\"\n },\n {\n \"msg_contents\": \"Fujii Masao <masao.fujii@gmail.com> writes:\\n> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\\n> <tsunakawa.takay@jp.fujitsu.com> wrote:\\n>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\\n\\n> This issue was reported about ten years ago and added as TODO item.\\n> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\\n\\n> I agree that this is a bug, and am tempted to back-patch to all the supported\\n> versions. But it had not been fixed in many years since the first report of\\n> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\\n> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\\n> more opinions about this.\\n\\nIt looks to me like the issue simply fell through the cracks because Joe\\nwasn't excited about fixing it. Now that we have a second complaint,\\nI think it's worth treating as a bug and back-patching.\\n\\n(I've not read this particular patch and am not expressing an opinion\\nwhether it's correct.)\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 09 Mar 2017 10:54:11 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [bug fix] dblink leaks unnamed connections\"\n },\n {\n \"msg_contents\": \"On 03/09/2017 07:54 AM, Tom Lane wrote:\\n> Fujii Masao <masao.fujii@gmail.com> writes:\\n>> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\\n>> <tsunakawa.takay@jp.fujitsu.com> wrote:\\n>>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\\n> \\n>> This issue was reported about ten years ago and added as TODO item.\\n>> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\\n> \\n>> I agree that this is a bug, and am tempted to back-patch to all the supported\\n>> versions. But it had not been fixed in many years since the first report of\\n>> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\\n>> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\\n>> more opinions about this.\\n> \\n> It looks to me like the issue simply fell through the cracks because Joe\\n> wasn't excited about fixing it. Now that we have a second complaint,\\n> I think it's worth treating as a bug and back-patching.\\n> \\n> (I've not read this particular patch and am not expressing an opinion\\n> whether it's correct.)\\n\\nOk, will take another look.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Thu, 9 Mar 2017 08:31:42 -0800\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [bug fix] dblink leaks unnamed connections\"\n },\n {\n \"msg_contents\": \"On 03/09/2017 08:31 AM, Joe Conway wrote:\\n> On 03/09/2017 07:54 AM, Tom Lane wrote:\\n>> Fujii Masao <masao.fujii@gmail.com> writes:\\n>>> On Wed, Mar 8, 2017 at 3:52 PM, Tsunakawa, Takayuki\\n>>> <tsunakawa.takay@jp.fujitsu.com> wrote:\\n>>>> dblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\\n>> \\n>>> This issue was reported about ten years ago and added as TODO item.\\n>>> http://archives.postgresql.org/pgsql-hackers/2007-10/msg00895.php\\n>> \\n>>> I agree that this is a bug, and am tempted to back-patch to all the supported\\n>>> versions. But it had not been fixed in many years since the first report of\\n>>> the issue. So I'm not sure if it's ok to just treat this as a bug right now and\\n>>> back-patch. Or we should fix this only in HEAD? Anyway I'd like to hear\\n>>> more opinions about this.\\n>> \\n>> It looks to me like the issue simply fell through the cracks because Joe\\n>> wasn't excited about fixing it. Now that we have a second complaint,\\n>> I think it's worth treating as a bug and back-patching.\\n>> \\n>> (I've not read this particular patch and am not expressing an opinion\\n>> whether it's correct.)\\n> \\n> Ok, will take another look.\\n\\nI pushed a fix to all supported branches.\\n\\nThanks,\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sat, 11 Mar 2017 13:37:36 -0800\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [bug fix] dblink leaks unnamed connections\"\n }\n]","isConversational":false}}},{"rowIdx":44087,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi,\n\nOver in another thread it was pointed out that a patch I submitted\nbroke a project rule by including \"postgres.h\" in a header. Here is a\npatch to remove it from dsa.h where I made the same mistake, and also\na case I found in bufmask.h by grepping.\n\nThere are also instances in regcustom.h and snowball's header.h -- are\nthose special cases?\n\n-- \nThomas Munro\nhttp://www.enterprisedb.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 8 Mar 2017 21:55:08 +1300","msg_from":"Thomas Munro <thomas.munro@enterprisedb.com>","msg_from_op":true,"msg_subject":"Removing #include \"postgres.h\" from a couple of headers"},{"msg_contents":"On Wed, Mar 8, 2017 at 5:55 PM, Thomas Munro\n<thomas.munro@enterprisedb.com> wrote:\n> Over in another thread it was pointed out that a patch I submitted\n> broke a project rule by including \"postgres.h\" in a header. Here is a\n> patch to remove it from dsa.h where I made the same mistake, and also\n> a case I found in bufmask.h by grepping.\n>\n> There are also instances in regcustom.h and snowball's header.h -- are\n> those special cases?\n\n--- a/src/include/access/bufmask.h\n+++ b/src/include/access/bufmask.h\n@@ -17,7 +17,6 @@\n #ifndef BUFMASK_H\n #define BUFMASK_H\n\n-#include \"postgres.h\"\n #include \"storage/block.h\"\n #include \"storage/bufmgr.h\"\nOops. This really escaped me...\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 8 Mar 2017 17:58:06 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: Removing #include \"postgres.h\" from a couple of headers"},{"msg_contents":"On Wed, Mar 8, 2017 at 3:55 AM, Thomas Munro\n<thomas.munro@enterprisedb.com> wrote:\n> Over in another thread it was pointed out that a patch I submitted\n> broke a project rule by including \"postgres.h\" in a header. Here is a\n> patch to remove it from dsa.h where I made the same mistake, and also\n> a case I found in bufmask.h by grepping.\n\nCommitted.\n\n> There are also instances in regcustom.h and snowball's header.h -- are\n> those special cases?\n\nI will leave this question to someone wiser (or more self-assured) than I.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 8 Mar 2017 08:25:27 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: Removing #include \"postgres.h\" from a couple of headers"},{"msg_contents":"Robert Haas <robertmhaas@gmail.com> writes:\n> On Wed, Mar 8, 2017 at 3:55 AM, Thomas Munro\n> <thomas.munro@enterprisedb.com> wrote:\n>> There are also instances in regcustom.h and snowball's header.h -- are\n>> those special cases?\n\n> I will leave this question to someone wiser (or more self-assured) than I.\n\nI'm pretty sure I'm to blame for both of those special cases. The genesis\nof both is that we are including these headers from externally-generated\n.c files, and it seemed like modifying the .c files would be a bigger\nproblem than violating the policy. I am not sure if I hold that position\nanymore for the regexp library; our copy has diverged substantially from\nTcl's anyway. It's still an issue for Snowball, because those .c files\nare actually machine-generated by a Snowball-to-C compiler. We haven't\nmodified them and probably shouldn't.\n\nIf we don't change the code layout, we should probably at least add\ncomments near these postgres.h inclusions explaining why they're violating\npolicy.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 08 Mar 2017 09:33:55 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: Removing #include \"postgres.h\" from a couple of headers"},{"msg_contents":"... BTW, a bit of grepping shows that there are still significant\nviolations of this policy with respect to header-driven inclusion of\npostgres_fe.h and c.h. Also, plpgsql is doing it in the unapproved\nfashion. Cleaning these up will take a bit more work, since we'll have\nto actually add #includes to some .c files that are currently relying on\nthose headers to get the core header. However, if I believe my own\nargument, then that's a good thing and we'd better go do it.\n\nAlso, ecpglib.h seems like a complete mess: it's relying on symbols\nlike ENABLE_NLS but there's no certainty as to whether pg_config.h\nhas been included first. It won't have been in the case where this\nheader is read from an ecpg-generated .c file. So this header will\nin fact be interpreted differently in ecpg-generated programs than\nin ecpglib and ecpg itself. Maybe this is okay but it sure smells\nlike trouble waiting to happen. I have no desire to try to fix it\nmyself though.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 08 Mar 2017 09:53:50 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: Removing #include \"postgres.h\" from a couple of headers"}],"string":"[\n {\n \"msg_contents\": \"Hi,\\n\\nOver in another thread it was pointed out that a patch I submitted\\nbroke a project rule by including \\\"postgres.h\\\" in a header. Here is a\\npatch to remove it from dsa.h where I made the same mistake, and also\\na case I found in bufmask.h by grepping.\\n\\nThere are also instances in regcustom.h and snowball's header.h -- are\\nthose special cases?\\n\\n-- \\nThomas Munro\\nhttp://www.enterprisedb.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 21:55:08 +1300\",\n \"msg_from\": \"Thomas Munro <thomas.munro@enterprisedb.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Removing #include \\\"postgres.h\\\" from a couple of headers\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 5:55 PM, Thomas Munro\\n<thomas.munro@enterprisedb.com> wrote:\\n> Over in another thread it was pointed out that a patch I submitted\\n> broke a project rule by including \\\"postgres.h\\\" in a header. Here is a\\n> patch to remove it from dsa.h where I made the same mistake, and also\\n> a case I found in bufmask.h by grepping.\\n>\\n> There are also instances in regcustom.h and snowball's header.h -- are\\n> those special cases?\\n\\n--- a/src/include/access/bufmask.h\\n+++ b/src/include/access/bufmask.h\\n@@ -17,7 +17,6 @@\\n #ifndef BUFMASK_H\\n #define BUFMASK_H\\n\\n-#include \\\"postgres.h\\\"\\n #include \\\"storage/block.h\\\"\\n #include \\\"storage/bufmgr.h\\\"\\nOops. This really escaped me...\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 8 Mar 2017 17:58:06 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Removing #include \\\"postgres.h\\\" from a couple of headers\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 3:55 AM, Thomas Munro\\n<thomas.munro@enterprisedb.com> wrote:\\n> Over in another thread it was pointed out that a patch I submitted\\n> broke a project rule by including \\\"postgres.h\\\" in a header. Here is a\\n> patch to remove it from dsa.h where I made the same mistake, and also\\n> a case I found in bufmask.h by grepping.\\n\\nCommitted.\\n\\n> There are also instances in regcustom.h and snowball's header.h -- are\\n> those special cases?\\n\\nI will leave this question to someone wiser (or more self-assured) than I.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 8 Mar 2017 08:25:27 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Removing #include \\\"postgres.h\\\" from a couple of headers\"\n },\n {\n \"msg_contents\": \"Robert Haas <robertmhaas@gmail.com> writes:\\n> On Wed, Mar 8, 2017 at 3:55 AM, Thomas Munro\\n> <thomas.munro@enterprisedb.com> wrote:\\n>> There are also instances in regcustom.h and snowball's header.h -- are\\n>> those special cases?\\n\\n> I will leave this question to someone wiser (or more self-assured) than I.\\n\\nI'm pretty sure I'm to blame for both of those special cases. The genesis\\nof both is that we are including these headers from externally-generated\\n.c files, and it seemed like modifying the .c files would be a bigger\\nproblem than violating the policy. I am not sure if I hold that position\\nanymore for the regexp library; our copy has diverged substantially from\\nTcl's anyway. It's still an issue for Snowball, because those .c files\\nare actually machine-generated by a Snowball-to-C compiler. We haven't\\nmodified them and probably shouldn't.\\n\\nIf we don't change the code layout, we should probably at least add\\ncomments near these postgres.h inclusions explaining why they're violating\\npolicy.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 08 Mar 2017 09:33:55 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Removing #include \\\"postgres.h\\\" from a couple of headers\"\n },\n {\n \"msg_contents\": \"... BTW, a bit of grepping shows that there are still significant\\nviolations of this policy with respect to header-driven inclusion of\\npostgres_fe.h and c.h. Also, plpgsql is doing it in the unapproved\\nfashion. Cleaning these up will take a bit more work, since we'll have\\nto actually add #includes to some .c files that are currently relying on\\nthose headers to get the core header. However, if I believe my own\\nargument, then that's a good thing and we'd better go do it.\\n\\nAlso, ecpglib.h seems like a complete mess: it's relying on symbols\\nlike ENABLE_NLS but there's no certainty as to whether pg_config.h\\nhas been included first. It won't have been in the case where this\\nheader is read from an ecpg-generated .c file. So this header will\\nin fact be interpreted differently in ecpg-generated programs than\\nin ecpglib and ecpg itself. Maybe this is okay but it sure smells\\nlike trouble waiting to happen. I have no desire to try to fix it\\nmyself though.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 08 Mar 2017 09:53:50 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Removing #include \\\"postgres.h\\\" from a couple of headers\"\n }\n]","isConversational":false}}},{"rowIdx":44088,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com>\nwrote:\n\n> I am attaching another version of the patch, as I found stupid mistake\n> in the earlier version of patch, where I missed to initialize initial\n> value to\n> WaitEventIO enum. Also earlier version was not getting cleanly apply on\n> the current version of sources.\n>\n\nI have applied attached patch, set shared_buffers to 128kB and ran pgbench,\nI am able to see below distinct IO events.\n\n--with ./pgbench -i -s 500 postgres\napplication_name wait_event_type wait_event query\n pgbench IO ExtendDataBlock\ncopy pgbench_account\n pgbench IO WriteXLog\n copy pgbench_account\n pgbench IO WriteDataBlock\n copy pgbench_account\n pgbench IO ReadDataBlock\n vacuum analyze pgben\n pgbench IO ReadBuffile\n alter table pgbench_\n\n--with ./pgbench -T 600 postgres (readwrite)\napplication_name wait_event_type wait_event query\n pgbench IO ReadDataBlock\n UPDATE pgbench_accou\n pgbench IO WriteDataBlock\nUPDATE pgbench_telle\n IO SyncDataBlock\n\n pgbench IO SyncDataBlock\n UPDATE pgbench_telle\n IO SyncDataBlock\n autovacuum: VACUUM A\n pgbench IO WriteXLog\n END;\n pgbench IO ExtendDataBlock\ncopy pgbench_account\n\n--with ./pgbench -T 600 -S postgres (select only)\napplication_name wait_event_type wait_event query\n pgbench IO ReadDataBlock\n SELECT abalance FROM\n\nAttached excel with all IO event values.\n\nThanks & Regards,\nRajkumar Raghuwanshi\nQMG, EnterpriseDB Corporation\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 8 Mar 2017 18:41:13 +0530","msg_from":"Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>","msg_from_op":true,"msg_subject":"Re: wait events for disk I/O"},{"msg_contents":"Thanks Rajkumar for performing tests on this patch.\n\nYes, I also noticed similar results in my testing. Additionally sometime I\nalso\nnoticed ReadSLRUPage event on my system.\n\nI also run the reindex database command and I notices below IO events.\n\nSyncImmedRelation,\nWriteDataBlock\nWriteBuffile,\nWriteXLog,\nReadDataBlock\n\n\n\nOn Wed, Mar 8, 2017 at 6:41 PM, Rajkumar Raghuwanshi <\nrajkumar.raghuwanshi@enterprisedb.com> wrote:\n\n> On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com>\n> wrote:\n>\n>> I am attaching another version of the patch, as I found stupid mistake\n>> in the earlier version of patch, where I missed to initialize initial\n>> value to\n>> WaitEventIO enum. Also earlier version was not getting cleanly apply on\n>> the current version of sources.\n>>\n>\n> I have applied attached patch, set shared_buffers to 128kB and ran\n> pgbench, I am able to see below distinct IO events.\n>\n> --with ./pgbench -i -s 500 postgres\n> application_name wait_event_type wait_event query\n> pgbench IO ExtendDataBlock\n> copy pgbench_account\n> pgbench IO WriteXLog\n> copy pgbench_account\n> pgbench IO WriteDataBlock\n> copy pgbench_account\n> pgbench IO ReadDataBlock\n> vacuum analyze pgben\n> pgbench IO ReadBuffile\n> alter table pgbench_\n>\n> --with ./pgbench -T 600 postgres (readwrite)\n> application_name wait_event_type wait_event query\n> pgbench IO ReadDataBlock\n> UPDATE pgbench_accou\n> pgbench IO WriteDataBlock\n> UPDATE pgbench_telle\n> IO SyncDataBlock\n>\n> pgbench IO SyncDataBlock\n> UPDATE pgbench_telle\n> IO SyncDataBlock\n> autovacuum: VACUUM A\n> pgbench IO WriteXLog\n> END;\n> pgbench IO ExtendDataBlock\n> copy pgbench_account\n>\n> --with ./pgbench -T 600 -S postgres (select only)\n> application_name wait_event_type wait_event query\n> pgbench IO ReadDataBlock\n> SELECT abalance FROM\n>\n> Attached excel with all IO event values.\n>\n> Thanks & Regards,\n> Rajkumar Raghuwanshi\n> QMG, EnterpriseDB Corporation\n>\n>\n>\n\n\n-- \nRushabh Lathia\nwww.EnterpriseDB.com\n\nThanks Rajkumar for performing tests on this patch. Yes, I also noticed similar results in my testing. Additionally sometime I alsonoticed ReadSLRUPage event on my system.I also run the reindex database command and I notices below IO events.SyncImmedRelation,WriteDataBlockWriteBuffile,WriteXLog,ReadDataBlockOn Wed, Mar 8, 2017 at 6:41 PM, Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com> wrote:On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com> wrote:I am attaching another version of the patch, as I found stupid mistakein the earlier version of patch, where I missed to initialize initial value toWaitEventIO enum. Also earlier version was not getting cleanly apply onthe current version of sources.I have applied attached patch, set shared_buffers to 128kB and ran pgbench, I am able to see below distinct IO events.--with ./pgbench -i -s 500 postgresapplication_name wait_event_type wait_event query pgbench IO ExtendDataBlock copy pgbench_account pgbench IO WriteXLog copy pgbench_account pgbench IO WriteDataBlock copy pgbench_account pgbench IO ReadDataBlock vacuum analyze pgben pgbench IO ReadBuffile alter table pgbench_--with ./pgbench -T 600 postgres (readwrite)application_name wait_event_type wait_event query pgbench IO ReadDataBlock UPDATE pgbench_accou pgbench IO WriteDataBlock UPDATE pgbench_telle IO SyncDataBlock pgbench IO SyncDataBlock UPDATE pgbench_telle IO SyncDataBlock autovacuum: VACUUM A pgbench IO WriteXLog END; pgbench IO ExtendDataBlock copy pgbench_account--with ./pgbench -T 600 -S postgres (select only)application_name wait_event_type wait_event query pgbench IO ReadDataBlock SELECT abalance FROMAttached excel with all IO event values.Thanks & Regards,Rajkumar RaghuwanshiQMG, EnterpriseDB Corporation\n-- Rushabh Lathiawww.EnterpriseDB.com","msg_date":"Thu, 9 Mar 2017 10:54:51 +0530","msg_from":"Rushabh Lathia <rushabh.lathia@gmail.com>","msg_from_op":false,"msg_subject":"Re: wait events for disk I/O"},{"msg_contents":"On Thu, Mar 9, 2017 at 10:54 AM, Rushabh Lathia\n<rushabh.lathia@gmail.com> wrote:\n> Thanks Rajkumar for performing tests on this patch.\n>\n> Yes, I also noticed similar results in my testing. Additionally sometime I\n> also\n> noticed ReadSLRUPage event on my system.\n>\n> I also run the reindex database command and I notices below IO events.\n>\n> SyncImmedRelation,\n> WriteDataBlock\n> WriteBuffile,\n> WriteXLog,\n> ReadDataBlock\n\nI have tried for generating some more events, by running pgbench on\nmaster/slave configuration, able to see three more events.\nWriteInitXLogFile\nSyncInitXLogFile\nReadXLog\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 13 Mar 2017 10:54:19 +0530","msg_from":"Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>","msg_from_op":true,"msg_subject":"Re: wait events for disk I/O"}],"string":"[\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com>\\nwrote:\\n\\n> I am attaching another version of the patch, as I found stupid mistake\\n> in the earlier version of patch, where I missed to initialize initial\\n> value to\\n> WaitEventIO enum. Also earlier version was not getting cleanly apply on\\n> the current version of sources.\\n>\\n\\nI have applied attached patch, set shared_buffers to 128kB and ran pgbench,\\nI am able to see below distinct IO events.\\n\\n--with ./pgbench -i -s 500 postgres\\napplication_name wait_event_type wait_event query\\n pgbench IO ExtendDataBlock\\ncopy pgbench_account\\n pgbench IO WriteXLog\\n copy pgbench_account\\n pgbench IO WriteDataBlock\\n copy pgbench_account\\n pgbench IO ReadDataBlock\\n vacuum analyze pgben\\n pgbench IO ReadBuffile\\n alter table pgbench_\\n\\n--with ./pgbench -T 600 postgres (readwrite)\\napplication_name wait_event_type wait_event query\\n pgbench IO ReadDataBlock\\n UPDATE pgbench_accou\\n pgbench IO WriteDataBlock\\nUPDATE pgbench_telle\\n IO SyncDataBlock\\n\\n pgbench IO SyncDataBlock\\n UPDATE pgbench_telle\\n IO SyncDataBlock\\n autovacuum: VACUUM A\\n pgbench IO WriteXLog\\n END;\\n pgbench IO ExtendDataBlock\\ncopy pgbench_account\\n\\n--with ./pgbench -T 600 -S postgres (select only)\\napplication_name wait_event_type wait_event query\\n pgbench IO ReadDataBlock\\n SELECT abalance FROM\\n\\nAttached excel with all IO event values.\\n\\nThanks & Regards,\\nRajkumar Raghuwanshi\\nQMG, EnterpriseDB Corporation\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 18:41:13 +0530\",\n \"msg_from\": \"Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: wait events for disk I/O\"\n },\n {\n \"msg_contents\": \"Thanks Rajkumar for performing tests on this patch.\\n\\nYes, I also noticed similar results in my testing. Additionally sometime I\\nalso\\nnoticed ReadSLRUPage event on my system.\\n\\nI also run the reindex database command and I notices below IO events.\\n\\nSyncImmedRelation,\\nWriteDataBlock\\nWriteBuffile,\\nWriteXLog,\\nReadDataBlock\\n\\n\\n\\nOn Wed, Mar 8, 2017 at 6:41 PM, Rajkumar Raghuwanshi <\\nrajkumar.raghuwanshi@enterprisedb.com> wrote:\\n\\n> On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com>\\n> wrote:\\n>\\n>> I am attaching another version of the patch, as I found stupid mistake\\n>> in the earlier version of patch, where I missed to initialize initial\\n>> value to\\n>> WaitEventIO enum. Also earlier version was not getting cleanly apply on\\n>> the current version of sources.\\n>>\\n>\\n> I have applied attached patch, set shared_buffers to 128kB and ran\\n> pgbench, I am able to see below distinct IO events.\\n>\\n> --with ./pgbench -i -s 500 postgres\\n> application_name wait_event_type wait_event query\\n> pgbench IO ExtendDataBlock\\n> copy pgbench_account\\n> pgbench IO WriteXLog\\n> copy pgbench_account\\n> pgbench IO WriteDataBlock\\n> copy pgbench_account\\n> pgbench IO ReadDataBlock\\n> vacuum analyze pgben\\n> pgbench IO ReadBuffile\\n> alter table pgbench_\\n>\\n> --with ./pgbench -T 600 postgres (readwrite)\\n> application_name wait_event_type wait_event query\\n> pgbench IO ReadDataBlock\\n> UPDATE pgbench_accou\\n> pgbench IO WriteDataBlock\\n> UPDATE pgbench_telle\\n> IO SyncDataBlock\\n>\\n> pgbench IO SyncDataBlock\\n> UPDATE pgbench_telle\\n> IO SyncDataBlock\\n> autovacuum: VACUUM A\\n> pgbench IO WriteXLog\\n> END;\\n> pgbench IO ExtendDataBlock\\n> copy pgbench_account\\n>\\n> --with ./pgbench -T 600 -S postgres (select only)\\n> application_name wait_event_type wait_event query\\n> pgbench IO ReadDataBlock\\n> SELECT abalance FROM\\n>\\n> Attached excel with all IO event values.\\n>\\n> Thanks & Regards,\\n> Rajkumar Raghuwanshi\\n> QMG, EnterpriseDB Corporation\\n>\\n>\\n>\\n\\n\\n-- \\nRushabh Lathia\\nwww.EnterpriseDB.com\\n\\nThanks Rajkumar for performing tests on this patch. Yes, I also noticed similar results in my testing. Additionally sometime I alsonoticed ReadSLRUPage event on my system.I also run the reindex database command and I notices below IO events.SyncImmedRelation,WriteDataBlockWriteBuffile,WriteXLog,ReadDataBlockOn Wed, Mar 8, 2017 at 6:41 PM, Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com> wrote:On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com> wrote:I am attaching another version of the patch, as I found stupid mistakein the earlier version of patch, where I missed to initialize initial value toWaitEventIO enum. Also earlier version was not getting cleanly apply onthe current version of sources.I have applied attached patch, set shared_buffers to 128kB and ran pgbench, I am able to see below distinct IO events.--with ./pgbench -i -s 500 postgresapplication_name wait_event_type wait_event query pgbench IO ExtendDataBlock copy pgbench_account pgbench IO WriteXLog copy pgbench_account pgbench IO WriteDataBlock copy pgbench_account pgbench IO ReadDataBlock vacuum analyze pgben pgbench IO ReadBuffile alter table pgbench_--with ./pgbench -T 600 postgres (readwrite)application_name wait_event_type wait_event query pgbench IO ReadDataBlock UPDATE pgbench_accou pgbench IO WriteDataBlock UPDATE pgbench_telle IO SyncDataBlock pgbench IO SyncDataBlock UPDATE pgbench_telle IO SyncDataBlock autovacuum: VACUUM A pgbench IO WriteXLog END; pgbench IO ExtendDataBlock copy pgbench_account--with ./pgbench -T 600 -S postgres (select only)application_name wait_event_type wait_event query pgbench IO ReadDataBlock SELECT abalance FROMAttached excel with all IO event values.Thanks & Regards,Rajkumar RaghuwanshiQMG, EnterpriseDB Corporation\\n-- Rushabh Lathiawww.EnterpriseDB.com\",\n \"msg_date\": \"Thu, 9 Mar 2017 10:54:51 +0530\",\n \"msg_from\": \"Rushabh Lathia <rushabh.lathia@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: wait events for disk I/O\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 10:54 AM, Rushabh Lathia\\n<rushabh.lathia@gmail.com> wrote:\\n> Thanks Rajkumar for performing tests on this patch.\\n>\\n> Yes, I also noticed similar results in my testing. Additionally sometime I\\n> also\\n> noticed ReadSLRUPage event on my system.\\n>\\n> I also run the reindex database command and I notices below IO events.\\n>\\n> SyncImmedRelation,\\n> WriteDataBlock\\n> WriteBuffile,\\n> WriteXLog,\\n> ReadDataBlock\\n\\nI have tried for generating some more events, by running pgbench on\\nmaster/slave configuration, able to see three more events.\\nWriteInitXLogFile\\nSyncInitXLogFile\\nReadXLog\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 13 Mar 2017 10:54:19 +0530\",\n \"msg_from\": \"Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: wait events for disk I/O\"\n }\n]","isConversational":false}}},{"rowIdx":44089,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Fix parallel index and index-only scans to fall back to serial.\n\nParallel executor nodes can't assume that parallel execution will\nhappen in every case where the plan calls for it, because it might\nnot work out that way. However, parallel index scan and parallel\nindex-only scan failed to do the right thing here. Repair.\n\nAmit Kapila, per a report from me.\n\nDiscussion: http://postgr.es/m/CAA4eK1Kq5qb_u2AOoda5XBB91vVWz90w=LgtRLgsssriS8pVTw@mail.gmail.com\n\nBranch\n------\nmaster\n\nDetails\n-------\nhttp://git.postgresql.org/pg/commitdiff/09529a70bb5a77935d086d651c63396767d240d7\n\nModified Files\n--------------\nsrc/backend/executor/nodeIndexonlyscan.c | 72 +++++++++++++------------\nsrc/backend/executor/nodeIndexscan.c | 90 ++++++++++++++++++++------------\n2 files changed, 96 insertions(+), 66 deletions(-)\n\n\n-- \nSent via pgsql-committers mailing list (pgsql-committers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-committers\n","msg_date":"Wed, 08 Mar 2017 13:17:36 +0000","msg_from":"Robert Haas <rhaas@postgresql.org>","msg_from_op":true,"msg_subject":"pgsql: Fix parallel index and index-only scans to fall back to\n serial."},{"msg_contents":"Hi!\n\nI just bumped into this comment, from commit 09529a70bb5, and I can't \nmake sense of it:\n\n> + /*\n> + * We reach here if the index only scan is not parallel, or if we're\n> + * executing a index only scan that was intended to be parallel\n> + * serially.\n> + */\n\nWhat was that intended to say?\n\n- Heikki\n\n","msg_date":"Fri, 13 Jul 2018 22:22:42 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: pgsql: Fix parallel index and index-only scans to fall back to\n serial."},{"msg_contents":"On Fri, Jul 13, 2018 at 12:22 PM, Heikki Linnakangas <hlinnaka@iki.fi>\nwrote:\n\n> Hi!\n>\n> I just bumped into this comment, from commit 09529a70bb5, and I can't make\n> sense of it:\n>\n> + /*\n>> + * We reach here if the index only scan is not parallel,\n>> or if we're\n>> + * executing a index only scan that was intended to be\n>> parallel\n>> + * serially.\n>> + */\n>>\n>\n> What was that intended to say?\n>\n>\nI think...\n\n...or if we're serially executing a[n] index only scan that was intended\nto be [executed in] parallel\n\ns/intended/planned/ ?\n\nDavid J.\n\nOn Fri, Jul 13, 2018 at 12:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:Hi!\n\nI just bumped into this comment, from commit 09529a70bb5, and I can't make sense of it:\n\n\n+ /*\n+ * We reach here if the index only scan is not parallel, or if we're\n+ * executing a index only scan that was intended to be parallel\n+ * serially.\n+ */\n\n\nWhat was that intended to say?I think......or if we're serially executing a[n] index only scan that was intended to be [executed in] parallels/intended/planned/ ?David J.","msg_date":"Fri, 13 Jul 2018 14:19:29 -0700","msg_from":"\"David G. Johnston\" <david.g.johnston@gmail.com>","msg_from_op":false,"msg_subject":"Re: pgsql: Fix parallel index and index-only scans to fall back to\n serial."},{"msg_contents":"On Fri, Jul 13, 2018 at 2:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n> I just bumped into this comment, from commit 09529a70bb5, and I can't make\n> sense of it:\n>\n>> + /*\n>> + * We reach here if the index only scan is not parallel,\n>> or if we're\n>> + * executing a index only scan that was intended to be\n>> parallel\n>> + * serially.\n>> + */\n>\n>\n> What was that intended to say?\n\nThere are two ways that you can reach that code. One is that you have\nthe thing that shows up in EXPLAIN output as \"Index-Only Scan\". The\nother is that you have the thing that shows up in EXPLAIN output as\n\"Parallel Index-Only Scan\", but you didn't get any workers, so now\nyou're falling back to running what was intended to be a parallel plan\nwithout parallelism i.e. serially. The comment is intended to alert\nyou to the fact that an intended-as-parallel scan can end up here in\ncorner cases where the plan doesn't end up being parallel. We've had\nsome difficulty in consistently getting that case correct.\n\nIf you decide to rephase the comment for clarity, note that there are\nthree other near-copies of it cf. git grep -C4 'We reach here if'\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n","msg_date":"Fri, 13 Jul 2018 16:56:40 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: pgsql: Fix parallel index and index-only scans to fall back to\n serial."},{"msg_contents":"On 14/07/18 00:56, Robert Haas wrote:\n> On Fri, Jul 13, 2018 at 2:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\n>> I just bumped into this comment, from commit 09529a70bb5, and I can't make\n>> sense of it:\n>>\n>>> + /*\n>>> + * We reach here if the index only scan is not parallel,\n>>> or if we're\n>>> + * executing a index only scan that was intended to be\n>>> parallel\n>>> + * serially.\n>>> + */\n>>\n>>\n>> What was that intended to say?\n> \n> There are two ways that you can reach that code. One is that you have\n> the thing that shows up in EXPLAIN output as \"Index-Only Scan\". The\n> other is that you have the thing that shows up in EXPLAIN output as\n> \"Parallel Index-Only Scan\", but you didn't get any workers, so now\n> you're falling back to running what was intended to be a parallel plan\n> without parallelism i.e. serially. The comment is intended to alert\n> you to the fact that an intended-as-parallel scan can end up here in\n> corner cases where the plan doesn't end up being parallel. We've had\n> some difficulty in consistently getting that case correct.\n\nAh, gotcha. I changed that (and the other copies) per David's suggestion.\n\n- Heikki\n\n","msg_date":"Thu, 19 Jul 2018 16:13:18 +0300","msg_from":"Heikki Linnakangas <hlinnaka@iki.fi>","msg_from_op":false,"msg_subject":"Re: pgsql: Fix parallel index and index-only scans to fall back to\n serial."}],"string":"[\n {\n \"msg_contents\": \"Fix parallel index and index-only scans to fall back to serial.\\n\\nParallel executor nodes can't assume that parallel execution will\\nhappen in every case where the plan calls for it, because it might\\nnot work out that way. However, parallel index scan and parallel\\nindex-only scan failed to do the right thing here. Repair.\\n\\nAmit Kapila, per a report from me.\\n\\nDiscussion: http://postgr.es/m/CAA4eK1Kq5qb_u2AOoda5XBB91vVWz90w=LgtRLgsssriS8pVTw@mail.gmail.com\\n\\nBranch\\n------\\nmaster\\n\\nDetails\\n-------\\nhttp://git.postgresql.org/pg/commitdiff/09529a70bb5a77935d086d651c63396767d240d7\\n\\nModified Files\\n--------------\\nsrc/backend/executor/nodeIndexonlyscan.c | 72 +++++++++++++------------\\nsrc/backend/executor/nodeIndexscan.c | 90 ++++++++++++++++++++------------\\n2 files changed, 96 insertions(+), 66 deletions(-)\\n\\n\\n-- \\nSent via pgsql-committers mailing list (pgsql-committers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-committers\\n\",\n \"msg_date\": \"Wed, 08 Mar 2017 13:17:36 +0000\",\n \"msg_from\": \"Robert Haas <rhaas@postgresql.org>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"pgsql: Fix parallel index and index-only scans to fall back to\\n serial.\"\n },\n {\n \"msg_contents\": \"Hi!\\n\\nI just bumped into this comment, from commit 09529a70bb5, and I can't \\nmake sense of it:\\n\\n> + /*\\n> + * We reach here if the index only scan is not parallel, or if we're\\n> + * executing a index only scan that was intended to be parallel\\n> + * serially.\\n> + */\\n\\nWhat was that intended to say?\\n\\n- Heikki\\n\\n\",\n \"msg_date\": \"Fri, 13 Jul 2018 22:22:42 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: pgsql: Fix parallel index and index-only scans to fall back to\\n serial.\"\n },\n {\n \"msg_contents\": \"On Fri, Jul 13, 2018 at 12:22 PM, Heikki Linnakangas <hlinnaka@iki.fi>\\nwrote:\\n\\n> Hi!\\n>\\n> I just bumped into this comment, from commit 09529a70bb5, and I can't make\\n> sense of it:\\n>\\n> + /*\\n>> + * We reach here if the index only scan is not parallel,\\n>> or if we're\\n>> + * executing a index only scan that was intended to be\\n>> parallel\\n>> + * serially.\\n>> + */\\n>>\\n>\\n> What was that intended to say?\\n>\\n>\\nI think...\\n\\n...or if we're serially executing a[n] index only scan that was intended\\nto be [executed in] parallel\\n\\ns/intended/planned/ ?\\n\\nDavid J.\\n\\nOn Fri, Jul 13, 2018 at 12:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:Hi!\\n\\nI just bumped into this comment, from commit 09529a70bb5, and I can't make sense of it:\\n\\n\\n+ /*\\n+ * We reach here if the index only scan is not parallel, or if we're\\n+ * executing a index only scan that was intended to be parallel\\n+ * serially.\\n+ */\\n\\n\\nWhat was that intended to say?I think......or if we're serially executing a[n] index only scan that was intended to be [executed in] parallels/intended/planned/ ?David J.\",\n \"msg_date\": \"Fri, 13 Jul 2018 14:19:29 -0700\",\n \"msg_from\": \"\\\"David G. Johnston\\\" <david.g.johnston@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: pgsql: Fix parallel index and index-only scans to fall back to\\n serial.\"\n },\n {\n \"msg_contents\": \"On Fri, Jul 13, 2018 at 2:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n> I just bumped into this comment, from commit 09529a70bb5, and I can't make\\n> sense of it:\\n>\\n>> + /*\\n>> + * We reach here if the index only scan is not parallel,\\n>> or if we're\\n>> + * executing a index only scan that was intended to be\\n>> parallel\\n>> + * serially.\\n>> + */\\n>\\n>\\n> What was that intended to say?\\n\\nThere are two ways that you can reach that code. One is that you have\\nthe thing that shows up in EXPLAIN output as \\\"Index-Only Scan\\\". The\\nother is that you have the thing that shows up in EXPLAIN output as\\n\\\"Parallel Index-Only Scan\\\", but you didn't get any workers, so now\\nyou're falling back to running what was intended to be a parallel plan\\nwithout parallelism i.e. serially. The comment is intended to alert\\nyou to the fact that an intended-as-parallel scan can end up here in\\ncorner cases where the plan doesn't end up being parallel. We've had\\nsome difficulty in consistently getting that case correct.\\n\\nIf you decide to rephase the comment for clarity, note that there are\\nthree other near-copies of it cf. git grep -C4 'We reach here if'\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\",\n \"msg_date\": \"Fri, 13 Jul 2018 16:56:40 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: pgsql: Fix parallel index and index-only scans to fall back to\\n serial.\"\n },\n {\n \"msg_contents\": \"On 14/07/18 00:56, Robert Haas wrote:\\n> On Fri, Jul 13, 2018 at 2:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:\\n>> I just bumped into this comment, from commit 09529a70bb5, and I can't make\\n>> sense of it:\\n>>\\n>>> + /*\\n>>> + * We reach here if the index only scan is not parallel,\\n>>> or if we're\\n>>> + * executing a index only scan that was intended to be\\n>>> parallel\\n>>> + * serially.\\n>>> + */\\n>>\\n>>\\n>> What was that intended to say?\\n> \\n> There are two ways that you can reach that code. One is that you have\\n> the thing that shows up in EXPLAIN output as \\\"Index-Only Scan\\\". The\\n> other is that you have the thing that shows up in EXPLAIN output as\\n> \\\"Parallel Index-Only Scan\\\", but you didn't get any workers, so now\\n> you're falling back to running what was intended to be a parallel plan\\n> without parallelism i.e. serially. The comment is intended to alert\\n> you to the fact that an intended-as-parallel scan can end up here in\\n> corner cases where the plan doesn't end up being parallel. We've had\\n> some difficulty in consistently getting that case correct.\\n\\nAh, gotcha. I changed that (and the other copies) per David's suggestion.\\n\\n- Heikki\\n\\n\",\n \"msg_date\": \"Thu, 19 Jul 2018 16:13:18 +0300\",\n \"msg_from\": \"Heikki Linnakangas <hlinnaka@iki.fi>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: pgsql: Fix parallel index and index-only scans to fall back to\\n serial.\"\n }\n]","isConversational":false}}},{"rowIdx":44090,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Here's a patch which updates parallel.sgml for the new parallel scan\ntypes which have recently been committed, and also expands the\nexplanation of parallel joins slightly. I hope everyone will find\nthis useful in understanding what new capabilities we now have, and\nwhat remains to be done in the future.\n\nBarring objections, I plan to commit this tomorrow.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 8 Mar 2017 12:46:42 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":true,"msg_subject":"updating parallel.sgml"},{"msg_contents":"On Wed, Mar 8, 2017 at 12:46 PM, Robert Haas <robertmhaas@gmail.com> wrote:\n> Here's a patch which updates parallel.sgml for the new parallel scan\n> types which have recently been committed, and also expands the\n> explanation of parallel joins slightly. I hope everyone will find\n> this useful in understanding what new capabilities we now have, and\n> what remains to be done in the future.\n>\n> Barring objections, I plan to commit this tomorrow.\n\nCommitted with a fix for a typo pointed out to me off-list.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 13:06:49 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":true,"msg_subject":"Re: updating parallel.sgml"}],"string":"[\n {\n \"msg_contents\": \"Here's a patch which updates parallel.sgml for the new parallel scan\\ntypes which have recently been committed, and also expands the\\nexplanation of parallel joins slightly. I hope everyone will find\\nthis useful in understanding what new capabilities we now have, and\\nwhat remains to be done in the future.\\n\\nBarring objections, I plan to commit this tomorrow.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 12:46:42 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"updating parallel.sgml\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 12:46 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> Here's a patch which updates parallel.sgml for the new parallel scan\\n> types which have recently been committed, and also expands the\\n> explanation of parallel joins slightly. I hope everyone will find\\n> this useful in understanding what new capabilities we now have, and\\n> what remains to be done in the future.\\n>\\n> Barring objections, I plan to commit this tomorrow.\\n\\nCommitted with a fix for a typo pointed out to me off-list.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 13:06:49 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: updating parallel.sgml\"\n }\n]","isConversational":false}}},{"rowIdx":44091,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi all\n\nI've found a minor off-by-one error in the resource retention logic\nfor logical slots, where we treat confirmed_flush as meaning \"flushed\nup to and including this LSN\". Seems reasonable, but the rest of the\ncode treats it as \"flushed up to but excluding this LSN\". In\nparticular, we treat confirmed_flush as the inclusive startpoint for a\nnew logical decoding session if no startpoint is provided by the\nclient and will replay a change whose commit record begins at exactly\nconfirmed_flush to the client.\n\nThis issue was identified while debugging an issue where duplicate\nrows were replicated after unclean shutdown by logical replication on\na table with no PK.\n\nI'd prefer to make confirmed_flush mean \"confirmed flushed up to and\nincluding\" everywhere, but the knock-on effects are too ugly. In\nparticular we'd then be changing the meaning of START_REPLICATION ...\nLOGICAL ... 's argument LSN to be \"start replay of commits after, but\nnot including, this LSN\". We can't just adjust it by 1, otherwise\nsuddenly we'd get different results if we passed the confirmed_flush\nvalue explicitly vs passing 0/0 and letting it be picked up\nimplicitly.\n\nTwo further minor patches follow, one to fix a harmless but confusing\nquirk in logical walsender init, and one that adds explanatory\ncomments to relevant parts of the code (and a couple of spots in the\ndocs) to avoid future occurrences of the above issues.\n\n-- \n Craig Ringer http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 9 Mar 2017 11:17:05 +0800","msg_from":"Craig Ringer <craig@2ndquadrant.com>","msg_from_op":true,"msg_subject":"[PATCH] Off-by-one error in logical slot resource retention"},{"msg_contents":"On 9 March 2017 at 11:17, Craig Ringer <craig@2ndquadrant.com> wrote:\n> Hi all\n>\n> I've found a minor off-by-one error in the resource retention logic\n> for logical slots, where we treat confirmed_flush as meaning \"flushed\n> up to and including this LSN\". Seems reasonable, but the rest of the\n> code treats it as \"flushed up to but excluding this LSN\". In\n> particular, we treat confirmed_flush as the inclusive startpoint for a\n> new logical decoding session if no startpoint is provided by the\n> client and will replay a change whose commit record begins at exactly\n> confirmed_flush to the client.\n\nThis should be backpatched to 9.4.\n\nThe others do not need backpatching.\n\n-- \n Craig Ringer http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 11:18:30 +0800","msg_from":"Craig Ringer <craig@2ndquadrant.com>","msg_from_op":true,"msg_subject":"Re: [PATCH] Off-by-one error in logical slot resource retention"},{"msg_contents":"The following review has been posted through the commitfest application:\nmake installcheck-world: not tested\nImplements feature: not tested\nSpec compliant: not tested\nDocumentation: not tested\n\nHi Craig,\r\n\r\nI'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\n\nThe new status of this patch is: Waiting on Author\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 01 Sep 2017 12:28:02 +0000","msg_from":"Aleksander Alekseev <a.alekseev@postgrespro.ru>","msg_from_op":false,"msg_subject":"Re: [PATCH] Off-by-one error in logical slot resource retention"},{"msg_contents":"> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\n> \n> The following review has been posted through the commitfest application:\n> make installcheck-world: not tested\n> Implements feature: not tested\n> Spec compliant: not tested\n> Documentation: not tested\n> \n> Hi Craig,\n> \n> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\n> \n> The new status of this patch is: Waiting on Author\n\nHave you had a chance to look at rebasing this patch so we can get it further\nin the process?\n\ncheers ./daniel\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 15 Sep 2017 13:26:39 +0200","msg_from":"Daniel Gustafsson <daniel@yesql.se>","msg_from_op":false,"msg_subject":"Re: [PATCH] Off-by-one error in logical slot resource retention"},{"msg_contents":"> On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\n> \n>> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\n>> \n>> The following review has been posted through the commitfest application:\n>> make installcheck-world: not tested\n>> Implements feature: not tested\n>> Spec compliant: not tested\n>> Documentation: not tested\n>> \n>> Hi Craig,\n>> \n>> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\n>> \n>> The new status of this patch is: Waiting on Author\n> \n> Have you had a chance to look at rebasing this patch so we can get it further\n> in the process?\n\nSince this patch has been in Waiting for Author state for the duration of the\ncommitfest without moving, I’m marking it Returned with Feedback. If there is\nstill interest in pursuing this patch, please re-submit it to the next\ncommitfest with the comments addressed.\n\ncheers ./daniel\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Sun, 1 Oct 2017 23:27:30 +0200","msg_from":"Daniel Gustafsson <daniel@yesql.se>","msg_from_op":false,"msg_subject":"Re: [PATCH] Off-by-one error in logical slot resource retention"},{"msg_contents":"On 2 October 2017 at 05:27, Daniel Gustafsson <daniel@yesql.se> wrote:\n\n> > On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\n> >\n> >> On 01 Sep 2017, at 14:28, Aleksander Alekseev <\n> a.alekseev@postgrespro.ru> wrote:\n> >>\n> >> The following review has been posted through the commitfest application:\n> >> make installcheck-world: not tested\n> >> Implements feature: not tested\n> >> Spec compliant: not tested\n> >> Documentation: not tested\n> >>\n> >> Hi Craig,\n> >>\n> >> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please\n> resolve the conflicts?\n> >>\n> >> The new status of this patch is: Waiting on Author\n> >\n> > Have you had a chance to look at rebasing this patch so we can get it\n> further\n> > in the process?\n>\n> Since this patch has been in Waiting for Author state for the duration of\n> the\n> commitfest without moving, I’m marking it Returned with Feedback. If\n> there is\n> still interest in pursuing this patch, please re-submit it to the next\n> commitfest with the comments addressed.\n>\n\nThanks. I'll revisit it next CF.\n\n-- \n Craig Ringer http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\nOn 2 October 2017 at 05:27, Daniel Gustafsson <daniel@yesql.se> wrote:> On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\n>\n>> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\n>>\n>> The following review has been posted through the commitfest application:\n>> make installcheck-world: not tested\n>> Implements feature: not tested\n>> Spec compliant: not tested\n>> Documentation: not tested\n>>\n>> Hi Craig,\n>>\n>> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\n>>\n>> The new status of this patch is: Waiting on Author\n>\n> Have you had a chance to look at rebasing this patch so we can get it further\n> in the process?\n\nSince this patch has been in Waiting for Author state for the duration of the\ncommitfest without moving, I’m marking it Returned with Feedback. If there is\nstill interest in pursuing this patch, please re-submit it to the next\ncommitfest with the comments addressed.Thanks. I'll revisit it next CF. -- Craig Ringer http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services","msg_date":"Mon, 2 Oct 2017 09:43:20 +0800","msg_from":"Craig Ringer <craig@2ndquadrant.com>","msg_from_op":true,"msg_subject":"Re: [PATCH] Off-by-one error in logical slot resource retention"}],"string":"[\n {\n \"msg_contents\": \"Hi all\\n\\nI've found a minor off-by-one error in the resource retention logic\\nfor logical slots, where we treat confirmed_flush as meaning \\\"flushed\\nup to and including this LSN\\\". Seems reasonable, but the rest of the\\ncode treats it as \\\"flushed up to but excluding this LSN\\\". In\\nparticular, we treat confirmed_flush as the inclusive startpoint for a\\nnew logical decoding session if no startpoint is provided by the\\nclient and will replay a change whose commit record begins at exactly\\nconfirmed_flush to the client.\\n\\nThis issue was identified while debugging an issue where duplicate\\nrows were replicated after unclean shutdown by logical replication on\\na table with no PK.\\n\\nI'd prefer to make confirmed_flush mean \\\"confirmed flushed up to and\\nincluding\\\" everywhere, but the knock-on effects are too ugly. In\\nparticular we'd then be changing the meaning of START_REPLICATION ...\\nLOGICAL ... 's argument LSN to be \\\"start replay of commits after, but\\nnot including, this LSN\\\". We can't just adjust it by 1, otherwise\\nsuddenly we'd get different results if we passed the confirmed_flush\\nvalue explicitly vs passing 0/0 and letting it be picked up\\nimplicitly.\\n\\nTwo further minor patches follow, one to fix a harmless but confusing\\nquirk in logical walsender init, and one that adds explanatory\\ncomments to relevant parts of the code (and a couple of spots in the\\ndocs) to avoid future occurrences of the above issues.\\n\\n-- \\n Craig Ringer http://www.2ndQuadrant.com/\\n PostgreSQL Development, 24x7 Support, Training & Services\\n\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 9 Mar 2017 11:17:05 +0800\",\n \"msg_from\": \"Craig Ringer <craig@2ndquadrant.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"[PATCH] Off-by-one error in logical slot resource retention\"\n },\n {\n \"msg_contents\": \"On 9 March 2017 at 11:17, Craig Ringer <craig@2ndquadrant.com> wrote:\\n> Hi all\\n>\\n> I've found a minor off-by-one error in the resource retention logic\\n> for logical slots, where we treat confirmed_flush as meaning \\\"flushed\\n> up to and including this LSN\\\". Seems reasonable, but the rest of the\\n> code treats it as \\\"flushed up to but excluding this LSN\\\". In\\n> particular, we treat confirmed_flush as the inclusive startpoint for a\\n> new logical decoding session if no startpoint is provided by the\\n> client and will replay a change whose commit record begins at exactly\\n> confirmed_flush to the client.\\n\\nThis should be backpatched to 9.4.\\n\\nThe others do not need backpatching.\\n\\n-- \\n Craig Ringer http://www.2ndQuadrant.com/\\n PostgreSQL Development, 24x7 Support, Training & Services\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 11:18:30 +0800\",\n \"msg_from\": \"Craig Ringer <craig@2ndquadrant.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [PATCH] Off-by-one error in logical slot resource retention\"\n },\n {\n \"msg_contents\": \"The following review has been posted through the commitfest application:\\nmake installcheck-world: not tested\\nImplements feature: not tested\\nSpec compliant: not tested\\nDocumentation: not tested\\n\\nHi Craig,\\r\\n\\r\\nI'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\\n\\nThe new status of this patch is: Waiting on Author\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 01 Sep 2017 12:28:02 +0000\",\n \"msg_from\": \"Aleksander Alekseev <a.alekseev@postgrespro.ru>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Off-by-one error in logical slot resource retention\"\n },\n {\n \"msg_contents\": \"> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\\n> \\n> The following review has been posted through the commitfest application:\\n> make installcheck-world: not tested\\n> Implements feature: not tested\\n> Spec compliant: not tested\\n> Documentation: not tested\\n> \\n> Hi Craig,\\n> \\n> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\\n> \\n> The new status of this patch is: Waiting on Author\\n\\nHave you had a chance to look at rebasing this patch so we can get it further\\nin the process?\\n\\ncheers ./daniel\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 15 Sep 2017 13:26:39 +0200\",\n \"msg_from\": \"Daniel Gustafsson <daniel@yesql.se>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Off-by-one error in logical slot resource retention\"\n },\n {\n \"msg_contents\": \"> On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\\n> \\n>> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\\n>> \\n>> The following review has been posted through the commitfest application:\\n>> make installcheck-world: not tested\\n>> Implements feature: not tested\\n>> Spec compliant: not tested\\n>> Documentation: not tested\\n>> \\n>> Hi Craig,\\n>> \\n>> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\\n>> \\n>> The new status of this patch is: Waiting on Author\\n> \\n> Have you had a chance to look at rebasing this patch so we can get it further\\n> in the process?\\n\\nSince this patch has been in Waiting for Author state for the duration of the\\ncommitfest without moving, I’m marking it Returned with Feedback. If there is\\nstill interest in pursuing this patch, please re-submit it to the next\\ncommitfest with the comments addressed.\\n\\ncheers ./daniel\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Sun, 1 Oct 2017 23:27:30 +0200\",\n \"msg_from\": \"Daniel Gustafsson <daniel@yesql.se>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Off-by-one error in logical slot resource retention\"\n },\n {\n \"msg_contents\": \"On 2 October 2017 at 05:27, Daniel Gustafsson <daniel@yesql.se> wrote:\\n\\n> > On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\\n> >\\n> >> On 01 Sep 2017, at 14:28, Aleksander Alekseev <\\n> a.alekseev@postgrespro.ru> wrote:\\n> >>\\n> >> The following review has been posted through the commitfest application:\\n> >> make installcheck-world: not tested\\n> >> Implements feature: not tested\\n> >> Spec compliant: not tested\\n> >> Documentation: not tested\\n> >>\\n> >> Hi Craig,\\n> >>\\n> >> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please\\n> resolve the conflicts?\\n> >>\\n> >> The new status of this patch is: Waiting on Author\\n> >\\n> > Have you had a chance to look at rebasing this patch so we can get it\\n> further\\n> > in the process?\\n>\\n> Since this patch has been in Waiting for Author state for the duration of\\n> the\\n> commitfest without moving, I’m marking it Returned with Feedback. If\\n> there is\\n> still interest in pursuing this patch, please re-submit it to the next\\n> commitfest with the comments addressed.\\n>\\n\\nThanks. I'll revisit it next CF.\\n\\n-- \\n Craig Ringer http://www.2ndQuadrant.com/\\n PostgreSQL Development, 24x7 Support, Training & Services\\n\\nOn 2 October 2017 at 05:27, Daniel Gustafsson <daniel@yesql.se> wrote:> On 15 Sep 2017, at 13:26, Daniel Gustafsson <daniel@yesql.se> wrote:\\n>\\n>> On 01 Sep 2017, at 14:28, Aleksander Alekseev <a.alekseev@postgrespro.ru> wrote:\\n>>\\n>> The following review has been posted through the commitfest application:\\n>> make installcheck-world: not tested\\n>> Implements feature: not tested\\n>> Spec compliant: not tested\\n>> Documentation: not tested\\n>>\\n>> Hi Craig,\\n>>\\n>> I'm afraid patches 0002 and 0003 don't apply anymore. Could you please resolve the conflicts?\\n>>\\n>> The new status of this patch is: Waiting on Author\\n>\\n> Have you had a chance to look at rebasing this patch so we can get it further\\n> in the process?\\n\\nSince this patch has been in Waiting for Author state for the duration of the\\ncommitfest without moving, I’m marking it Returned with Feedback. If there is\\nstill interest in pursuing this patch, please re-submit it to the next\\ncommitfest with the comments addressed.Thanks. I'll revisit it next CF. -- Craig Ringer http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services\",\n \"msg_date\": \"Mon, 2 Oct 2017 09:43:20 +0800\",\n \"msg_from\": \"Craig Ringer <craig@2ndquadrant.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [PATCH] Off-by-one error in logical slot resource retention\"\n }\n]","isConversational":false}}},{"rowIdx":44092,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi!\n\nI'd like to offer a forward port from a change I'm contributing\nthe Greenplum code base over here:\n https://github.com/greenplum-db/gpdb/pull/1983\n\nWhile it is true that 64 bit atomics are mostly a no-op\nfor Postgres right now, perhaps this patch could still\nbe useful.\n\nTo make it easier to review, let me repeat the analysis I did\non the pull request in this email:\n\nI built the following tiny example with -fno-inline-small-functions to\nmake sure you get to see all the guts (without the\n-fno-inline-small-functions all of this gets inlined and somewhat\nfurther optimized):\n\nextern pg_atomic_uint64 test1;\nextern uint64 test2;\nextern uint64 test3;\n\nint main() {\n pg_atomic_compare_exchange_u64(&test1, &test2, test3);\n pg_atomic_add_fetch_u64(&test1, 1);\n pg_atomic_add_fetch_u64(&test1, 2);\n}\n\nHere's the assembly I got from it:\n\nDisassembly of section .text:\n\n0000000000000000 <pg_atomic_compare_exchange_u64_impl>:\n 0: d10083ff sub sp, sp, #0x20\n 4: f9000fe0 str x0, [sp,#24]\n 8: f9000be1 str x1, [sp,#16]\n c: f90007e2 str x2, [sp,#8]\n 10: f9400fe1 ldr x1, [sp,#24]\n 14: f9400be0 ldr x0, [sp,#16]\n 18: f9400003 ldr x3, [x0]\n 1c: f94007e4 ldr x4, [sp,#8]\n 20: c85ffc22 ldaxr x2, [x1]\n 24: eb03005f cmp x2, x3\n 28: 54000061 b.ne 34 <pg_atomic_compare_exchange_u64_impl+0x34>\n 2c: c805fc24 stlxr w5, x4, [x1]\n 30: 35ffff85 cbnz w5, 20 <pg_atomic_compare_exchange_u64_impl+0x20>\n 34: 1a9f17e1 cset w1, eq\n 38: 7100003f cmp w1, #0x0\n 3c: 54000041 b.ne 44 <pg_atomic_compare_exchange_u64_impl+0x44>\n 40: f9000002 str x2, [x0]\n 44: 2a0103e0 mov w0, w1\n 48: 910083ff add sp, sp, #0x20\n 4c: d65f03c0 ret\n\n0000000000000050 <pg_atomic_fetch_add_u64_impl>:\n 50: d10043ff sub sp, sp, #0x10\n 54: f90007e0 str x0, [sp,#8]\n 58: f90003e1 str x1, [sp]\n 5c: f94007e0 ldr x0, [sp,#8]\n 60: f94003e1 ldr x1, [sp]\n 64: c85f7c02 ldxr x2, [x0]\n 68: 8b010043 add x3, x2, x1\n 6c: c804fc03 stlxr w4, x3, [x0]\n 70: 35ffffa4 cbnz w4, 64 <pg_atomic_fetch_add_u64_impl+0x14>\n 74: d5033bbf dmb ish\n 78: aa0203e0 mov x0, x2\n 7c: 910043ff add sp, sp, #0x10\n 80: d65f03c0 ret\n\n0000000000000084 <pg_atomic_add_fetch_u64_impl>:\n 84: a9be7bfd stp x29, x30, [sp,#-32]!\n 88: 910003fd mov x29, sp\n 8c: f9000fa0 str x0, [x29,#24]\n 90: f9000ba1 str x1, [x29,#16]\n 94: f9400ba1 ldr x1, [x29,#16]\n 98: f9400fa0 ldr x0, [x29,#24]\n 9c: 97ffffed bl 50 <pg_atomic_fetch_add_u64_impl>\n a0: aa0003e1 mov x1, x0\n a4: f9400ba0 ldr x0, [x29,#16]\n a8: 8b000020 add x0, x1, x0\n ac: a8c27bfd ldp x29, x30, [sp],#32\n b0: d65f03c0 ret\n\n00000000000000b4 <pg_atomic_compare_exchange_u64>:\n b4: a9bd7bfd stp x29, x30, [sp,#-48]!\n b8: 910003fd mov x29, sp\n bc: f90017a0 str x0, [x29,#40]\n c0: f90013a1 str x1, [x29,#32]\n c4: f9000fa2 str x2, [x29,#24]\n c8: f9400fa2 ldr x2, [x29,#24]\n cc: f94013a1 ldr x1, [x29,#32]\n d0: f94017a0 ldr x0, [x29,#40]\n d4: 97ffffcb bl 0 <pg_atomic_compare_exchange_u64_impl>\n d8: 53001c00 uxtb w0, w0\n dc: a8c37bfd ldp x29, x30, [sp],#48\n e0: d65f03c0 ret\n\n00000000000000e4 <pg_atomic_add_fetch_u64>:\n e4: a9be7bfd stp x29, x30, [sp,#-32]!\n e8: 910003fd mov x29, sp\n ec: f9000fa0 str x0, [x29,#24]\n f0: f9000ba1 str x1, [x29,#16]\n f4: f9400ba1 ldr x1, [x29,#16]\n f8: f9400fa0 ldr x0, [x29,#24]\n fc: 97ffffe2 bl 84 <pg_atomic_add_fetch_u64_impl>\n 100: a8c27bfd ldp x29, x30, [sp],#32\n 104: d65f03c0 ret\n\n0000000000000108 <main>:\n 108: a9bf7bfd stp x29, x30, [sp,#-16]!\n 10c: 910003fd mov x29, sp\n 110: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\n 114: f9400000 ldr x0, [x0]\n 118: f9400002 ldr x2, [x0]\n 11c: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\n 120: f9400001 ldr x1, [x0]\n 124: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\n 128: f9400000 ldr x0, [x0]\n 12c: 97ffffe2 bl b4 <pg_atomic_compare_exchange_u64>\n 130: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\n 134: f9400000 ldr x0, [x0]\n 138: d2800021 mov x1, #0x1 // #1\n 13c: 97ffffea bl e4 <pg_atomic_add_fetch_u64>\n 140: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\n 144: f9400000 ldr x0, [x0]\n 148: d2800041 mov x1, #0x2 // #2\n 14c: 97ffffe6 bl e4 <pg_atomic_add_fetch_u64>\n 150: 52800000 mov w0, #0x0 // #0\n 154: a8c17bfd ldp x29, x30, [sp],#16\n 158: d65f03c0 ret\n\n\nThanks,\nRoman.\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 8 Mar 2017 19:18:04 -0800","msg_from":"Roman Shaposhnik <rvs@apache.org>","msg_from_op":true,"msg_subject":"[PATCH] Enabling atomics on ARM64"},{"msg_contents":"On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\n> I'd like to offer a forward port from a change I'm contributing\n> the Greenplum code base over here:\n> https://github.com/greenplum-db/gpdb/pull/1983\n\nThanks. Please add this to https://commitfest.postgresql.org/14/ so\nit doesn't get forgotten. I'm afraid your patch is probably too late\nfor v10 (the deadline for patches to be considered for v10 was March\n1st, though somebody might propose to make an exception), but v11 will\nbe here soon enough.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 13:01:07 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"On Fri, Mar 10, 2017 at 3:01 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n> On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\n>> I'd like to offer a forward port from a change I'm contributing\n>> the Greenplum code base over here:\n>> https://github.com/greenplum-db/gpdb/pull/1983\n\n+1.\n\n> Thanks. Please add this to https://commitfest.postgresql.org/14/ so\n> it doesn't get forgotten. I'm afraid your patch is probably too late\n> for v10 (the deadline for patches to be considered for v10 was March\n> 1st, though somebody might propose to make an exception), but v11 will\n> be here soon enough.\n\nYeah, one month earlier would have been fine though. This is not\nreally invasive.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 10:29:39 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"On 2017-03-10 10:29:39 +0900, Michael Paquier wrote:\n> On Fri, Mar 10, 2017 at 3:01 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n> > On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\n> >> I'd like to offer a forward port from a change I'm contributing\n> >> the Greenplum code base over here:\n> >> https://github.com/greenplum-db/gpdb/pull/1983\n> \n> +1.\n> \n> > Thanks. Please add this to https://commitfest.postgresql.org/14/ so\n> > it doesn't get forgotten. I'm afraid your patch is probably too late\n> > for v10 (the deadline for patches to be considered for v10 was March\n> > 1st, though somebody might propose to make an exception), but v11 will\n> > be here soon enough.\n> \n> Yeah, one month earlier would have been fine though. This is not\n> really invasive.\n\nFWIW, Unless somebody seriously protests, I'm going to commit this soon-ish.\n\n- Andres\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 17:54:44 -0800","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> FWIW, Unless somebody seriously protests, I'm going to commit this soon-ish.\n\nSure, if you want to take responsibility for it, push away.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 00:03:46 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"Hi,\n\nOn 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\n> I'd like to offer a forward port from a change I'm contributing\n> the Greenplum code base over here\n\nI think the change makes sense. Any chance we could convince you to\nbring up an arm64 buildfarm animal (our test infrastructure) up, so we\ncan be sure arm64 stays working? See https://buildfarm.postgresql.org/\n\nPushed, thanks for your contribution.\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 11:20:43 -0800","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"On Fri, Mar 10, 2017 at 11:20 AM, Andres Freund <andres@anarazel.de> wrote:\n> Hi,\n>\n> On 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\n>> I'd like to offer a forward port from a change I'm contributing\n>> the Greenplum code base over here\n>\n> I think the change makes sense. Any chance we could convince you to\n> bring up an arm64 buildfarm animal (our test infrastructure) up, so we\n> can be sure arm64 stays working? See https://buildfarm.postgresql.org/\n\nWhat does it take to be a your build farm slave? Can it be done via\na Docker container that calls out to your infrastructure? I can probably\nrun that as part of the Apache Bigtop CI.\n\n> Pushed, thanks for your contribution.\n\nThanks for taking care of it so quickly!\n\nThanks,\nRoman.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 18:14:13 -0800","msg_from":"Roman Shaposhnik <rvs@apache.org>","msg_from_op":true,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"},{"msg_contents":"On 2017-03-10 18:14:13 -0800, Roman Shaposhnik wrote:\n> On Fri, Mar 10, 2017 at 11:20 AM, Andres Freund <andres@anarazel.de> wrote:\n> > Hi,\n> >\n> > On 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\n> >> I'd like to offer a forward port from a change I'm contributing\n> >> the Greenplum code base over here\n> >\n> > I think the change makes sense. Any chance we could convince you to\n> > bring up an arm64 buildfarm animal (our test infrastructure) up, so we\n> > can be sure arm64 stays working? See https://buildfarm.postgresql.org/\n> \n> What does it take to be a your build farm slave? Can it be done via\n> a Docker container that calls out to your infrastructure? I can probably\n> run that as part of the Apache Bigtop CI.\n\nYes, it can. The buildfarm stuff is all push from the test machine's\nside, not pull. You basically configure a cron job as described in\nhttps://wiki.postgresql.org/wiki/PostgreSQL_Buildfarm_Howto and then\nregister your animal with the form on\nhttps://buildfarm.postgresql.org/cgi-bin/register-form.pl\n\nWould be great. We have a bunch of 32bit arm animal, but no 64bit ones:\nhttps://buildfarm.postgresql.org/cgi-bin/show_members.pl\nhttps://buildfarm.postgresql.org/cgi-bin/show_status.pl\n\nRegards,\n\nAndres\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 18:17:17 -0800","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: [PATCH] Enabling atomics on ARM64"}],"string":"[\n {\n \"msg_contents\": \"Hi!\\n\\nI'd like to offer a forward port from a change I'm contributing\\nthe Greenplum code base over here:\\n https://github.com/greenplum-db/gpdb/pull/1983\\n\\nWhile it is true that 64 bit atomics are mostly a no-op\\nfor Postgres right now, perhaps this patch could still\\nbe useful.\\n\\nTo make it easier to review, let me repeat the analysis I did\\non the pull request in this email:\\n\\nI built the following tiny example with -fno-inline-small-functions to\\nmake sure you get to see all the guts (without the\\n-fno-inline-small-functions all of this gets inlined and somewhat\\nfurther optimized):\\n\\nextern pg_atomic_uint64 test1;\\nextern uint64 test2;\\nextern uint64 test3;\\n\\nint main() {\\n pg_atomic_compare_exchange_u64(&test1, &test2, test3);\\n pg_atomic_add_fetch_u64(&test1, 1);\\n pg_atomic_add_fetch_u64(&test1, 2);\\n}\\n\\nHere's the assembly I got from it:\\n\\nDisassembly of section .text:\\n\\n0000000000000000 <pg_atomic_compare_exchange_u64_impl>:\\n 0: d10083ff sub sp, sp, #0x20\\n 4: f9000fe0 str x0, [sp,#24]\\n 8: f9000be1 str x1, [sp,#16]\\n c: f90007e2 str x2, [sp,#8]\\n 10: f9400fe1 ldr x1, [sp,#24]\\n 14: f9400be0 ldr x0, [sp,#16]\\n 18: f9400003 ldr x3, [x0]\\n 1c: f94007e4 ldr x4, [sp,#8]\\n 20: c85ffc22 ldaxr x2, [x1]\\n 24: eb03005f cmp x2, x3\\n 28: 54000061 b.ne 34 <pg_atomic_compare_exchange_u64_impl+0x34>\\n 2c: c805fc24 stlxr w5, x4, [x1]\\n 30: 35ffff85 cbnz w5, 20 <pg_atomic_compare_exchange_u64_impl+0x20>\\n 34: 1a9f17e1 cset w1, eq\\n 38: 7100003f cmp w1, #0x0\\n 3c: 54000041 b.ne 44 <pg_atomic_compare_exchange_u64_impl+0x44>\\n 40: f9000002 str x2, [x0]\\n 44: 2a0103e0 mov w0, w1\\n 48: 910083ff add sp, sp, #0x20\\n 4c: d65f03c0 ret\\n\\n0000000000000050 <pg_atomic_fetch_add_u64_impl>:\\n 50: d10043ff sub sp, sp, #0x10\\n 54: f90007e0 str x0, [sp,#8]\\n 58: f90003e1 str x1, [sp]\\n 5c: f94007e0 ldr x0, [sp,#8]\\n 60: f94003e1 ldr x1, [sp]\\n 64: c85f7c02 ldxr x2, [x0]\\n 68: 8b010043 add x3, x2, x1\\n 6c: c804fc03 stlxr w4, x3, [x0]\\n 70: 35ffffa4 cbnz w4, 64 <pg_atomic_fetch_add_u64_impl+0x14>\\n 74: d5033bbf dmb ish\\n 78: aa0203e0 mov x0, x2\\n 7c: 910043ff add sp, sp, #0x10\\n 80: d65f03c0 ret\\n\\n0000000000000084 <pg_atomic_add_fetch_u64_impl>:\\n 84: a9be7bfd stp x29, x30, [sp,#-32]!\\n 88: 910003fd mov x29, sp\\n 8c: f9000fa0 str x0, [x29,#24]\\n 90: f9000ba1 str x1, [x29,#16]\\n 94: f9400ba1 ldr x1, [x29,#16]\\n 98: f9400fa0 ldr x0, [x29,#24]\\n 9c: 97ffffed bl 50 <pg_atomic_fetch_add_u64_impl>\\n a0: aa0003e1 mov x1, x0\\n a4: f9400ba0 ldr x0, [x29,#16]\\n a8: 8b000020 add x0, x1, x0\\n ac: a8c27bfd ldp x29, x30, [sp],#32\\n b0: d65f03c0 ret\\n\\n00000000000000b4 <pg_atomic_compare_exchange_u64>:\\n b4: a9bd7bfd stp x29, x30, [sp,#-48]!\\n b8: 910003fd mov x29, sp\\n bc: f90017a0 str x0, [x29,#40]\\n c0: f90013a1 str x1, [x29,#32]\\n c4: f9000fa2 str x2, [x29,#24]\\n c8: f9400fa2 ldr x2, [x29,#24]\\n cc: f94013a1 ldr x1, [x29,#32]\\n d0: f94017a0 ldr x0, [x29,#40]\\n d4: 97ffffcb bl 0 <pg_atomic_compare_exchange_u64_impl>\\n d8: 53001c00 uxtb w0, w0\\n dc: a8c37bfd ldp x29, x30, [sp],#48\\n e0: d65f03c0 ret\\n\\n00000000000000e4 <pg_atomic_add_fetch_u64>:\\n e4: a9be7bfd stp x29, x30, [sp,#-32]!\\n e8: 910003fd mov x29, sp\\n ec: f9000fa0 str x0, [x29,#24]\\n f0: f9000ba1 str x1, [x29,#16]\\n f4: f9400ba1 ldr x1, [x29,#16]\\n f8: f9400fa0 ldr x0, [x29,#24]\\n fc: 97ffffe2 bl 84 <pg_atomic_add_fetch_u64_impl>\\n 100: a8c27bfd ldp x29, x30, [sp],#32\\n 104: d65f03c0 ret\\n\\n0000000000000108 <main>:\\n 108: a9bf7bfd stp x29, x30, [sp,#-16]!\\n 10c: 910003fd mov x29, sp\\n 110: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\\n 114: f9400000 ldr x0, [x0]\\n 118: f9400002 ldr x2, [x0]\\n 11c: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\\n 120: f9400001 ldr x1, [x0]\\n 124: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\\n 128: f9400000 ldr x0, [x0]\\n 12c: 97ffffe2 bl b4 <pg_atomic_compare_exchange_u64>\\n 130: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\\n 134: f9400000 ldr x0, [x0]\\n 138: d2800021 mov x1, #0x1 // #1\\n 13c: 97ffffea bl e4 <pg_atomic_add_fetch_u64>\\n 140: 90000000 adrp x0, 0 <_GLOBAL_OFFSET_TABLE_>\\n 144: f9400000 ldr x0, [x0]\\n 148: d2800041 mov x1, #0x2 // #2\\n 14c: 97ffffe6 bl e4 <pg_atomic_add_fetch_u64>\\n 150: 52800000 mov w0, #0x0 // #0\\n 154: a8c17bfd ldp x29, x30, [sp],#16\\n 158: d65f03c0 ret\\n\\n\\nThanks,\\nRoman.\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 8 Mar 2017 19:18:04 -0800\",\n \"msg_from\": \"Roman Shaposhnik <rvs@apache.org>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"[PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\\n> I'd like to offer a forward port from a change I'm contributing\\n> the Greenplum code base over here:\\n> https://github.com/greenplum-db/gpdb/pull/1983\\n\\nThanks. Please add this to https://commitfest.postgresql.org/14/ so\\nit doesn't get forgotten. I'm afraid your patch is probably too late\\nfor v10 (the deadline for patches to be considered for v10 was March\\n1st, though somebody might propose to make an exception), but v11 will\\nbe here soon enough.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 13:01:07 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 10, 2017 at 3:01 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\\n>> I'd like to offer a forward port from a change I'm contributing\\n>> the Greenplum code base over here:\\n>> https://github.com/greenplum-db/gpdb/pull/1983\\n\\n+1.\\n\\n> Thanks. Please add this to https://commitfest.postgresql.org/14/ so\\n> it doesn't get forgotten. I'm afraid your patch is probably too late\\n> for v10 (the deadline for patches to be considered for v10 was March\\n> 1st, though somebody might propose to make an exception), but v11 will\\n> be here soon enough.\\n\\nYeah, one month earlier would have been fine though. This is not\\nreally invasive.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 10:29:39 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"On 2017-03-10 10:29:39 +0900, Michael Paquier wrote:\\n> On Fri, Mar 10, 2017 at 3:01 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> > On Wed, Mar 8, 2017 at 10:18 PM, Roman Shaposhnik <rvs@apache.org> wrote:\\n> >> I'd like to offer a forward port from a change I'm contributing\\n> >> the Greenplum code base over here:\\n> >> https://github.com/greenplum-db/gpdb/pull/1983\\n> \\n> +1.\\n> \\n> > Thanks. Please add this to https://commitfest.postgresql.org/14/ so\\n> > it doesn't get forgotten. I'm afraid your patch is probably too late\\n> > for v10 (the deadline for patches to be considered for v10 was March\\n> > 1st, though somebody might propose to make an exception), but v11 will\\n> > be here soon enough.\\n> \\n> Yeah, one month earlier would have been fine though. This is not\\n> really invasive.\\n\\nFWIW, Unless somebody seriously protests, I'm going to commit this soon-ish.\\n\\n- Andres\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 17:54:44 -0800\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> FWIW, Unless somebody seriously protests, I'm going to commit this soon-ish.\\n\\nSure, if you want to take responsibility for it, push away.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 00:03:46 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"Hi,\\n\\nOn 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\\n> I'd like to offer a forward port from a change I'm contributing\\n> the Greenplum code base over here\\n\\nI think the change makes sense. Any chance we could convince you to\\nbring up an arm64 buildfarm animal (our test infrastructure) up, so we\\ncan be sure arm64 stays working? See https://buildfarm.postgresql.org/\\n\\nPushed, thanks for your contribution.\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 11:20:43 -0800\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 10, 2017 at 11:20 AM, Andres Freund <andres@anarazel.de> wrote:\\n> Hi,\\n>\\n> On 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\\n>> I'd like to offer a forward port from a change I'm contributing\\n>> the Greenplum code base over here\\n>\\n> I think the change makes sense. Any chance we could convince you to\\n> bring up an arm64 buildfarm animal (our test infrastructure) up, so we\\n> can be sure arm64 stays working? See https://buildfarm.postgresql.org/\\n\\nWhat does it take to be a your build farm slave? Can it be done via\\na Docker container that calls out to your infrastructure? I can probably\\nrun that as part of the Apache Bigtop CI.\\n\\n> Pushed, thanks for your contribution.\\n\\nThanks for taking care of it so quickly!\\n\\nThanks,\\nRoman.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 18:14:13 -0800\",\n \"msg_from\": \"Roman Shaposhnik <rvs@apache.org>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n },\n {\n \"msg_contents\": \"On 2017-03-10 18:14:13 -0800, Roman Shaposhnik wrote:\\n> On Fri, Mar 10, 2017 at 11:20 AM, Andres Freund <andres@anarazel.de> wrote:\\n> > Hi,\\n> >\\n> > On 2017-03-08 19:18:04 -0800, Roman Shaposhnik wrote:\\n> >> I'd like to offer a forward port from a change I'm contributing\\n> >> the Greenplum code base over here\\n> >\\n> > I think the change makes sense. Any chance we could convince you to\\n> > bring up an arm64 buildfarm animal (our test infrastructure) up, so we\\n> > can be sure arm64 stays working? See https://buildfarm.postgresql.org/\\n> \\n> What does it take to be a your build farm slave? Can it be done via\\n> a Docker container that calls out to your infrastructure? I can probably\\n> run that as part of the Apache Bigtop CI.\\n\\nYes, it can. The buildfarm stuff is all push from the test machine's\\nside, not pull. You basically configure a cron job as described in\\nhttps://wiki.postgresql.org/wiki/PostgreSQL_Buildfarm_Howto and then\\nregister your animal with the form on\\nhttps://buildfarm.postgresql.org/cgi-bin/register-form.pl\\n\\nWould be great. We have a bunch of 32bit arm animal, but no 64bit ones:\\nhttps://buildfarm.postgresql.org/cgi-bin/show_members.pl\\nhttps://buildfarm.postgresql.org/cgi-bin/show_status.pl\\n\\nRegards,\\n\\nAndres\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 18:17:17 -0800\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCH] Enabling atomics on ARM64\"\n }\n]","isConversational":false}}},{"rowIdx":44093,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi,\n\nWe are having some performance issues after we upgraded to newest\nversion of PostgreSQL, before it everything was fast and smooth.\n\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\nupgraded to 9.6.2 with no improvement.\n\nSome information about our setup: Freebsd, Solaris (SmartOS), simple\nmaster-slave using streaming replication.\n\nProblem:\nVery high system CPU when master is streaming replication data, CPU\ngoes up to 77%. Only one process is generating this load, it's a\npostgresql startup process. When I attached a truss to this process I\nsaw a lot o read calls with almost the same number of errors (EAGAIN).\n\nroot(at)d8:~ # truss -c -p 38091\n^Csyscall seconds calls errors\nsemop 0.001611782 198 0\nwrite 0.000074404 2 0\nread 2.281535100 17266 12375\nopenat 0.000683532 48 0\nlseek 0.177612479 20443 0\nclose 0.000395549 48 0\n ------------- ------- -------\n 2.461912846 38005 12375\n\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)\nlseek(778,0x0,SEEK_END) = 57344 (0xe000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)\nread(894,\" ~\\0\\08\\a\\M--m\\0\\0\\^A\\0\\M^T\\0000\"...,8192) = 8192 (0x2000)\nlseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)\nread(3,\"\\M^S\\M-P\\^E\\0\\^A\\0\\0\\0\\0`\\M-z\"...,8192) = 8192 (0x2000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)\nread(894,\"\\^_~\\0\\0\\M-H\\M-H\\M-B\\M-b\\0\\0\\^E\"...,8192) = 8192 (0x2000)\nlseek(818,0x0,SEEK_END) = 57344 (0xe000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)\nlseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)\nread(442,\"\\^_~\\0\\0\\M-P\\M-+\\M-1\\M-b\\0\\0\\0\\0\"...,8192) = 8192 (0x2000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n\nDescriptor 6 is a pipe\n\nRead call try to read one byte over and over, I looked up to source\ncode and I think this file is responsible for this behavior\nsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nHi,We are having some performance issues after we upgraded to newestversion of PostgreSQL, before it everything was fast and smooth.Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now weupgraded to 9.6.2 with no improvement.Some information about our setup: Freebsd, Solaris (SmartOS), simplemaster-slave using streaming replication.Problem:Very high system CPU when master is streaming replication data, CPUgoes up to 77%. Only one process is generating this load, it's apostgresql startup process. When I attached a truss to this process Isaw a lot o read calls with almost the same number of errors (EAGAIN).root(at)d8:~ # truss -c -p 38091^Csyscall seconds calls errorssemop 0.001611782 198 0write 0.000074404 2 0read 2.281535100 17266 12375openat 0.000683532 48 0lseek 0.177612479 20443 0close 0.000395549 48 0 ------------- ------- ------- 2.461912846 38005 12375read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)lseek(778,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)read(894,\" ~\\0\\08\\a\\M--m\\0\\0\\^A\\0\\M^T\\0000\"...,8192) = 8192 (0x2000)lseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)read(3,\"\\M^S\\M-P\\^E\\0\\^A\\0\\0\\0\\0`\\M-z\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)read(894,\"\\^_~\\0\\0\\M-H\\M-H\\M-B\\M-b\\0\\0\\^E\"...,8192) = 8192 (0x2000)lseek(818,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)lseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)read(442,\"\\^_~\\0\\0\\M-P\\M-+\\M-1\\M-b\\0\\0\\0\\0\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'Descriptor 6 is a pipeRead call try to read one byte over and over, I looked up to sourcecode and I think this file is responsible for this behaviorsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.","msg_date":"Thu, 9 Mar 2017 13:39:40 +0100","msg_from":"Naytro Naytro <naytro@googlemail.com>","msg_from_op":true,"msg_subject":"Performance issue after upgrading from 9.4 to 9.6"}],"string":"[\n {\n \"msg_contents\": \"Hi,\\n\\nWe are having some performance issues after we upgraded to newest\\nversion of PostgreSQL, before it everything was fast and smooth.\\n\\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\nupgraded to 9.6.2 with no improvement.\\n\\nSome information about our setup: Freebsd, Solaris (SmartOS), simple\\nmaster-slave using streaming replication.\\n\\nProblem:\\nVery high system CPU when master is streaming replication data, CPU\\ngoes up to 77%. Only one process is generating this load, it's a\\npostgresql startup process. When I attached a truss to this process I\\nsaw a lot o read calls with almost the same number of errors (EAGAIN).\\n\\nroot(at)d8:~ # truss -c -p 38091\\n^Csyscall seconds calls errors\\nsemop 0.001611782 198 0\\nwrite 0.000074404 2 0\\nread 2.281535100 17266 12375\\nopenat 0.000683532 48 0\\nlseek 0.177612479 20443 0\\nclose 0.000395549 48 0\\n ------------- ------- -------\\n 2.461912846 38005 12375\\n\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)\\nlseek(778,0x0,SEEK_END) = 57344 (0xe000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)\\nread(894,\\\" ~\\\\0\\\\08\\\\a\\\\M--m\\\\0\\\\0\\\\^A\\\\0\\\\M^T\\\\0000\\\"...,8192) = 8192 (0x2000)\\nlseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)\\nread(3,\\\"\\\\M^S\\\\M-P\\\\^E\\\\0\\\\^A\\\\0\\\\0\\\\0\\\\0`\\\\M-z\\\"...,8192) = 8192 (0x2000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)\\nread(894,\\\"\\\\^_~\\\\0\\\\0\\\\M-H\\\\M-H\\\\M-B\\\\M-b\\\\0\\\\0\\\\^E\\\"...,8192) = 8192 (0x2000)\\nlseek(818,0x0,SEEK_END) = 57344 (0xe000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)\\nlseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)\\nread(442,\\\"\\\\^_~\\\\0\\\\0\\\\M-P\\\\M-+\\\\M-1\\\\M-b\\\\0\\\\0\\\\0\\\\0\\\"...,8192) = 8192 (0x2000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n\\nDescriptor 6 is a pipe\\n\\nRead call try to read one byte over and over, I looked up to source\\ncode and I think this file is responsible for this behavior\\nsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nHi,We are having some performance issues after we upgraded to newestversion of PostgreSQL, before it everything was fast and smooth.Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now weupgraded to 9.6.2 with no improvement.Some information about our setup: Freebsd, Solaris (SmartOS), simplemaster-slave using streaming replication.Problem:Very high system CPU when master is streaming replication data, CPUgoes up to 77%. Only one process is generating this load, it's apostgresql startup process. When I attached a truss to this process Isaw a lot o read calls with almost the same number of errors (EAGAIN).root(at)d8:~ # truss -c -p 38091^Csyscall seconds calls errorssemop 0.001611782 198 0write 0.000074404 2 0read 2.281535100 17266 12375openat 0.000683532 48 0lseek 0.177612479 20443 0close 0.000395549 48 0 ------------- ------- ------- 2.461912846 38005 12375read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)lseek(778,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)read(894,\\\" ~\\\\0\\\\08\\\\a\\\\M--m\\\\0\\\\0\\\\^A\\\\0\\\\M^T\\\\0000\\\"...,8192) = 8192 (0x2000)lseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)read(3,\\\"\\\\M^S\\\\M-P\\\\^E\\\\0\\\\^A\\\\0\\\\0\\\\0\\\\0`\\\\M-z\\\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)read(894,\\\"\\\\^_~\\\\0\\\\0\\\\M-H\\\\M-H\\\\M-B\\\\M-b\\\\0\\\\0\\\\^E\\\"...,8192) = 8192 (0x2000)lseek(818,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)lseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)read(442,\\\"\\\\^_~\\\\0\\\\0\\\\M-P\\\\M-+\\\\M-1\\\\M-b\\\\0\\\\0\\\\0\\\\0\\\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'Descriptor 6 is a pipeRead call try to read one byte over and over, I looked up to sourcecode and I think this file is responsible for this behaviorsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\",\n \"msg_date\": \"Thu, 9 Mar 2017 13:39:40 +0100\",\n \"msg_from\": \"Naytro Naytro <naytro@googlemail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Performance issue after upgrading from 9.4 to 9.6\"\n }\n]","isConversational":false}}},{"rowIdx":44094,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi,\n\nWe are having some performance issues after we upgraded to newest\nversion of PostgreSQL, before it everything was fast and smooth.\n\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\nupgraded to 9.6.2 with no improvement.\n\nSome information about our setup: Freebsd, Solaris (SmartOS), simple\nmaster-slave using streaming replication.\n\nProblem:\nVery high system CPU when master is streaming replication data, CPU\ngoes up to 77%. Only one process is generating this load, it's a\npostgresql startup process. When I attached a truss to this process I\nsaw a lot o read calls with almost the same number of errors (EAGAIN).\n\nroot(at)d8:~ # truss -c -p 38091\n^Csyscall seconds calls errors\nsemop 0.001611782 198 0\nwrite 0.000074404 2 0\nread 2.281535100 17266 12375\nopenat 0.000683532 48 0\nlseek 0.177612479 20443 0\nclose 0.000395549 48 0\n ------------- ------- -------\n 2.461912846 38005 12375\n\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)\nlseek(778,0x0,SEEK_END) = 57344 (0xe000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)\nread(894,\" ~\\0\\08\\a\\M--m\\0\\0\\^A\\0\\M^T\\0000\"...,8192) = 8192 (0x2000)\nlseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)\nread(3,\"\\M^S\\M-P\\^E\\0\\^A\\0\\0\\0\\0`\\M-z\"...,8192) = 8192 (0x2000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\nlseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)\nread(894,\"\\^_~\\0\\0\\M-H\\M-H\\M-B\\M-b\\0\\0\\^E\"...,8192) = 8192 (0x2000)\nlseek(818,0x0,SEEK_END) = 57344 (0xe000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\nlseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)\nlseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)\nread(442,\"\\^_~\\0\\0\\M-P\\M-+\\M-1\\M-b\\0\\0\\0\\0\"...,8192) = 8192 (0x2000)\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n\nDescriptor 6 is a pipe\n\nRead call try to read one byte over and over, I looked up to source\ncode and I think this file is responsible for this behavior\nsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nHi,We are having some performance issues after we upgraded to newestversion of PostgreSQL, before it everything was fast and smooth.Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now weupgraded to 9.6.2 with no improvement.Some information about our setup: Freebsd, Solaris (SmartOS), simplemaster-slave using streaming replication.Problem:Very high system CPU when master is streaming replication data, CPUgoes up to 77%. Only one process is generating this load, it's apostgresql startup process. When I attached a truss to this process Isaw a lot o read calls with almost the same number of errors (EAGAIN).root(at)d8:~ # truss -c -p 38091^Csyscall seconds calls errorssemop 0.001611782 198 0write 0.000074404 2 0read 2.281535100 17266 12375openat 0.000683532 48 0lseek 0.177612479 20443 0close 0.000395549 48 0 ------------- ------- ------- 2.461912846 38005 12375read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)lseek(778,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)read(894,\" ~\\0\\08\\a\\M--m\\0\\0\\^A\\0\\M^T\\0000\"...,8192) = 8192 (0x2000)lseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)read(3,\"\\M^S\\M-P\\^E\\0\\^A\\0\\0\\0\\0`\\M-z\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)read(894,\"\\^_~\\0\\0\\M-H\\M-H\\M-B\\M-b\\0\\0\\^E\"...,8192) = 8192 (0x2000)lseek(818,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)lseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)read(442,\"\\^_~\\0\\0\\M-P\\M-+\\M-1\\M-b\\0\\0\\0\\0\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'Descriptor 6 is a pipeRead call try to read one byte over and over, I looked up to sourcecode and I think this file is responsible for this behaviorsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.","msg_date":"Thu, 9 Mar 2017 13:47:35 +0100","msg_from":"Naytro Naytro <naytro@googlemail.com>","msg_from_op":true,"msg_subject":"Performance issue after upgrading from 9.4 to 9.6"},{"msg_contents":"On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com> wrote:\n> We are having some performance issues after we upgraded to newest\n> version of PostgreSQL, before it everything was fast and smooth.\n>\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> upgraded to 9.6.2 with no improvement.\n>\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> master-slave using streaming replication.\n>\n> Problem:\n> Very high system CPU when master is streaming replication data, CPU\n> goes up to 77%. Only one process is generating this load, it's a\n> postgresql startup process. When I attached a truss to this process I\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n>\n> Descriptor 6 is a pipe\n>\n> Read call try to read one byte over and over, I looked up to source\n> code and I think this file is responsible for this behavior\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nOur latch implementation did get overhauled pretty thoroughly in 9.6;\nsee primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\ncan't guess what is going wrong here based on this information. It\nmight help if you can pull some stack backtraces from the startup\nprocess.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 12:28:19 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: Performance issue after upgrading from 9.4 to 9.6"},{"msg_contents":"Hi,\n\nOn 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\n> We are having some performance issues after we upgraded to newest\n> version of PostgreSQL, before it everything was fast and smooth.\n> \n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> upgraded to 9.6.2 with no improvement.\n> \n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> master-slave using streaming replication.\n\nWhich node is on which of those, and where is the high load?\n\n\n> Problem:\n> Very high system CPU when master is streaming replication data, CPU\n> goes up to 77%. Only one process is generating this load, it's a\n> postgresql startup process. When I attached a truss to this process I\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\n\nHm. Just to clarify: The load is on the *receiving* side, in the startup\nprocess? Because the load doesn't quite look that way...\n\n\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n> \n> Descriptor 6 is a pipe\n\nThat's presumably a latches internal pipe. Could you redo that\ntruss/strace with timestamps attached? Does truss show signals\nreceived? The above profile would e.g. make a lot more sense if not. Is\nthe wal receiver sending signals?\n\n\n> Read call try to read one byte over and over, I looked up to source\n> code and I think this file is responsible for this behavior\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nIt was \"just\" moved (and expanded), used to be at\nsrc/backend/port/unix_latch.c.\n\nThere normally shouldn't be that much \"latch traffic\" in the startup\nprocess, we'd expect to block from within WaitForWALToBecomeAvailable().\n\nHm. Any chance you've configured a recovery_min_apply_delay? Although\nI'd expect more timestamp calls in that case.\n\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 11:19:49 -0800","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: Performance issue after upgrading from 9.4 to 9.6"},{"msg_contents":"2017-03-09 20:19 GMT+01:00 Andres Freund <andres@anarazel.de>:\n\n> Hi,\n>\n> On 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\n> > We are having some performance issues after we upgraded to newest\n> > version of PostgreSQL, before it everything was fast and smooth.\n> >\n> > Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> > upgraded to 9.6.2 with no improvement.\n> >\n> > Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> > master-slave using streaming replication.\n>\n> Which node is on which of those, and where is the high load?\n>\n>\nHigh load in only on slaves, FreeBSD (master+slave) and Solaris (only\nslaves)\n\n\n>\n> > Problem:\n> > Very high system CPU when master is streaming replication data, CPU\n> > goes up to 77%. Only one process is generating this load, it's a\n> > postgresql startup process. When I attached a truss to this process I\n> > saw a lot o read calls with almost the same number of errors (EAGAIN).\n>\n> Hm. Just to clarify: The load is on the *receiving* side, in the startup\n> process? Because the load doesn't quite look that way...\n>\n>\nYes\n\n\n>\n> > read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n> >\n> > Descriptor 6 is a pipe\n>\n> That's presumably a latches internal pipe. Could you redo that\n> truss/strace with timestamps attached? Does truss show signals\n> received? The above profile would e.g. make a lot more sense if not. Is\n> the wal receiver sending signals?\n>\n>\nTruss from Solaris: http://pastebin.com/WajedZ8Y and FreeBSD:\nhttp://pastebin.com/DB5iT8na\nFreeBSD truss should show signals by default\n\nDtrace from solaris: http://pastebin.com/u03uVKbr\n\n\n>\n> > Read call try to read one byte over and over, I looked up to source\n> > code and I think this file is responsible for this behavior\n> > src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n>\n> It was \"just\" moved (and expanded), used to be at\n> src/backend/port/unix_latch.c.\n>\n> There normally shouldn't be that much \"latch traffic\" in the startup\n> process, we'd expect to block from within WaitForWALToBecomeAvailable().\n>\n> Hm. Any chance you've configured a recovery_min_apply_delay? Although\n> I'd expect more timestamp calls in that case.\n>\n>\nNo, we don't have this option configured\n\n\n>\n> Greetings,\n>\n> Andres Freund\n>\n\n2017-03-09 20:19 GMT+01:00 Andres Freund <andres@anarazel.de>:Hi,\n\nOn 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\n> We are having some performance issues after we upgraded to newest\n> version of PostgreSQL, before it everything was fast and smooth.\n>\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> upgraded to 9.6.2 with no improvement.\n>\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> master-slave using streaming replication.\n\nWhich node is on which of those, and where is the high load?\nHigh load in only on slaves, FreeBSD (master+slave) and Solaris (only slaves) \n\n> Problem:\n> Very high system CPU when master is streaming replication data, CPU\n> goes up to 77%. Only one process is generating this load, it's a\n> postgresql startup process. When I attached a truss to this process I\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\n\nHm. Just to clarify: The load is on the *receiving* side, in the startup\nprocess? Because the load doesn't quite look that way...\nYes \n\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n>\n> Descriptor 6 is a pipe\n\nThat's presumably a latches internal pipe. Could you redo that\ntruss/strace with timestamps attached? Does truss show signals\nreceived? The above profile would e.g. make a lot more sense if not. Is\nthe wal receiver sending signals?\nTruss from Solaris: http://pastebin.com/WajedZ8Y and FreeBSD: http://pastebin.com/DB5iT8naFreeBSD truss should show signals by defaultDtrace from solaris: http://pastebin.com/u03uVKbr \n\n> Read call try to read one byte over and over, I looked up to source\n> code and I think this file is responsible for this behavior\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nIt was \"just\" moved (and expanded), used to be at\nsrc/backend/port/unix_latch.c.\n\nThere normally shouldn't be that much \"latch traffic\" in the startup\nprocess, we'd expect to block from within WaitForWALToBecomeAvailable().\n\nHm. Any chance you've configured a recovery_min_apply_delay? Although\nI'd expect more timestamp calls in that case.\nNo, we don't have this option configured \n\nGreetings,\n\nAndres Freund","msg_date":"Thu, 9 Mar 2017 23:38:46 +0100","msg_from":"Naytro Naytro <naytro@googlemail.com>","msg_from_op":true,"msg_subject":"Re: Performance issue after upgrading from 9.4 to 9.6"},{"msg_contents":"2017-03-09 18:28 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:\n\n> On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com>\n> wrote:\n> > We are having some performance issues after we upgraded to newest\n> > version of PostgreSQL, before it everything was fast and smooth.\n> >\n> > Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> > upgraded to 9.6.2 with no improvement.\n> >\n> > Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> > master-slave using streaming replication.\n> >\n> > Problem:\n> > Very high system CPU when master is streaming replication data, CPU\n> > goes up to 77%. Only one process is generating this load, it's a\n> > postgresql startup process. When I attached a truss to this process I\n> > saw a lot o read calls with almost the same number of errors (EAGAIN).\n> > read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n> >\n> > Descriptor 6 is a pipe\n> >\n> > Read call try to read one byte over and over, I looked up to source\n> > code and I think this file is responsible for this behavior\n> > src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n>\n> Our latch implementation did get overhauled pretty thoroughly in 9.6;\n> see primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\n> can't guess what is going wrong here based on this information. It\n> might help if you can pull some stack backtraces from the startup\n> process.\n\n\nDtrace from solaris: http://pastebin.com/u03uVKbr\n\n2017-03-09 18:28 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com> wrote:\n> We are having some performance issues after we upgraded to newest\n> version of PostgreSQL, before it everything was fast and smooth.\n>\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\n> upgraded to 9.6.2 with no improvement.\n>\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\n> master-slave using streaming replication.\n>\n> Problem:\n> Very high system CPU when master is streaming replication data, CPU\n> goes up to 77%. Only one process is generating this load, it's a\n> postgresql startup process. When I attached a truss to this process I\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\n>\n> Descriptor 6 is a pipe\n>\n> Read call try to read one byte over and over, I looked up to source\n> code and I think this file is responsible for this behavior\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\n\nOur latch implementation did get overhauled pretty thoroughly in 9.6;\nsee primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\ncan't guess what is going wrong here based on this information. It\nmight help if you can pull some stack backtraces from the startup\nprocess.Dtrace from solaris: http://pastebin.com/u03uVKbr","msg_date":"Thu, 9 Mar 2017 23:40:09 +0100","msg_from":"Naytro Naytro <naytro@googlemail.com>","msg_from_op":true,"msg_subject":"Re: Performance issue after upgrading from 9.4 to 9.6"}],"string":"[\n {\n \"msg_contents\": \"Hi,\\n\\nWe are having some performance issues after we upgraded to newest\\nversion of PostgreSQL, before it everything was fast and smooth.\\n\\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\nupgraded to 9.6.2 with no improvement.\\n\\nSome information about our setup: Freebsd, Solaris (SmartOS), simple\\nmaster-slave using streaming replication.\\n\\nProblem:\\nVery high system CPU when master is streaming replication data, CPU\\ngoes up to 77%. Only one process is generating this load, it's a\\npostgresql startup process. When I attached a truss to this process I\\nsaw a lot o read calls with almost the same number of errors (EAGAIN).\\n\\nroot(at)d8:~ # truss -c -p 38091\\n^Csyscall seconds calls errors\\nsemop 0.001611782 198 0\\nwrite 0.000074404 2 0\\nread 2.281535100 17266 12375\\nopenat 0.000683532 48 0\\nlseek 0.177612479 20443 0\\nclose 0.000395549 48 0\\n ------------- ------- -------\\n 2.461912846 38005 12375\\n\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)\\nlseek(778,0x0,SEEK_END) = 57344 (0xe000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)\\nread(894,\\\" ~\\\\0\\\\08\\\\a\\\\M--m\\\\0\\\\0\\\\^A\\\\0\\\\M^T\\\\0000\\\"...,8192) = 8192 (0x2000)\\nlseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)\\nread(3,\\\"\\\\M^S\\\\M-P\\\\^E\\\\0\\\\^A\\\\0\\\\0\\\\0\\\\0`\\\\M-z\\\"...,8192) = 8192 (0x2000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)\\nlseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)\\nread(894,\\\"\\\\^_~\\\\0\\\\0\\\\M-H\\\\M-H\\\\M-B\\\\M-b\\\\0\\\\0\\\\^E\\\"...,8192) = 8192 (0x2000)\\nlseek(818,0x0,SEEK_END) = 57344 (0xe000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\nlseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)\\nlseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)\\nread(442,\\\"\\\\^_~\\\\0\\\\0\\\\M-P\\\\M-+\\\\M-1\\\\M-b\\\\0\\\\0\\\\0\\\\0\\\"...,8192) = 8192 (0x2000)\\nread(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n\\nDescriptor 6 is a pipe\\n\\nRead call try to read one byte over and over, I looked up to source\\ncode and I think this file is responsible for this behavior\\nsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nHi,We are having some performance issues after we upgraded to newestversion of PostgreSQL, before it everything was fast and smooth.Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now weupgraded to 9.6.2 with no improvement.Some information about our setup: Freebsd, Solaris (SmartOS), simplemaster-slave using streaming replication.Problem:Very high system CPU when master is streaming replication data, CPUgoes up to 77%. Only one process is generating this load, it's apostgresql startup process. When I attached a truss to this process Isaw a lot o read calls with almost the same number of errors (EAGAIN).root(at)d8:~ # truss -c -p 38091^Csyscall seconds calls errorssemop 0.001611782 198 0write 0.000074404 2 0read 2.281535100 17266 12375openat 0.000683532 48 0lseek 0.177612479 20443 0close 0.000395549 48 0 ------------- ------- ------- 2.461912846 38005 12375read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(444,0x0,SEEK_END) = 32571392 (0x1f10000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(7,0x0,SEEK_END) = 164487168 (0x9cde000)lseek(778,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x4b7e000,SEEK_SET) = 79159296 (0x4b7e000)read(894,\\\" ~\\\\0\\\\08\\\\a\\\\M--m\\\\0\\\\0\\\\^A\\\\0\\\\M^T\\\\0000\\\"...,8192) = 8192 (0x2000)lseek(3,0xfa6000,SEEK_SET) = 16408576 (0xfa6000)read(3,\\\"\\\\M^S\\\\M-P\\\\^E\\\\0\\\\^A\\\\0\\\\0\\\\0\\\\0`\\\\M-z\\\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x0,SEEK_END) = 139296768 (0x84d8000)lseek(894,0x449c000,SEEK_SET) = 71942144 (0x449c000)read(894,\\\"\\\\^_~\\\\0\\\\0\\\\M-H\\\\M-H\\\\M-B\\\\M-b\\\\0\\\\0\\\\^E\\\"...,8192) = 8192 (0x2000)lseek(818,0x0,SEEK_END) = 57344 (0xe000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'lseek(442,0x0,SEEK_END) = 10174464 (0x9b4000)lseek(442,0x4c4000,SEEK_SET) = 4997120 (0x4c4000)read(442,\\\"\\\\^_~\\\\0\\\\0\\\\M-P\\\\M-+\\\\M-1\\\\M-b\\\\0\\\\0\\\\0\\\\0\\\"...,8192) = 8192 (0x2000)read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'Descriptor 6 is a pipeRead call try to read one byte over and over, I looked up to sourcecode and I think this file is responsible for this behaviorsrc/backend/storage/ipc/latch.c. There was no such file in 9.4.\",\n \"msg_date\": \"Thu, 9 Mar 2017 13:47:35 +0100\",\n \"msg_from\": \"Naytro Naytro <naytro@googlemail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Performance issue after upgrading from 9.4 to 9.6\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com> wrote:\\n> We are having some performance issues after we upgraded to newest\\n> version of PostgreSQL, before it everything was fast and smooth.\\n>\\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> upgraded to 9.6.2 with no improvement.\\n>\\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> master-slave using streaming replication.\\n>\\n> Problem:\\n> Very high system CPU when master is streaming replication data, CPU\\n> goes up to 77%. Only one process is generating this load, it's a\\n> postgresql startup process. When I attached a truss to this process I\\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n>\\n> Descriptor 6 is a pipe\\n>\\n> Read call try to read one byte over and over, I looked up to source\\n> code and I think this file is responsible for this behavior\\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nOur latch implementation did get overhauled pretty thoroughly in 9.6;\\nsee primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\\ncan't guess what is going wrong here based on this information. It\\nmight help if you can pull some stack backtraces from the startup\\nprocess.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 12:28:19 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Performance issue after upgrading from 9.4 to 9.6\"\n },\n {\n \"msg_contents\": \"Hi,\\n\\nOn 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\\n> We are having some performance issues after we upgraded to newest\\n> version of PostgreSQL, before it everything was fast and smooth.\\n> \\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> upgraded to 9.6.2 with no improvement.\\n> \\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> master-slave using streaming replication.\\n\\nWhich node is on which of those, and where is the high load?\\n\\n\\n> Problem:\\n> Very high system CPU when master is streaming replication data, CPU\\n> goes up to 77%. Only one process is generating this load, it's a\\n> postgresql startup process. When I attached a truss to this process I\\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\\n\\nHm. Just to clarify: The load is on the *receiving* side, in the startup\\nprocess? Because the load doesn't quite look that way...\\n\\n\\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n> \\n> Descriptor 6 is a pipe\\n\\nThat's presumably a latches internal pipe. Could you redo that\\ntruss/strace with timestamps attached? Does truss show signals\\nreceived? The above profile would e.g. make a lot more sense if not. Is\\nthe wal receiver sending signals?\\n\\n\\n> Read call try to read one byte over and over, I looked up to source\\n> code and I think this file is responsible for this behavior\\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nIt was \\\"just\\\" moved (and expanded), used to be at\\nsrc/backend/port/unix_latch.c.\\n\\nThere normally shouldn't be that much \\\"latch traffic\\\" in the startup\\nprocess, we'd expect to block from within WaitForWALToBecomeAvailable().\\n\\nHm. Any chance you've configured a recovery_min_apply_delay? Although\\nI'd expect more timestamp calls in that case.\\n\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 11:19:49 -0800\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Performance issue after upgrading from 9.4 to 9.6\"\n },\n {\n \"msg_contents\": \"2017-03-09 20:19 GMT+01:00 Andres Freund <andres@anarazel.de>:\\n\\n> Hi,\\n>\\n> On 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\\n> > We are having some performance issues after we upgraded to newest\\n> > version of PostgreSQL, before it everything was fast and smooth.\\n> >\\n> > Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> > upgraded to 9.6.2 with no improvement.\\n> >\\n> > Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> > master-slave using streaming replication.\\n>\\n> Which node is on which of those, and where is the high load?\\n>\\n>\\nHigh load in only on slaves, FreeBSD (master+slave) and Solaris (only\\nslaves)\\n\\n\\n>\\n> > Problem:\\n> > Very high system CPU when master is streaming replication data, CPU\\n> > goes up to 77%. Only one process is generating this load, it's a\\n> > postgresql startup process. When I attached a truss to this process I\\n> > saw a lot o read calls with almost the same number of errors (EAGAIN).\\n>\\n> Hm. Just to clarify: The load is on the *receiving* side, in the startup\\n> process? Because the load doesn't quite look that way...\\n>\\n>\\nYes\\n\\n\\n>\\n> > read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n> >\\n> > Descriptor 6 is a pipe\\n>\\n> That's presumably a latches internal pipe. Could you redo that\\n> truss/strace with timestamps attached? Does truss show signals\\n> received? The above profile would e.g. make a lot more sense if not. Is\\n> the wal receiver sending signals?\\n>\\n>\\nTruss from Solaris: http://pastebin.com/WajedZ8Y and FreeBSD:\\nhttp://pastebin.com/DB5iT8na\\nFreeBSD truss should show signals by default\\n\\nDtrace from solaris: http://pastebin.com/u03uVKbr\\n\\n\\n>\\n> > Read call try to read one byte over and over, I looked up to source\\n> > code and I think this file is responsible for this behavior\\n> > src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n>\\n> It was \\\"just\\\" moved (and expanded), used to be at\\n> src/backend/port/unix_latch.c.\\n>\\n> There normally shouldn't be that much \\\"latch traffic\\\" in the startup\\n> process, we'd expect to block from within WaitForWALToBecomeAvailable().\\n>\\n> Hm. Any chance you've configured a recovery_min_apply_delay? Although\\n> I'd expect more timestamp calls in that case.\\n>\\n>\\nNo, we don't have this option configured\\n\\n\\n>\\n> Greetings,\\n>\\n> Andres Freund\\n>\\n\\n2017-03-09 20:19 GMT+01:00 Andres Freund <andres@anarazel.de>:Hi,\\n\\nOn 2017-03-09 13:47:35 +0100, Naytro Naytro wrote:\\n> We are having some performance issues after we upgraded to newest\\n> version of PostgreSQL, before it everything was fast and smooth.\\n>\\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> upgraded to 9.6.2 with no improvement.\\n>\\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> master-slave using streaming replication.\\n\\nWhich node is on which of those, and where is the high load?\\nHigh load in only on slaves, FreeBSD (master+slave) and Solaris (only slaves) \\n\\n> Problem:\\n> Very high system CPU when master is streaming replication data, CPU\\n> goes up to 77%. Only one process is generating this load, it's a\\n> postgresql startup process. When I attached a truss to this process I\\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\\n\\nHm. Just to clarify: The load is on the *receiving* side, in the startup\\nprocess? Because the load doesn't quite look that way...\\nYes \\n\\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n>\\n> Descriptor 6 is a pipe\\n\\nThat's presumably a latches internal pipe. Could you redo that\\ntruss/strace with timestamps attached? Does truss show signals\\nreceived? The above profile would e.g. make a lot more sense if not. Is\\nthe wal receiver sending signals?\\nTruss from Solaris: http://pastebin.com/WajedZ8Y and FreeBSD: http://pastebin.com/DB5iT8naFreeBSD truss should show signals by defaultDtrace from solaris: http://pastebin.com/u03uVKbr \\n\\n> Read call try to read one byte over and over, I looked up to source\\n> code and I think this file is responsible for this behavior\\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nIt was \\\"just\\\" moved (and expanded), used to be at\\nsrc/backend/port/unix_latch.c.\\n\\nThere normally shouldn't be that much \\\"latch traffic\\\" in the startup\\nprocess, we'd expect to block from within WaitForWALToBecomeAvailable().\\n\\nHm. Any chance you've configured a recovery_min_apply_delay? Although\\nI'd expect more timestamp calls in that case.\\nNo, we don't have this option configured \\n\\nGreetings,\\n\\nAndres Freund\",\n \"msg_date\": \"Thu, 9 Mar 2017 23:38:46 +0100\",\n \"msg_from\": \"Naytro Naytro <naytro@googlemail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Performance issue after upgrading from 9.4 to 9.6\"\n },\n {\n \"msg_contents\": \"2017-03-09 18:28 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:\\n\\n> On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com>\\n> wrote:\\n> > We are having some performance issues after we upgraded to newest\\n> > version of PostgreSQL, before it everything was fast and smooth.\\n> >\\n> > Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> > upgraded to 9.6.2 with no improvement.\\n> >\\n> > Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> > master-slave using streaming replication.\\n> >\\n> > Problem:\\n> > Very high system CPU when master is streaming replication data, CPU\\n> > goes up to 77%. Only one process is generating this load, it's a\\n> > postgresql startup process. When I attached a truss to this process I\\n> > saw a lot o read calls with almost the same number of errors (EAGAIN).\\n> > read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n> >\\n> > Descriptor 6 is a pipe\\n> >\\n> > Read call try to read one byte over and over, I looked up to source\\n> > code and I think this file is responsible for this behavior\\n> > src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n>\\n> Our latch implementation did get overhauled pretty thoroughly in 9.6;\\n> see primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\\n> can't guess what is going wrong here based on this information. It\\n> might help if you can pull some stack backtraces from the startup\\n> process.\\n\\n\\nDtrace from solaris: http://pastebin.com/u03uVKbr\\n\\n2017-03-09 18:28 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:On Thu, Mar 9, 2017 at 7:47 AM, Naytro Naytro <naytro@googlemail.com> wrote:\\n> We are having some performance issues after we upgraded to newest\\n> version of PostgreSQL, before it everything was fast and smooth.\\n>\\n> Upgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\\n> upgraded to 9.6.2 with no improvement.\\n>\\n> Some information about our setup: Freebsd, Solaris (SmartOS), simple\\n> master-slave using streaming replication.\\n>\\n> Problem:\\n> Very high system CPU when master is streaming replication data, CPU\\n> goes up to 77%. Only one process is generating this load, it's a\\n> postgresql startup process. When I attached a truss to this process I\\n> saw a lot o read calls with almost the same number of errors (EAGAIN).\\n> read(6,0x7fffffffa0c7,1) ERR#35 'Resource temporarily unavailable'\\n>\\n> Descriptor 6 is a pipe\\n>\\n> Read call try to read one byte over and over, I looked up to source\\n> code and I think this file is responsible for this behavior\\n> src/backend/storage/ipc/latch.c. There was no such file in 9.4.\\n\\nOur latch implementation did get overhauled pretty thoroughly in 9.6;\\nsee primarily commit 98a64d0bd713cb89e61bef6432befc4b7b5da59e. But I\\ncan't guess what is going wrong here based on this information. It\\nmight help if you can pull some stack backtraces from the startup\\nprocess.Dtrace from solaris: http://pastebin.com/u03uVKbr\",\n \"msg_date\": \"Thu, 9 Mar 2017 23:40:09 +0100\",\n \"msg_from\": \"Naytro Naytro <naytro@googlemail.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Performance issue after upgrading from 9.4 to 9.6\"\n }\n]","isConversational":false}}},{"rowIdx":44095,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Greetings,\n\nWhile reviewing Amit Langote's patch to handle partitioned tables\nproperly in various contrib modules (mostly by throwing an error since\nthings like pageinspect aren't going to work on the empty 'parent'\ntable), I went looking through contrib for other modules that do\nsomething with relkind and noticed that postgres_fdw's IMPORT SCHEMA\nwould pull in the child tables (relkind = 'r') but would ignore the\nparent table (relkind = 'P', or soon to be 'p', I guess).\n\nI tend to view this as an issue which should be added to the open items\nlist and resolved before PG10 (though perhaps it could be done after\nfeature freeze), but I could see an argument that it should be just a\ndocumented limitation of postgres_fdw and that adding such support would\nbe a new feature.\n\nIn any case, this seems like an issue that should be addressed one way\nor the other, so I'll add it to the open items list. I'm not planning\nto work on fixing it myself, but if someone proposes a patch which looks\nreasonable, I'll try to find time for it.\n\nThanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 09:15:31 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Thu, Mar 9, 2017 at 11:15 PM, Stephen Frost <sfrost@snowman.net> wrote:\n> While reviewing Amit Langote's patch to handle partitioned tables\n> properly in various contrib modules (mostly by throwing an error since\n> things like pageinspect aren't going to work on the empty 'parent'\n> table), I went looking through contrib for other modules that do\n> something with relkind and noticed that postgres_fdw's IMPORT SCHEMA\n> would pull in the child tables (relkind = 'r') but would ignore the\n> parent table (relkind = 'P', or soon to be 'p', I guess).\n\nIt is not as straight-forward as it seems. A foreign table can be\ndefined as a child (use of PARTITION OF), but not as a parent (use\nPARTITION BY), and IMPORT SCHEMA has to issue queries to create\nforeign tables. It seems to me that the correct fix here is to ignore\nchild tables that are part of a partition, and just include the parent\nin what is imported so as when querying the parent through\npostgres_fdw all the child partitions are considered automatically.\nThoughts?\n\nOf course this should be documented.\n\n> I tend to view this as an issue which should be added to the open items\n> list and resolved before PG10 (though perhaps it could be done after\n> feature freeze), but I could see an argument that it should be just a\n> documented limitation of postgres_fdw and that adding such support would\n> be a new feature.\n\nAgreed. I think that this is a bug, because any database having one\npartitioning set of tables would fail to import automatically except\nby excluding some tables, and that's annoying for the user.\n\n> In any case, this seems like an issue that should be addressed one way\n> or the other, so I'll add it to the open items list. I'm not planning\n> to work on fixing it myself, but if someone proposes a patch which looks\n> reasonable, I'll try to find time for it.\n\nWell, I worked on IMPORT SCHEMA, so I'm fine to do something. After\nthe CF is done though.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 10:26:22 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On 2017/03/10 10:26, Michael Paquier wrote:\n> On Thu, Mar 9, 2017 at 11:15 PM, Stephen Frost <sfrost@snowman.net> wrote:\n>> While reviewing Amit Langote's patch to handle partitioned tables\n>> properly in various contrib modules (mostly by throwing an error since\n>> things like pageinspect aren't going to work on the empty 'parent'\n>> table), I went looking through contrib for other modules that do\n>> something with relkind and noticed that postgres_fdw's IMPORT SCHEMA\n>> would pull in the child tables (relkind = 'r') but would ignore the\n>> parent table (relkind = 'P', or soon to be 'p', I guess).\n> \n> It is not as straight-forward as it seems. A foreign table can be\n> defined as a child (use of PARTITION OF), but not as a parent (use\n> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\n> foreign tables. It seems to me that the correct fix here is to ignore\n> child tables that are part of a partition, and just include the parent\n> in what is imported so as when querying the parent through\n> postgres_fdw all the child partitions are considered automatically.\n> Thoughts?\n\nI think that makes sense. The query in postgresImportForeignSchema() that\nfetches the information about remote tables should be fixed to include\nrelkind = 'P' tables (partitioned tables) but exclude relispartition =\ntrue (partitions). Something like below:\n\n- \"WHERE c.relkind IN ('r', 'v', 'f', 'm') \"\n+ \"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \"\n+ \" AND NOT c.relispartition \"\n\nIt means we don't import tables that are supposed to be partitions of some\ntable. If we allow importing the latter, we get access to those\npartitions anyway.\n\nI would like to hear more opinions of course.\n\n> Of course this should be documented.\n\n+1\n\n>> I tend to view this as an issue which should be added to the open items\n>> list and resolved before PG10 (though perhaps it could be done after\n>> feature freeze), but I could see an argument that it should be just a\n>> documented limitation of postgres_fdw and that adding such support would\n>> be a new feature.\n> \n> Agreed. I think that this is a bug, because any database having one\n> partitioning set of tables would fail to import automatically except\n> by excluding some tables, and that's annoying for the user.\n\nAgreed too.\n\n>> In any case, this seems like an issue that should be addressed one way\n>> or the other, so I'll add it to the open items list. I'm not planning\n>> to work on fixing it myself, but if someone proposes a patch which looks\n>> reasonable, I'll try to find time for it.\n> \n> Well, I worked on IMPORT SCHEMA, so I'm fine to do something. After\n> the CF is done though.\n\nThanks.\n\nRegards,\nAmit\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 10:47:01 +0900","msg_from":"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Thu, Mar 9, 2017 at 8:47 PM, Amit Langote\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\n>> It is not as straight-forward as it seems. A foreign table can be\n>> defined as a child (use of PARTITION OF), but not as a parent (use\n>> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\n>> foreign tables. It seems to me that the correct fix here is to ignore\n>> child tables that are part of a partition, and just include the parent\n>> in what is imported so as when querying the parent through\n>> postgres_fdw all the child partitions are considered automatically.\n>> Thoughts?\n>\n> I think that makes sense. The query in postgresImportForeignSchema() that\n> fetches the information about remote tables should be fixed to include\n> relkind = 'P' tables (partitioned tables) but exclude relispartition =\n> true (partitions). Something like below:\n>\n> - \"WHERE c.relkind IN ('r', 'v', 'f', 'm') \"\n> + \"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \"\n> + \" AND NOT c.relispartition \"\n>\n> It means we don't import tables that are supposed to be partitions of some\n> table. If we allow importing the latter, we get access to those\n> partitions anyway.\n>\n> I would like to hear more opinions of course.\n\nFor the most part, I'm not very exercised about this either way. I\nthink that the above definition seems reasonably likely to be a useful\none, but I certainly wouldn't try to insist on it in the face of\nopposition; I don't think it's 100% clear what users will want here.\n\nHowever, if we're going to do something about this, I think it should\nbe done soon. Otherwise, I'm going to advocate for reclassifying this\nissue from \"open item\" to \"possible area for future development\".\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Tue, 28 Mar 2017 23:22:54 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Wed, Mar 29, 2017 at 12:22 PM, Robert Haas <robertmhaas@gmail.com> wrote:\n> On Thu, Mar 9, 2017 at 8:47 PM, Amit Langote\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\n>>> It is not as straight-forward as it seems. A foreign table can be\n>>> defined as a child (use of PARTITION OF), but not as a parent (use\n>>> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\n>>> foreign tables. It seems to me that the correct fix here is to ignore\n>>> child tables that are part of a partition, and just include the parent\n>>> in what is imported so as when querying the parent through\n>>> postgres_fdw all the child partitions are considered automatically.\n>>> Thoughts?\n>>\n>> I think that makes sense. The query in postgresImportForeignSchema() that\n>> fetches the information about remote tables should be fixed to include\n>> relkind = 'P' tables (partitioned tables) but exclude relispartition =\n>> true (partitions). Something like below:\n>>\n>> - \"WHERE c.relkind IN ('r', 'v', 'f', 'm') \"\n>> + \"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \"\n>> + \" AND NOT c.relispartition \"\n>>\n>> It means we don't import tables that are supposed to be partitions of some\n>> table. If we allow importing the latter, we get access to those\n>> partitions anyway.\n>>\n>> I would like to hear more opinions of course.\n>\n> For the most part, I'm not very exercised about this either way. I\n> think that the above definition seems reasonably likely to be a useful\n> one, but I certainly wouldn't try to insist on it in the face of\n> opposition; I don't think it's 100% clear what users will want here.\n\nUsers like things that are friendly, and we are most likely going to\npiss them off when using postgres_fdw if they need to list manually\neach parent table from the IMPORT FOREIGN SCHEMA command.\n\n> However, if we're going to do something about this, I think it should\n> be done soon. Otherwise, I'm going to advocate for reclassifying this\n> issue from \"open item\" to \"possible area for future development\".\n\nI was just waiting for the end of the CF before sending in a patch,\nallocating now some time to look at some patches pending for reviews.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 29 Mar 2017 12:30:24 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Wed, Mar 29, 2017 at 12:30 PM, Michael Paquier\n<michael.paquier@gmail.com> wrote:\n> Users like things that are friendly, and we are most likely going to\n> piss them off when using postgres_fdw if they need to list manually\n> each parent table from the IMPORT FOREIGN SCHEMA command.\n>\n>> However, if we're going to do something about this, I think it should\n>> be done soon. Otherwise, I'm going to advocate for reclassifying this\n>> issue from \"open item\" to \"possible area for future development\".\n>\n> I was just waiting for the end of the CF before sending in a patch,\n> allocating now some time to look at some patches pending for reviews.\n\nAnd here is the promised patch to address this open item.\n-- \nMichael\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Fri, 31 Mar 2017 13:23:34 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On 2017/03/31 13:23, Michael Paquier wrote:\n> On Wed, Mar 29, 2017 at 12:30 PM, Michael Paquier\n> <michael.paquier@gmail.com> wrote:\n>> Users like things that are friendly, and we are most likely going to\n>> piss them off when using postgres_fdw if they need to list manually\n>> each parent table from the IMPORT FOREIGN SCHEMA command.\n>>\n>>> However, if we're going to do something about this, I think it should\n>>> be done soon. Otherwise, I'm going to advocate for reclassifying this\n>>> issue from \"open item\" to \"possible area for future development\".\n>>\n>> I was just waiting for the end of the CF before sending in a patch,\n>> allocating now some time to look at some patches pending for reviews.\n> \n> And here is the promised patch to address this open item.\n\nLooks good to me, except maybe:\n\n+\n+ <para>\n+ For partitioned tables, partitions are automatically excluded from the\n+ schema data imported. Only the definition of partitioned tables is\nincluded\n+ to give access to the full data set of all partitions present remotely.\n+ </para>\n+\n\nOnly the definitions of \"root\" partitioned tables, because when using\nmulti-level partitioning, there would be partitioned tables that won't be\nincluded (because, relispartition=true).\n\nIf you agree, then this code comment too could use the same terminology:\n\n+ * Ignore table data for partitions and only include the parent\n+ * definitions to allow access to the complete remote data set\n+ * locally in the schema imported.\n+ *\n\nThanks,\nAmit\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 13:37:55 +0900","msg_from":"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Fri, Mar 31, 2017 at 1:37 PM, Amit Langote\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\n> + <para>\n> + For partitioned tables, partitions are automatically excluded from the\n> + schema data imported. Only the definition of partitioned tables is\n> included\n> + to give access to the full data set of all partitions present remotely.\n> + </para>\n> +\n>\n> Only the definitions of \"root\" partitioned tables, because when using\n> multi-level partitioning, there would be partitioned tables that won't be\n> included (because, relispartition=true).\n>\n> If you agree, then this code comment too could use the same terminology:\n>\n> + * Ignore table data for partitions and only include the parent\n> + * definitions to allow access to the complete remote data set\n> + * locally in the schema imported.\n> + *\n\nThat makes sense. Updated this way as I have my hands on it now.\n-- \nMichael\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Fri, 31 Mar 2017 13:46:28 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On 2017/03/31 13:46, Michael Paquier wrote:\n> On Fri, Mar 31, 2017 at 1:37 PM, Amit Langote\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\n>> + <para>\n>> + For partitioned tables, partitions are automatically excluded from the\n>> + schema data imported. Only the definition of partitioned tables is\n>> included\n>> + to give access to the full data set of all partitions present remotely.\n>> + </para>\n>> +\n>>\n>> Only the definitions of \"root\" partitioned tables, because when using\n>> multi-level partitioning, there would be partitioned tables that won't be\n>> included (because, relispartition=true).\n>>\n>> If you agree, then this code comment too could use the same terminology:\n>>\n>> + * Ignore table data for partitions and only include the parent\n>> + * definitions to allow access to the complete remote data set\n>> + * locally in the schema imported.\n>> + *\n> \n> That makes sense. Updated this way as I have my hands on it now.\n\nThanks, no more comments from my side.\n\nRegards,\nAmit\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 13:51:59 +0900","msg_from":"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Fri, Mar 31, 2017 at 12:51 AM, Amit Langote\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\n> Thanks, no more comments from my side.\n\nCommitted after rewording the documentation.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 15:25:19 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On 2017-03-31 15:25:19 -0400, Robert Haas wrote:\n> On Fri, Mar 31, 2017 at 12:51 AM, Amit Langote\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\n> > Thanks, no more comments from my side.\n> \n> Committed after rewording the documentation.\n\nHm. Wonder if something like that shouldn't be backpatched - because\notherwise using postgres_fdw from an old server against a newer one will\ndo weird stuff. I don't know what kind of policy we've committed to\nwith postgresImportForeignSchema...\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 12:31:35 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\n>> Committed after rewording the documentation.\n>\n> Hm. Wonder if something like that shouldn't be backpatched - because\n> otherwise using postgres_fdw from an old server against a newer one will\n> do weird stuff. I don't know what kind of policy we've committed to\n> with postgresImportForeignSchema...\n\nI don't think I'd like to promise that postgres_fdw will always be\nforward-compatible. Backward-compatibility is hard enough already.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 15:32:33 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"Robert Haas <robertmhaas@gmail.com> writes:\n> On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\n>> Hm. Wonder if something like that shouldn't be backpatched - because\n>> otherwise using postgres_fdw from an old server against a newer one will\n>> do weird stuff. I don't know what kind of policy we've committed to\n>> with postgresImportForeignSchema...\n\n> I don't think I'd like to promise that postgres_fdw will always be\n> forward-compatible. Backward-compatibility is hard enough already.\n\nUnless I'm missing something, the behavior will be that an older\nversion will simply ignore remote partitioned tables (they will not\npass the relkind filter in the query). Seems pretty fail-soft,\nso I think it's fine.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 15:55:30 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"},{"msg_contents":"On Sat, Apr 1, 2017 at 4:55 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\n> Robert Haas <robertmhaas@gmail.com> writes:\n>> On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\n>>> Hm. Wonder if something like that shouldn't be backpatched - because\n>>> otherwise using postgres_fdw from an old server against a newer one will\n>>> do weird stuff. I don't know what kind of policy we've committed to\n>>> with postgresImportForeignSchema...\n>\n>> I don't think I'd like to promise that postgres_fdw will always be\n>> forward-compatible. Backward-compatibility is hard enough already.\n\nThanks for the commit.\n\n> Unless I'm missing something, the behavior will be that an older\n> version will simply ignore remote partitioned tables (they will not\n> pass the relkind filter in the query). Seems pretty fail-soft,\n> so I think it's fine.\n\nYeah, I would suggest to revisit that if we get actual complaints, but\nI would not push much in favor of it. It's not an area where nothing\ncan be done to improve the user experience.\n-- \nMichael\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Sat, 1 Apr 2017 08:19:04 +0900","msg_from":"Michael Paquier <michael.paquier@gmail.com>","msg_from_op":false,"msg_subject":"Re: postgres_fdw IMPORT SCHEMA and partitioned tables"}],"string":"[\n {\n \"msg_contents\": \"Greetings,\\n\\nWhile reviewing Amit Langote's patch to handle partitioned tables\\nproperly in various contrib modules (mostly by throwing an error since\\nthings like pageinspect aren't going to work on the empty 'parent'\\ntable), I went looking through contrib for other modules that do\\nsomething with relkind and noticed that postgres_fdw's IMPORT SCHEMA\\nwould pull in the child tables (relkind = 'r') but would ignore the\\nparent table (relkind = 'P', or soon to be 'p', I guess).\\n\\nI tend to view this as an issue which should be added to the open items\\nlist and resolved before PG10 (though perhaps it could be done after\\nfeature freeze), but I could see an argument that it should be just a\\ndocumented limitation of postgres_fdw and that adding such support would\\nbe a new feature.\\n\\nIn any case, this seems like an issue that should be addressed one way\\nor the other, so I'll add it to the open items list. I'm not planning\\nto work on fixing it myself, but if someone proposes a patch which looks\\nreasonable, I'll try to find time for it.\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 09:15:31 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 11:15 PM, Stephen Frost <sfrost@snowman.net> wrote:\\n> While reviewing Amit Langote's patch to handle partitioned tables\\n> properly in various contrib modules (mostly by throwing an error since\\n> things like pageinspect aren't going to work on the empty 'parent'\\n> table), I went looking through contrib for other modules that do\\n> something with relkind and noticed that postgres_fdw's IMPORT SCHEMA\\n> would pull in the child tables (relkind = 'r') but would ignore the\\n> parent table (relkind = 'P', or soon to be 'p', I guess).\\n\\nIt is not as straight-forward as it seems. A foreign table can be\\ndefined as a child (use of PARTITION OF), but not as a parent (use\\nPARTITION BY), and IMPORT SCHEMA has to issue queries to create\\nforeign tables. It seems to me that the correct fix here is to ignore\\nchild tables that are part of a partition, and just include the parent\\nin what is imported so as when querying the parent through\\npostgres_fdw all the child partitions are considered automatically.\\nThoughts?\\n\\nOf course this should be documented.\\n\\n> I tend to view this as an issue which should be added to the open items\\n> list and resolved before PG10 (though perhaps it could be done after\\n> feature freeze), but I could see an argument that it should be just a\\n> documented limitation of postgres_fdw and that adding such support would\\n> be a new feature.\\n\\nAgreed. I think that this is a bug, because any database having one\\npartitioning set of tables would fail to import automatically except\\nby excluding some tables, and that's annoying for the user.\\n\\n> In any case, this seems like an issue that should be addressed one way\\n> or the other, so I'll add it to the open items list. I'm not planning\\n> to work on fixing it myself, but if someone proposes a patch which looks\\n> reasonable, I'll try to find time for it.\\n\\nWell, I worked on IMPORT SCHEMA, so I'm fine to do something. After\\nthe CF is done though.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 10:26:22 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On 2017/03/10 10:26, Michael Paquier wrote:\\n> On Thu, Mar 9, 2017 at 11:15 PM, Stephen Frost <sfrost@snowman.net> wrote:\\n>> While reviewing Amit Langote's patch to handle partitioned tables\\n>> properly in various contrib modules (mostly by throwing an error since\\n>> things like pageinspect aren't going to work on the empty 'parent'\\n>> table), I went looking through contrib for other modules that do\\n>> something with relkind and noticed that postgres_fdw's IMPORT SCHEMA\\n>> would pull in the child tables (relkind = 'r') but would ignore the\\n>> parent table (relkind = 'P', or soon to be 'p', I guess).\\n> \\n> It is not as straight-forward as it seems. A foreign table can be\\n> defined as a child (use of PARTITION OF), but not as a parent (use\\n> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\\n> foreign tables. It seems to me that the correct fix here is to ignore\\n> child tables that are part of a partition, and just include the parent\\n> in what is imported so as when querying the parent through\\n> postgres_fdw all the child partitions are considered automatically.\\n> Thoughts?\\n\\nI think that makes sense. The query in postgresImportForeignSchema() that\\nfetches the information about remote tables should be fixed to include\\nrelkind = 'P' tables (partitioned tables) but exclude relispartition =\\ntrue (partitions). Something like below:\\n\\n- \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm') \\\"\\n+ \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \\\"\\n+ \\\" AND NOT c.relispartition \\\"\\n\\nIt means we don't import tables that are supposed to be partitions of some\\ntable. If we allow importing the latter, we get access to those\\npartitions anyway.\\n\\nI would like to hear more opinions of course.\\n\\n> Of course this should be documented.\\n\\n+1\\n\\n>> I tend to view this as an issue which should be added to the open items\\n>> list and resolved before PG10 (though perhaps it could be done after\\n>> feature freeze), but I could see an argument that it should be just a\\n>> documented limitation of postgres_fdw and that adding such support would\\n>> be a new feature.\\n> \\n> Agreed. I think that this is a bug, because any database having one\\n> partitioning set of tables would fail to import automatically except\\n> by excluding some tables, and that's annoying for the user.\\n\\nAgreed too.\\n\\n>> In any case, this seems like an issue that should be addressed one way\\n>> or the other, so I'll add it to the open items list. I'm not planning\\n>> to work on fixing it myself, but if someone proposes a patch which looks\\n>> reasonable, I'll try to find time for it.\\n> \\n> Well, I worked on IMPORT SCHEMA, so I'm fine to do something. After\\n> the CF is done though.\\n\\nThanks.\\n\\nRegards,\\nAmit\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 10:47:01 +0900\",\n \"msg_from\": \"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 8:47 PM, Amit Langote\\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n>> It is not as straight-forward as it seems. A foreign table can be\\n>> defined as a child (use of PARTITION OF), but not as a parent (use\\n>> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\\n>> foreign tables. It seems to me that the correct fix here is to ignore\\n>> child tables that are part of a partition, and just include the parent\\n>> in what is imported so as when querying the parent through\\n>> postgres_fdw all the child partitions are considered automatically.\\n>> Thoughts?\\n>\\n> I think that makes sense. The query in postgresImportForeignSchema() that\\n> fetches the information about remote tables should be fixed to include\\n> relkind = 'P' tables (partitioned tables) but exclude relispartition =\\n> true (partitions). Something like below:\\n>\\n> - \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm') \\\"\\n> + \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \\\"\\n> + \\\" AND NOT c.relispartition \\\"\\n>\\n> It means we don't import tables that are supposed to be partitions of some\\n> table. If we allow importing the latter, we get access to those\\n> partitions anyway.\\n>\\n> I would like to hear more opinions of course.\\n\\nFor the most part, I'm not very exercised about this either way. I\\nthink that the above definition seems reasonably likely to be a useful\\none, but I certainly wouldn't try to insist on it in the face of\\nopposition; I don't think it's 100% clear what users will want here.\\n\\nHowever, if we're going to do something about this, I think it should\\nbe done soon. Otherwise, I'm going to advocate for reclassifying this\\nissue from \\\"open item\\\" to \\\"possible area for future development\\\".\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Tue, 28 Mar 2017 23:22:54 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 29, 2017 at 12:22 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Thu, Mar 9, 2017 at 8:47 PM, Amit Langote\\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n>>> It is not as straight-forward as it seems. A foreign table can be\\n>>> defined as a child (use of PARTITION OF), but not as a parent (use\\n>>> PARTITION BY), and IMPORT SCHEMA has to issue queries to create\\n>>> foreign tables. It seems to me that the correct fix here is to ignore\\n>>> child tables that are part of a partition, and just include the parent\\n>>> in what is imported so as when querying the parent through\\n>>> postgres_fdw all the child partitions are considered automatically.\\n>>> Thoughts?\\n>>\\n>> I think that makes sense. The query in postgresImportForeignSchema() that\\n>> fetches the information about remote tables should be fixed to include\\n>> relkind = 'P' tables (partitioned tables) but exclude relispartition =\\n>> true (partitions). Something like below:\\n>>\\n>> - \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm') \\\"\\n>> + \\\"WHERE c.relkind IN ('r', 'v', 'f', 'm', 'P') \\\"\\n>> + \\\" AND NOT c.relispartition \\\"\\n>>\\n>> It means we don't import tables that are supposed to be partitions of some\\n>> table. If we allow importing the latter, we get access to those\\n>> partitions anyway.\\n>>\\n>> I would like to hear more opinions of course.\\n>\\n> For the most part, I'm not very exercised about this either way. I\\n> think that the above definition seems reasonably likely to be a useful\\n> one, but I certainly wouldn't try to insist on it in the face of\\n> opposition; I don't think it's 100% clear what users will want here.\\n\\nUsers like things that are friendly, and we are most likely going to\\npiss them off when using postgres_fdw if they need to list manually\\neach parent table from the IMPORT FOREIGN SCHEMA command.\\n\\n> However, if we're going to do something about this, I think it should\\n> be done soon. Otherwise, I'm going to advocate for reclassifying this\\n> issue from \\\"open item\\\" to \\\"possible area for future development\\\".\\n\\nI was just waiting for the end of the CF before sending in a patch,\\nallocating now some time to look at some patches pending for reviews.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 29 Mar 2017 12:30:24 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 29, 2017 at 12:30 PM, Michael Paquier\\n<michael.paquier@gmail.com> wrote:\\n> Users like things that are friendly, and we are most likely going to\\n> piss them off when using postgres_fdw if they need to list manually\\n> each parent table from the IMPORT FOREIGN SCHEMA command.\\n>\\n>> However, if we're going to do something about this, I think it should\\n>> be done soon. Otherwise, I'm going to advocate for reclassifying this\\n>> issue from \\\"open item\\\" to \\\"possible area for future development\\\".\\n>\\n> I was just waiting for the end of the CF before sending in a patch,\\n> allocating now some time to look at some patches pending for reviews.\\n\\nAnd here is the promised patch to address this open item.\\n-- \\nMichael\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 31 Mar 2017 13:23:34 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On 2017/03/31 13:23, Michael Paquier wrote:\\n> On Wed, Mar 29, 2017 at 12:30 PM, Michael Paquier\\n> <michael.paquier@gmail.com> wrote:\\n>> Users like things that are friendly, and we are most likely going to\\n>> piss them off when using postgres_fdw if they need to list manually\\n>> each parent table from the IMPORT FOREIGN SCHEMA command.\\n>>\\n>>> However, if we're going to do something about this, I think it should\\n>>> be done soon. Otherwise, I'm going to advocate for reclassifying this\\n>>> issue from \\\"open item\\\" to \\\"possible area for future development\\\".\\n>>\\n>> I was just waiting for the end of the CF before sending in a patch,\\n>> allocating now some time to look at some patches pending for reviews.\\n> \\n> And here is the promised patch to address this open item.\\n\\nLooks good to me, except maybe:\\n\\n+\\n+ <para>\\n+ For partitioned tables, partitions are automatically excluded from the\\n+ schema data imported. Only the definition of partitioned tables is\\nincluded\\n+ to give access to the full data set of all partitions present remotely.\\n+ </para>\\n+\\n\\nOnly the definitions of \\\"root\\\" partitioned tables, because when using\\nmulti-level partitioning, there would be partitioned tables that won't be\\nincluded (because, relispartition=true).\\n\\nIf you agree, then this code comment too could use the same terminology:\\n\\n+ * Ignore table data for partitions and only include the parent\\n+ * definitions to allow access to the complete remote data set\\n+ * locally in the schema imported.\\n+ *\\n\\nThanks,\\nAmit\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 13:37:55 +0900\",\n \"msg_from\": \"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 1:37 PM, Amit Langote\\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n> + <para>\\n> + For partitioned tables, partitions are automatically excluded from the\\n> + schema data imported. Only the definition of partitioned tables is\\n> included\\n> + to give access to the full data set of all partitions present remotely.\\n> + </para>\\n> +\\n>\\n> Only the definitions of \\\"root\\\" partitioned tables, because when using\\n> multi-level partitioning, there would be partitioned tables that won't be\\n> included (because, relispartition=true).\\n>\\n> If you agree, then this code comment too could use the same terminology:\\n>\\n> + * Ignore table data for partitions and only include the parent\\n> + * definitions to allow access to the complete remote data set\\n> + * locally in the schema imported.\\n> + *\\n\\nThat makes sense. Updated this way as I have my hands on it now.\\n-- \\nMichael\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 31 Mar 2017 13:46:28 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On 2017/03/31 13:46, Michael Paquier wrote:\\n> On Fri, Mar 31, 2017 at 1:37 PM, Amit Langote\\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n>> + <para>\\n>> + For partitioned tables, partitions are automatically excluded from the\\n>> + schema data imported. Only the definition of partitioned tables is\\n>> included\\n>> + to give access to the full data set of all partitions present remotely.\\n>> + </para>\\n>> +\\n>>\\n>> Only the definitions of \\\"root\\\" partitioned tables, because when using\\n>> multi-level partitioning, there would be partitioned tables that won't be\\n>> included (because, relispartition=true).\\n>>\\n>> If you agree, then this code comment too could use the same terminology:\\n>>\\n>> + * Ignore table data for partitions and only include the parent\\n>> + * definitions to allow access to the complete remote data set\\n>> + * locally in the schema imported.\\n>> + *\\n> \\n> That makes sense. Updated this way as I have my hands on it now.\\n\\nThanks, no more comments from my side.\\n\\nRegards,\\nAmit\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 13:51:59 +0900\",\n \"msg_from\": \"Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 12:51 AM, Amit Langote\\n<Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n> Thanks, no more comments from my side.\\n\\nCommitted after rewording the documentation.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 15:25:19 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On 2017-03-31 15:25:19 -0400, Robert Haas wrote:\\n> On Fri, Mar 31, 2017 at 12:51 AM, Amit Langote\\n> <Langote_Amit_f8@lab.ntt.co.jp> wrote:\\n> > Thanks, no more comments from my side.\\n> \\n> Committed after rewording the documentation.\\n\\nHm. Wonder if something like that shouldn't be backpatched - because\\notherwise using postgres_fdw from an old server against a newer one will\\ndo weird stuff. I don't know what kind of policy we've committed to\\nwith postgresImportForeignSchema...\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 12:31:35 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\\n>> Committed after rewording the documentation.\\n>\\n> Hm. Wonder if something like that shouldn't be backpatched - because\\n> otherwise using postgres_fdw from an old server against a newer one will\\n> do weird stuff. I don't know what kind of policy we've committed to\\n> with postgresImportForeignSchema...\\n\\nI don't think I'd like to promise that postgres_fdw will always be\\nforward-compatible. Backward-compatibility is hard enough already.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 15:32:33 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"Robert Haas <robertmhaas@gmail.com> writes:\\n> On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\\n>> Hm. Wonder if something like that shouldn't be backpatched - because\\n>> otherwise using postgres_fdw from an old server against a newer one will\\n>> do weird stuff. I don't know what kind of policy we've committed to\\n>> with postgresImportForeignSchema...\\n\\n> I don't think I'd like to promise that postgres_fdw will always be\\n> forward-compatible. Backward-compatibility is hard enough already.\\n\\nUnless I'm missing something, the behavior will be that an older\\nversion will simply ignore remote partitioned tables (they will not\\npass the relkind filter in the query). Seems pretty fail-soft,\\nso I think it's fine.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 15:55:30 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n },\n {\n \"msg_contents\": \"On Sat, Apr 1, 2017 at 4:55 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\\n> Robert Haas <robertmhaas@gmail.com> writes:\\n>> On Fri, Mar 31, 2017 at 3:31 PM, Andres Freund <andres@anarazel.de> wrote:\\n>>> Hm. Wonder if something like that shouldn't be backpatched - because\\n>>> otherwise using postgres_fdw from an old server against a newer one will\\n>>> do weird stuff. I don't know what kind of policy we've committed to\\n>>> with postgresImportForeignSchema...\\n>\\n>> I don't think I'd like to promise that postgres_fdw will always be\\n>> forward-compatible. Backward-compatibility is hard enough already.\\n\\nThanks for the commit.\\n\\n> Unless I'm missing something, the behavior will be that an older\\n> version will simply ignore remote partitioned tables (they will not\\n> pass the relkind filter in the query). Seems pretty fail-soft,\\n> so I think it's fine.\\n\\nYeah, I would suggest to revisit that if we get actual complaints, but\\nI would not push much in favor of it. It's not an area where nothing\\ncan be done to improve the user experience.\\n-- \\nMichael\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Sat, 1 Apr 2017 08:19:04 +0900\",\n \"msg_from\": \"Michael Paquier <michael.paquier@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: postgres_fdw IMPORT SCHEMA and partitioned tables\"\n }\n]","isConversational":false}}},{"rowIdx":44096,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Greetings,\n\nWhile going over the contrib modules, I noticed that sepgsql was not\nupdated for partitioned tables. What that appears to mean is that it's\nnot possible to define labels on partitioned tables. As I recall,\naccessing the parent of a table will, similar to the GRANT system, not\nperform checkes against the child tables, meaning that there's no way to\nhave SELinux checks properly enforced when partitioned tables are being\nused.\n\nThis is an issue which should be resolved for PG10, so I'll add it to\nthe open items list.\n\nThanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 09:47:18 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"partitioned tables and contrib/sepgsql"},{"msg_contents":"On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n> Greetings,\n>\n> While going over the contrib modules, I noticed that sepgsql was not\n> updated for partitioned tables. What that appears to mean is that it's\n> not possible to define labels on partitioned tables. As I recall,\n> accessing the parent of a table will, similar to the GRANT system, not\n> perform checkes against the child tables, meaning that there's no way to\n> have SELinux checks properly enforced when partitioned tables are being\n> used.\n\nI'll start taking a look at this. Presumably we'd just extend existing\nobject_access_hooks to cover partitioned tables?\n\n>\n> This is an issue which should be resolved for PG10, so I'll add it to\n> the open items list.\n\nI'll grab it. Thanks.\n\n--Mike\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 09:59:14 -0500","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Mike,\n\n* Mike Palmiotto (mike.palmiotto@crunchydata.com) wrote:\n> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n> > While going over the contrib modules, I noticed that sepgsql was not\n> > updated for partitioned tables. What that appears to mean is that it's\n> > not possible to define labels on partitioned tables. As I recall,\n> > accessing the parent of a table will, similar to the GRANT system, not\n> > perform checkes against the child tables, meaning that there's no way to\n> > have SELinux checks properly enforced when partitioned tables are being\n> > used.\n> \n> I'll start taking a look at this. Presumably we'd just extend existing\n> object_access_hooks to cover partitioned tables?\n\nAt least on first blush that seems like the right approach. We'll need\nto make sure that the SECURITY LABEL system will properly work with\npartitioned tables too, of course, and that the checks are called when a\nuser queries a partitioned table. Then we'll need regression tests to\nmake sure we get it all correct and don't screw it up in the future. ;)\n\n> > This is an issue which should be resolved for PG10, so I'll add it to\n> > the open items list.\n> \n> I'll grab it. Thanks.\n\nExcellent, thanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 10:05:00 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n> While going over the contrib modules, I noticed that sepgsql was not\n> updated for partitioned tables. What that appears to mean is that it's\n> not possible to define labels on partitioned tables.\n\nIt works for me:\n\nrhaas=# load 'dummy_seclabel';\nLOAD\nrhaas=# create table foo (a int, b text) partition by range (a);\nCREATE TABLE\nrhaas=# security label on table foo is 'classified';\nSECURITY LABEL\n\nWhat exactly is the problem you're seeing?\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 27 Mar 2017 10:47:31 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n>> While going over the contrib modules, I noticed that sepgsql was not\n>> updated for partitioned tables. What that appears to mean is that it's\n>> not possible to define labels on partitioned tables.\n>\n> It works for me:\n>\n> rhaas=# load 'dummy_seclabel';\n> LOAD\n> rhaas=# create table foo (a int, b text) partition by range (a);\n> CREATE TABLE\n> rhaas=# security label on table foo is 'classified';\n> SECURITY LABEL\n>\n> What exactly is the problem you're seeing?\n\nIIRC the initial concern was that contrib/sepgsql was not updated for\nRELKIND_PARTITIONED_TABLE, so the post-create hook would not work\nproperly for partitioned tables.\n\nI've looked into it a bit and saw a post-alter hook in\nStoreCatalogInheritance1. It seems like it may just be an issue of\nadding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\n\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 27 Mar 2017 11:22:36 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Mon, Mar 27, 2017 at 11:22 AM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n>> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n>>> While going over the contrib modules, I noticed that sepgsql was not\n>>> updated for partitioned tables. What that appears to mean is that it's\n>>> not possible to define labels on partitioned tables.\n>>\n>> It works for me:\n>>\n>> rhaas=# load 'dummy_seclabel';\n>> LOAD\n>> rhaas=# create table foo (a int, b text) partition by range (a);\n>> CREATE TABLE\n>> rhaas=# security label on table foo is 'classified';\n>> SECURITY LABEL\n>>\n>> What exactly is the problem you're seeing?\n>\n> IIRC the initial concern was that contrib/sepgsql was not updated for\n> RELKIND_PARTITIONED_TABLE, so the post-create hook would not work\n> properly for partitioned tables.\n>\n> I've looked into it a bit and saw a post-alter hook in\n> StoreCatalogInheritance1. It seems like it may just be an issue of\n> adding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\n\nI thought that kind of thing might be the issue, but it didn't seem to\nmatch Stephen's description. Adding RELKIND_PARTITIONED_TABLE to that\nfunction seems like it would probably be sufficient to make that hook\nwork, although that would need to be tested, but there are numerous\nother references to RELKIND_RELATION in contrib/sepgsql, some of which\nprobably also need to be updated to consider\nRELKIND_PARTITIONED_TABLE.\n\nIn my view, this is really an enhancement to sepgsql to handle a new\nPostgreSQL feature rather than a defect in the partitioning commit, so\nI don't think it falls on Amit Langote (as the author) or me (as the\ncommitter) to fix. There might similarly be updates to sepgsql to do\nsomething with permissions on logical replication's new publication\nand subscription objects, but we should similarly regard those as\npossible new features, not things that Petr or Peter need to work on.\n Note that sepgsql hasn't been updated to work with RLS yet, either,\nbut we didn't regard that as an open item for RLS, or if we did the\nresolution was just to document it. I am not opposed to giving a\nlittle more time to get this straightened out, but if a patch doesn't\nshow up fairly soon then I think we should just document that sepgsql\ndoesn't support partitioned tables in v10. sepgsql has a fairly\nlengthy list of implementation restrictions already, so one more is\nnot going to kill anybody -- or if it will then that person should\nproduce a patch soon.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 27 Mar 2017 11:46:42 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n> On Mon, Mar 27, 2017 at 11:22 AM, Mike Palmiotto\n> <mike.palmiotto@crunchydata.com> wrote:\n>> On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n>>> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\n>>>> While going over the contrib modules, I noticed that sepgsql was not\n>>>> updated for partitioned tables. What that appears to mean is that it's\n>>>> not possible to define labels on partitioned tables.\n>>>\n>>> It works for me:\n>>>\n>>> rhaas=# load 'dummy_seclabel';\n>>> LOAD\n>>> rhaas=# create table foo (a int, b text) partition by range (a);\n>>> CREATE TABLE\n>>> rhaas=# security label on table foo is 'classified';\n>>> SECURITY LABEL\n>>>\n>>> What exactly is the problem you're seeing?\n>>\n>> IIRC the initial concern was that contrib/sepgsql was not updated for\n>> RELKIND_PARTITIONED_TABLE, so the post-create hook would not work\n>> properly for partitioned tables.\n>>\n>> I've looked into it a bit and saw a post-alter hook in\n>> StoreCatalogInheritance1. It seems like it may just be an issue of\n>> adding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\n>\n> I thought that kind of thing might be the issue, but it didn't seem to\n> match Stephen's description. Adding RELKIND_PARTITIONED_TABLE to that\n> function seems like it would probably be sufficient to make that hook\n> work, although that would need to be tested, but there are numerous\n> other references to RELKIND_RELATION in contrib/sepgsql, some of which\n> probably also need to be updated to consider\n> RELKIND_PARTITIONED_TABLE.\n\nAgreed.\n\n>\n> In my view, this is really an enhancement to sepgsql to handle a new\n> PostgreSQL feature rather than a defect in the partitioning commit, so\n> I don't think it falls on Amit Langote (as the author) or me (as the\n> committer) to fix. There might similarly be updates to sepgsql to do\n> something with permissions on logical replication's new publication\n> and subscription objects, but we should similarly regard those as\n> possible new features, not things that Petr or Peter need to work on.\n\nI'll keep these items in mind for future reference. Thanks.\n\n> Note that sepgsql hasn't been updated to work with RLS yet, either,\n> but we didn't regard that as an open item for RLS, or if we did the\n> resolution was just to document it. I am not opposed to giving a\n> little more time to get this straightened out, but if a patch doesn't\n> show up fairly soon then I think we should just document that sepgsql\n> doesn't support partitioned tables in v10. sepgsql has a fairly\n> lengthy list of implementation restrictions already, so one more is\n> not going to kill anybody -- or if it will then that person should\n> produce a patch soon.\n\nOkay, I'll make sure I get something fleshed out today or tomorrow.\n\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 27 Mar 2017 12:09:08 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Mon, Mar 27, 2017 at 12:09 PM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n>> <snip>\n>> Note that sepgsql hasn't been updated to work with RLS yet, either,\n>> but we didn't regard that as an open item for RLS, or if we did the\n>> resolution was just to document it. I am not opposed to giving a\n>> little more time to get this straightened out, but if a patch doesn't\n>> show up fairly soon then I think we should just document that sepgsql\n>> doesn't support partitioned tables in v10. sepgsql has a fairly\n>> lengthy list of implementation restrictions already, so one more is\n>> not going to kill anybody -- or if it will then that person should\n>> produce a patch soon.\n>\n> Okay, I'll make sure I get something fleshed out today or tomorrow.\n\nApologies for the delay. I was waffling over whether to reference\nPartitionedRelationId in sepgsql, but ended up deciding to just handle\nRELKIND_PARTITIONED_TABLE and treat the classOid as\nRelationRelationId. Seeing as there is a relid in pg_class which\ncorresponds to the partitioned table, this chosen route seemed\nacceptable.\n\nHere is a demonstration of the partitioned table working with sepgsql hooks:\nhttps://gist.github.com/anonymous/b10f476a95ae9cdd39b83ef872d4b1e6\n\nAttached you will find two patches, which were rebased on master as of\n156d388 (applied with `git am --revert [patch file]`). The first gets\nrid of some pesky compiler warnings and the second implements the\nsepgsql handling of partitioned tables.\n\nThanks,\n\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Fri, 31 Mar 2017 14:14:59 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> On Mon, Mar 27, 2017 at 12:09 PM, Mike Palmiotto\n> <mike.palmiotto@crunchydata.com> wrote:\n>> On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\n>>> <snip>\n>>> Note that sepgsql hasn't been updated to work with RLS yet, either,\n>>> but we didn't regard that as an open item for RLS, or if we did the\n>>> resolution was just to document it. I am not opposed to giving a\n>>> little more time to get this straightened out, but if a patch doesn't\n>>> show up fairly soon then I think we should just document that sepgsql\n>>> doesn't support partitioned tables in v10. sepgsql has a fairly\n>>> lengthy list of implementation restrictions already, so one more is\n>>> not going to kill anybody -- or if it will then that person should\n>>> produce a patch soon.\n>>\n>> Okay, I'll make sure I get something fleshed out today or tomorrow.\n>\n> Apologies for the delay. I was waffling over whether to reference\n> PartitionedRelationId in sepgsql, but ended up deciding to just handle\n> RELKIND_PARTITIONED_TABLE and treat the classOid as\n> RelationRelationId. Seeing as there is a relid in pg_class which\n> corresponds to the partitioned table, this chosen route seemed\n> acceptable.\n\nNewest patches remove cruft from said waffling. No need to include\npg_partitioned_table.h if we're not referencing PartitionedRelationId.\n\n>\n> Here is a demonstration of the partitioned table working with sepgsql hooks:\n> https://gist.github.com/anonymous/b10f476a95ae9cdd39b83ef872d4b1e6\n>\n> Attached you will find two patches, which were rebased on master as of\n> 156d388 (applied with `git am --revert [patch file]`). The first gets\n> rid of some pesky compiler warnings and the second implements the\n> sepgsql handling of partitioned tables.\n\nThat should have read `git am --reject [patch file]`. Apologies for\nthe inaccuracy.\n\nThanks,\n\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Fri, 31 Mar 2017 14:42:00 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> Attached you will find two patches, which were rebased on master as of\n> 156d388 (applied with `git am --revert [patch file]`). The first gets\n> rid of some pesky compiler warnings and the second implements the\n> sepgsql handling of partitioned tables.\n\n0001 has the problem that we have a firm rule against putting any\n#includes whatsoever before \"postgres.h\". This stdbool issue has come\nup before, though, and I fear we're going to need to do something\nabout it.\n\n0002 looks extremely straightforward, but I wonder if we could get one\nof the people on this list who knows about sepgsql to have a look?\n(Stephen Frost, Joe Conway, KaiGai Kohei...)\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 31 Mar 2017 20:23:39 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Fri, Mar 31, 2017 at 8:23 PM, Robert Haas <robertmhaas@gmail.com> wrote:\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\n> <mike.palmiotto@crunchydata.com> wrote:\n>> Attached you will find two patches, which were rebased on master as of\n>> 156d388 (applied with `git am --revert [patch file]`). The first gets\n>> rid of some pesky compiler warnings and the second implements the\n>> sepgsql handling of partitioned tables.\n>\n> 0001 has the problem that we have a firm rule against putting any\n> #includes whatsoever before \"postgres.h\". This stdbool issue has come\n> up before, though, and I fear we're going to need to do something\n> about it.\n\nYeah, I recall this rule. The only things I can really think of to\nsolve the problem are:\n1) #define _STDBOOL_H in postgres's c.h once bool and the like have\nbeen defined, so we can avoid re-definition.\n2) Enforce that any code utilizing the stdbool header manage the\nre-definition with some combination of #undef/#define/#typedef and\ndocument the issue somewhere.\n\nI'd be more inclined to believe 2) is the correct solution, since 1)\nis more of a hack than anything.\n\nThoughts?\n\n>\n> 0002 looks extremely straightforward, but I wonder if we could get one\n> of the people on this list who knows about sepgsql to have a look?\n> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n\nI welcome any and all feedback.\n\nThanks,\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Mon, 3 Apr 2017 11:38:59 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 03/31/2017 05:23 PM, Robert Haas wrote:\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\n> <mike.palmiotto@crunchydata.com> wrote:\n>> Attached you will find two patches, which were rebased on master as of\n>> 156d388 (applied with `git am --revert [patch file]`). The first gets\n>> rid of some pesky compiler warnings and the second implements the\n>> sepgsql handling of partitioned tables.\n> \n> 0001 has the problem that we have a firm rule against putting any\n> #includes whatsoever before \"postgres.h\". This stdbool issue has come\n> up before, though, and I fear we're going to need to do something\n> about it.\n> \n> 0002 looks extremely straightforward, but I wonder if we could get one\n> of the people on this list who knows about sepgsql to have a look?\n> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n\nWill have a look later today.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Mon, 3 Apr 2017 09:02:57 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\n>> 0002 looks extremely straightforward, but I wonder if we could get one\n>> of the people on this list who knows about sepgsql to have a look?\n>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n>\n> Will have a look later today.\n\nI think it is now tomorrow, unless your time zone is someplace very\nfar away. :-)\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Tue, 4 Apr 2017 09:45:26 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/04/2017 06:45 AM, Robert Haas wrote:\n> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\n>>> 0002 looks extremely straightforward, but I wonder if we could get one\n>>> of the people on this list who knows about sepgsql to have a look?\n>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n>>\n>> Will have a look later today.\n> \n> I think it is now tomorrow, unless your time zone is someplace very\n> far away. :-)\n\nOBE -- I have scheduled time in 30 minutes from now, after I have gotten\nmy first cup of coffee ;-)\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Tue, 4 Apr 2017 07:19:43 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Tue, Apr 4, 2017 at 10:19 AM, Joe Conway <mail@joeconway.com> wrote:\n> On 04/04/2017 06:45 AM, Robert Haas wrote:\n>> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\n>>>> 0002 looks extremely straightforward, but I wonder if we could get one\n>>>> of the people on this list who knows about sepgsql to have a look?\n>>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n>>>\n>>> Will have a look later today.\n>>\n>> I think it is now tomorrow, unless your time zone is someplace very\n>> far away. :-)\n>\n> OBE -- I have scheduled time in 30 minutes from now, after I have gotten\n> my first cup of coffee ;-)\n\nAfter some discussion off-list, I've rebased and udpated the patches.\nPlease see attached for further review.\n\nThanks,\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Tue, 4 Apr 2017 12:55:11 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\n> On Tue, Apr 4, 2017 at 10:19 AM, Joe Conway <mail@joeconway.com> wrote:\n>> On 04/04/2017 06:45 AM, Robert Haas wrote:\n>>> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\n>>>>> 0002 looks extremely straightforward, but I wonder if we could get one\n>>>>> of the people on this list who knows about sepgsql to have a look?\n>>>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\n>>>>\n>>>> Will have a look later today.\n>>>\n>>> I think it is now tomorrow, unless your time zone is someplace very\n>>> far away. :-)\n>>\n>> OBE -- I have scheduled time in 30 minutes from now, after I have gotten\n>> my first cup of coffee ;-)\n> \n> After some discussion off-list, I've rebased and udpated the patches.\n> Please see attached for further review.\n\nThanks -- will have another look and test on a machine with selinux\nsetup. Robert, did you want me to take responsibility to commit on this\nor just provide review/feedback?\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Tue, 4 Apr 2017 10:02:11 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/04/2017 10:02 AM, Joe Conway wrote:\n> On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\n>> After some discussion off-list, I've rebased and udpated the patches.\n>> Please see attached for further review.\n> \n> Thanks -- will have another look and test on a machine with selinux\n> setup. Robert, did you want me to take responsibility to commit on this\n> or just provide review/feedback?\n\nI did some editorializing on these.\n\nIn particular I did not like the approach to fixing \"warning: ‘tclass’\nmay be used uninitialized\" and ended up just doing it the same as was\ndone elsewhere in relation.c already (set tclass = 0 in the variable\ndeclaration). Along the way I also changed one instance of tclass from\nuint16 to uint16_t for the sake of consistency.\n\nInterestingly we figured out that the warning was present with -Og, but\nnot present with -O0, -O2, or -O3.\n\nIf you want to test, apply 0001a and 0001b before 0002.\n\nAny objections?\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Tue, 4 Apr 2017 15:56:28 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\n> On 04/04/2017 10:02 AM, Joe Conway wrote:\n>> On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\n>>> After some discussion off-list, I've rebased and udpated the patches.\n>>> Please see attached for further review.\n>>\n>> Thanks -- will have another look and test on a machine with selinux\n>> setup. Robert, did you want me to take responsibility to commit on this\n>> or just provide review/feedback?\n>\n> I did some editorializing on these.\n>\n> In particular I did not like the approach to fixing \"warning: ‘tclass’\n> may be used uninitialized\" and ended up just doing it the same as was\n> done elsewhere in relation.c already (set tclass = 0 in the variable\n> declaration). Along the way I also changed one instance of tclass from\n> uint16 to uint16_t for the sake of consistency.\n>\n> Interestingly we figured out that the warning was present with -Og, but\n> not present with -O0, -O2, or -O3.\n>\n> If you want to test, apply 0001a and 0001b before 0002.\n>\n> Any objections?\n\nI'm guessing Tom's going to have a strong feeling about whether 0001a\nis the right way to address the stdbool issue, but I don't personally\nknow what the right thing to do is so I will avoid opining on that\ntopic.\n\nSo, nope, no objections here to you committing those.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Tue, 4 Apr 2017 21:29:19 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Robert Haas <robertmhaas@gmail.com> writes:\n> On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\n>> Any objections?\n\n> I'm guessing Tom's going to have a strong feeling about whether 0001a\n> is the right way to address the stdbool issue,\n\nI will? [ looks ... ] Yup, you're right.\n\nI doubt that works at all, TBH. What I'd expect to happen with a\ntypical compiler is a complaint about redefinition of typedef bool,\nbecause c.h already declared it and here this fragment is doing\nso again. It'd make sense to me to do\n\n+ #ifdef bool\n+ #undef bool\n+ #endif\n\nto get rid of the macro definition of bool that stdbool.h is\nsupposed to provide. But there should be no reason to declare\nour typedef a second time.\n\nAnother issue is whether you won't get compiler complaints about\nredefinition of the \"true\" and \"false\" macros. But those would\nlikely only be warnings, not flat-out errors.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 00:58:07 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 4/5/17 00:58, Tom Lane wrote:\n> Another issue is whether you won't get compiler complaints about\n> redefinition of the \"true\" and \"false\" macros. But those would\n> likely only be warnings, not flat-out errors.\n\nThe complaint about bool is also just a warning.\n\n-- \nPeter Eisentraut http://www.2ndQuadrant.com/\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 01:17:01 -0400","msg_from":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\n> On 4/5/17 00:58, Tom Lane wrote:\n>> Another issue is whether you won't get compiler complaints about\n>> redefinition of the \"true\" and \"false\" macros. But those would\n>> likely only be warnings, not flat-out errors.\n\n> The complaint about bool is also just a warning.\n\nReally?\n\n$ cat test.c\ntypedef char bool;\ntypedef char bool;\n$ gcc -c test.c\ntest.c:2: error: redefinition of typedef 'bool'\ntest.c:1: note: previous declaration of 'bool' was here\n\nThis is with gcc 4.4.7.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 01:20:20 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-04-05 00:58:07 -0400, Tom Lane wrote:\n> Robert Haas <robertmhaas@gmail.com> writes:\n> > On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\n> >> Any objections?\n> \n> > I'm guessing Tom's going to have a strong feeling about whether 0001a\n> > is the right way to address the stdbool issue,\n> \n> I will? [ looks ... ] Yup, you're right.\n> \n> I doubt that works at all, TBH. What I'd expect to happen with a\n> typical compiler is a complaint about redefinition of typedef bool,\n> because c.h already declared it and here this fragment is doing\n> so again. It'd make sense to me to do\n> \n> + #ifdef bool\n> + #undef bool\n> + #endif\n> \n> to get rid of the macro definition of bool that stdbool.h is\n> supposed to provide. But there should be no reason to declare\n> our typedef a second time.\n\n> Another issue is whether you won't get compiler complaints about\n> redefinition of the \"true\" and \"false\" macros. But those would\n> likely only be warnings, not flat-out errors.\n\nI argued before that we should migrate to stdbool.h by default, because\nit's only going to get more common. We already do so in a way for c++\ncompilers...\n\n- Andres\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 00:02:49 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> I argued before that we should migrate to stdbool.h by default, because\n> it's only going to get more common. We already do so in a way for c++\n> compilers...\n\nYeah, I was just thinking about that. The core problem though is that\nwe need the \"bool\" fields in the system catalog structs (or anyplace\nelse that it represents an on-disk bool datum) to be understood as\nbeing 1 byte wide. I do not think we can assume that that's true of\nevery compiler's _Bool type. So we'd need some workaround for that.\nThere are probably other places such as isnull arrays where it'd be\nwise to force the width to be 1 byte.\n\nIn any case, that's a research project that's not getting done for v10.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 09:43:53 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Wed, Apr 5, 2017 at 12:58 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\n> Robert Haas <robertmhaas@gmail.com> writes:\n>> On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\n>>> Any objections?\n>\n>> I'm guessing Tom's going to have a strong feeling about whether 0001a\n>> is the right way to address the stdbool issue,\n>\n> I will? [ looks ... ] Yup, you're right.\n\nIt happens occasionally. :-)\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 09:48:20 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\n> Andres Freund <andres@anarazel.de> writes:\n> > I argued before that we should migrate to stdbool.h by default, because\n> > it's only going to get more common. We already do so in a way for c++\n> > compilers...\n> \n> Yeah, I was just thinking about that. The core problem though is that\n> we need the \"bool\" fields in the system catalog structs (or anyplace\n> else that it represents an on-disk bool datum) to be understood as\n> being 1 byte wide. I do not think we can assume that that's true of\n> every compiler's _Bool type. So we'd need some workaround for that.\n> There are probably other places such as isnull arrays where it'd be\n> wise to force the width to be 1 byte.\n\nI wonder if there's any compiler that has _Bool/stdbool.h where it's not\n1 byte sized. It's definitely not guaranteed by the standard.\n\nHaving a seperate booldatum type or such shouldn't be too bad, either\nway.\n\n\n> In any case, that's a research project that's not getting done for v10.\n\nAgreed.\n\n\n- Andres\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 07:28:46 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\n> <mike.palmiotto@crunchydata.com> wrote:\n> > Attached you will find two patches, which were rebased on master as of\n> > 156d388 (applied with `git am --revert [patch file]`). The first gets\n> > rid of some pesky compiler warnings and the second implements the\n> > sepgsql handling of partitioned tables.\n> \n> 0001 has the problem that we have a firm rule against putting any\n> #includes whatsoever before \"postgres.h\". This stdbool issue has come\n> up before, though, and I fear we're going to need to do something\n> about it.\n\nI think in this case it makes sense to deviate from that rule,\ntemporarily and locally.\n\n- Andres\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 07:30:42 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\n>> Yeah, I was just thinking about that. The core problem though is that\n>> we need the \"bool\" fields in the system catalog structs (or anyplace\n>> else that it represents an on-disk bool datum) to be understood as\n>> being 1 byte wide. I do not think we can assume that that's true of\n>> every compiler's _Bool type. So we'd need some workaround for that.\n>> There are probably other places such as isnull arrays where it'd be\n>> wise to force the width to be 1 byte.\n\n> I wonder if there's any compiler that has _Bool/stdbool.h where it's not\n> 1 byte sized. It's definitely not guaranteed by the standard.\n\nHm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\n\nIf it is char-sized almost everywhere, we could create a policy of\nusing stdbool.h unless configure sees that _Bool is not char-sized.\nOTOH, that doesn't improve our existing situation that we have\nplatform-dependent semantics around \"bool\" (eg, what happens when\na non-char-sized value is assigned). It would just change which\none is the majority case, and not in a very safe direction :-(\n\n>> In any case, that's a research project that's not getting done for v10.\n\n> Agreed.\n\nYeah, it's off-topic for right now.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 10:45:06 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\n>> 0001 has the problem that we have a firm rule against putting any\n>> #includes whatsoever before \"postgres.h\". This stdbool issue has come\n>> up before, though, and I fear we're going to need to do something\n>> about it.\n\n> I think in this case it makes sense to deviate from that rule,\n> temporarily and locally.\n\nI'd really rather not. It might be safe here, because this code\nonly works on Linux anyway, but it's still a dangerous precedent.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 10:48:27 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-04-05 10:45:06 -0400, Tom Lane wrote:\n> Andres Freund <andres@anarazel.de> writes:\n> > On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\n> >> Yeah, I was just thinking about that. The core problem though is that\n> >> we need the \"bool\" fields in the system catalog structs (or anyplace\n> >> else that it represents an on-disk bool datum) to be understood as\n> >> being 1 byte wide. I do not think we can assume that that's true of\n> >> every compiler's _Bool type. So we'd need some workaround for that.\n> >> There are probably other places such as isnull arrays where it'd be\n> >> wise to force the width to be 1 byte.\n> \n> > I wonder if there's any compiler that has _Bool/stdbool.h where it's not\n> > 1 byte sized. It's definitely not guaranteed by the standard.\n> \n> Hm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\n\nI think nearly all x86 compilers use 1byte, but I assume there's some\narchitectures where that'd be expensive.\n\n\n> If it is char-sized almost everywhere, we could create a policy of\n> using stdbool.h unless configure sees that _Bool is not char-sized.\n> OTOH, that doesn't improve our existing situation that we have\n> platform-dependent semantics around \"bool\" (eg, what happens when\n> a non-char-sized value is assigned). It would just change which\n> one is the majority case, and not in a very safe direction :-(\n\nYea, no, that doesn't like fun :(.\n\n\nMight make sense to temporarily add a configure test checking for\n_Bool/stdbool size, so we have some idea what we're talking about.\n\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 07:52:00 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-04-05 10:48:27 -0400, Tom Lane wrote:\n> Andres Freund <andres@anarazel.de> writes:\n> > On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\n> >> 0001 has the problem that we have a firm rule against putting any\n> >> #includes whatsoever before \"postgres.h\". This stdbool issue has come\n> >> up before, though, and I fear we're going to need to do something\n> >> about it.\n> \n> > I think in this case it makes sense to deviate from that rule,\n> > temporarily and locally.\n> \n> I'd really rather not. It might be safe here, because this code\n> only works on Linux anyway, but it's still a dangerous precedent.\n\nWell, what's the alternative for v10? There's already precedent btw.,\ncf plperl.h undefining bool.\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 07:53:54 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> On 2017-04-05 10:48:27 -0400, Tom Lane wrote:\n>> I'd really rather not. It might be safe here, because this code\n>> only works on Linux anyway, but it's still a dangerous precedent.\n\n> Well, what's the alternative for v10? There's already precedent btw.,\n> cf plperl.h undefining bool.\n\nWell, the 0001a patch amounted to undefining bool, once you drop\nthe redundant typedef redeclaration. So that's a direction we have\nsome track record with.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 11:06:32 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> On 2017-04-05 10:45:06 -0400, Tom Lane wrote:\n>> Andres Freund <andres@anarazel.de> writes:\n>>> I wonder if there's any compiler that has _Bool/stdbool.h where it's not\n>>> 1 byte sized. It's definitely not guaranteed by the standard.\n\n>> Hm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\n\n> I think nearly all x86 compilers use 1byte, but I assume there's some\n> architectures where that'd be expensive.\n\n[ pokes around... ] prairiedog's gcc (Apple PPC) thinks sizeof(_Bool)\nis 4.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 11:25:46 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/04/2017 09:58 PM, Tom Lane wrote:\n> I doubt that works at all, TBH. What I'd expect to happen with a\n> typical compiler is a complaint about redefinition of typedef bool,\n> because c.h already declared it and here this fragment is doing\n> so again. It'd make sense to me to do\n> \n> + #ifdef bool\n> + #undef bool\n> + #endif\n> \n> to get rid of the macro definition of bool that stdbool.h is\n> supposed to provide. But there should be no reason to declare\n> our typedef a second time.\n\nmakes sense\n\n> Another issue is whether you won't get compiler complaints about\n> redefinition of the \"true\" and \"false\" macros. But those would\n> likely only be warnings, not flat-out errors.\n\nI have not been able to generate warnings or errors around \"true\" and\n\"false\".\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Wed, 5 Apr 2017 08:31:43 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 4/5/17 01:20, Tom Lane wrote:\n>> The complaint about bool is also just a warning.\n> \n> Really?\n> \n> $ cat test.c\n> typedef char bool;\n> typedef char bool;\n> $ gcc -c test.c\n> test.c:2: error: redefinition of typedef 'bool'\n> test.c:1: note: previous declaration of 'bool' was here\n> \n> This is with gcc 4.4.7.\n\nBut the above is not how the current code looks.\n\nstdbool.h does\n\n#define bool _Bool\n\nc.h does\n\n#ifndef bool\ntypedef char bool;\n#endif\n\nSo if you get stdbool.h first, then c.h does nothing. If you get c.h\nfirst, then the macro from stdbool.h will mask the c.h typedef.\n\nWhere this gets really fun is when you include other header files\nbetween, say, postgres.h and selinux/label.h, because then the function\ndefinitions from those header files will be incompatible before and\nafter the stdbool.h inclusion. Which indeed currently leads to the\nfollowing compilation error:\n\nlabel.c: In function �sepgsql_init_client_label�:\nlabel.c:437:18: error: assignment from incompatible pointer type\n[-Werror=incompatible-pointer-types]\n needs_fmgr_hook = sepgsql_needs_fmgr_hook;\n ^\n\nSo I can't reproduce the original complaint, but this is potentially worse.\n\n(The above is on Fedora 25. You need a fairly new selinux.)\n\n-- \nPeter Eisentraut http://www.2ndQuadrant.com/\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 11:56:31 -0400","msg_from":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Joe Conway <mail@joeconway.com> writes:\n> On 04/04/2017 09:58 PM, Tom Lane wrote:\n>> Another issue is whether you won't get compiler complaints about\n>> redefinition of the \"true\" and \"false\" macros. But those would\n>> likely only be warnings, not flat-out errors.\n\n> I have not been able to generate warnings or errors around \"true\" and\n> \"false\".\n\nInteresting. Poking at it on a Fedora 25 machine, I also see a\nbool-type-related warning in sepgsql/label.c, but nothing around macro\nredefinitions. In particular, I find that\n\n#include \"postgres.h\"\n\n#include <stdbool.h>\n\nis completely silent. On the other hand,\n\n#include \"postgres.h\"\n\n#define bool _Bool\n#define true 1\n#define false 0\n\ngenerates the warnings I expected about \"true\" and \"false\" being\nredefined. Which is damn odd, because I copied-and-pasted those\nlines out of\n/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h\n\nConclusion: Fedora's gcc is playing fast and loose somehow with the\ncommand \"#include <stdbool.h>\"; that does not include the file\nyou'd think it does, it does something magic inside the compiler.\nThe magic evidently includes not complaining about duplicate macro\ndefinitions for true and false.\n\nAnyway, I'd recommend that we do something like\n\n #include <selinux/label.h>\n+\n+/*\n+ * <selinux/label.h> includes <stdbool.h>, which creates an incompatible\n+ * #define for bool. Get rid of that so we can use our own typedef.\n+ * (For obscure reasons, the \"true\" and \"false\" macros don't cause issues.)\n+ */\n+ #undef bool\n\nand call it good.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 12:04:00 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"\n\nOn April 5, 2017 9:04:00 AM PDT, Tom Lane <tgl@sss.pgh.pa.us> wrote:\n>Joe Conway <mail@joeconway.com> writes:\n>> On 04/04/2017 09:58 PM, Tom Lane wrote:\n>>> Another issue is whether you won't get compiler complaints about\n>>> redefinition of the \"true\" and \"false\" macros. But those would\n>>> likely only be warnings, not flat-out errors.\n>\n>> I have not been able to generate warnings or errors around \"true\" and\n>> \"false\".\n>\n>Interesting. Poking at it on a Fedora 25 machine, I also see a\n>bool-type-related warning in sepgsql/label.c, but nothing around macro\n>redefinitions. In particular, I find that\n>\n>#include \"postgres.h\"\n>\n>#include <stdbool.h>\n>\n>is completely silent. On the other hand,\n>\n>#include \"postgres.h\"\n>\n>#define bool _Bool\n>#define true 1\n>#define false 0\n>\n>generates the warnings I expected about \"true\" and \"false\" being\n>redefined. Which is damn odd, because I copied-and-pasted those\n>lines out of\n>/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h\n>\n>Conclusion: Fedora's gcc is playing fast and loose somehow with the\n>command \"#include <stdbool.h>\"; that does not include the file\n>you'd think it does, it does something magic inside the compiler.\n>The magic evidently includes not complaining about duplicate macro\n>definitions for true and false.\n>\n>Anyway, I'd recommend that we do something like\n>\n> #include <selinux/label.h>\n>+\n>+/*\n>+ * <selinux/label.h> includes <stdbool.h>, which creates an\n>incompatible\n>+ * #define for bool. Get rid of that so we can use our own typedef.\n>+ * (For obscure reasons, the \"true\" and \"false\" macros don't cause\n>issues.)\n>+ */\n\nGCC generally doesn't warn about macro redefinitions, if both definitions are equivalent. IIRC we have some examples of that in the tree. There's a flag to warn regardless.\n\nAndres\n-- \nSent from my Android device with K-9 Mail. Please excuse my brevity.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 09:06:52 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\n> On 4/5/17 01:20, Tom Lane wrote:\n>> $ cat test.c\n>> typedef char bool;\n>> typedef char bool;\n>> $ gcc -c test.c\n>> test.c:2: error: redefinition of typedef 'bool'\n>> test.c:1: note: previous declaration of 'bool' was here\n\n> But the above is not how the current code looks.\n\nNo, but it is what the proposed 0001a patch would result in.\n\nThe experiments I just did prove that F25's gcc is cheating like mad in\nits implementation of \"#include <stdbool.h>\". I suspect that it's somehow\nturning off the warning that you ought to get about redefining typedef\nbool, as well, though I'm very unclear on how. In any case we don't need\nthe extra typedef and should leave it out.\n\n> So if you get stdbool.h first, then c.h does nothing. If you get c.h\n> first, then the macro from stdbool.h will mask the c.h typedef.\n\nRight, and *either* of those is disastrous if sizeof(_Bool) isn't 1.\n\n> Where this gets really fun is when you include other header files\n> between, say, postgres.h and selinux/label.h, because then the function\n> definitions from those header files will be incompatible before and\n> after the stdbool.h inclusion. Which indeed currently leads to the\n> following compilation error:\n\nThe compilation complaint comes from the fact that we have\n\nstatic bool\nsepgsql_needs_fmgr_hook(Oid functionId)\n\nappearing after the inclusion of <stdbool.h>, causing it to mean\nsomething else than what a Postgres programmer would expect.\nThe other headers aren't at issue really.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 12:14:05 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Andres Freund <andres@anarazel.de> writes:\n> GCC generally doesn't warn about macro redefinitions, if both\n> definitions are equivalent.\n\nBut they're *not* equivalent. c.h has\n\n#define true\t((bool) 1)\n\nwhereas so far as I can see the <stdbool.h> definition is just\n\n#define true\t1\n\nAnd if you put those two definitions together you will get a warning.\nThe lack of one with c.h + <stdbool.h> means somebody is cheating.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 12:26:25 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 4/5/17 12:04, Tom Lane wrote:\n> Conclusion: Fedora's gcc is playing fast and loose somehow with the\n> command \"#include <stdbool.h>\"; that does not include the file\n> you'd think it does, it does something magic inside the compiler.\n> The magic evidently includes not complaining about duplicate macro\n> definitions for true and false.\n\nPerhaps -Wsystem-headers would change the outcome in your case.\n\n-- \nPeter Eisentraut http://www.2ndQuadrant.com/\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 12:53:58 -0400","msg_from":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\n> On 4/5/17 12:04, Tom Lane wrote:\n>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\n>> command \"#include <stdbool.h>\"; that does not include the file\n>> you'd think it does, it does something magic inside the compiler.\n>> The magic evidently includes not complaining about duplicate macro\n>> definitions for true and false.\n\n> Perhaps -Wsystem-headers would change the outcome in your case.\n\nHah, you're right: with that,\n\nIn file included from /usr/include/selinux/label.h:9:0,\n from label.c:40:\n/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \"true\" redefined\n #define true 1\n \nIn file included from ../../src/include/postgres.h:47:0,\n from label.c:11:\n../../src/include/c.h:206:0: note: this is the location of the previous definition\n #define true ((bool) 1)\n\nand likewise for \"false\". So that mystery is explained.\n\nI stand by my previous patch suggestion, except that we can replace\nthe parenthetical comment with something like \"(We don't care if\n<stdbool.h> redefines \"true\"/\"false\"; those are close enough.)\".\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 05 Apr 2017 13:19:10 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Wed, Apr 5, 2017 at 1:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\n> Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\n>> On 4/5/17 12:04, Tom Lane wrote:\n>>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\n>>> command \"#include <stdbool.h>\"; that does not include the file\n>>> you'd think it does, it does something magic inside the compiler.\n>>> The magic evidently includes not complaining about duplicate macro\n>>> definitions for true and false.\n>\n>> Perhaps -Wsystem-headers would change the outcome in your case.\n>\n> Hah, you're right: with that,\n>\n> In file included from /usr/include/selinux/label.h:9:0,\n> from label.c:40:\n> /usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \"true\" redefined\n> #define true 1\n>\n> In file included from ../../src/include/postgres.h:47:0,\n> from label.c:11:\n> ../../src/include/c.h:206:0: note: this is the location of the previous definition\n> #define true ((bool) 1)\n>\n> and likewise for \"false\". So that mystery is explained.\n>\n> I stand by my previous patch suggestion, except that we can replace\n> the parenthetical comment with something like \"(We don't care if\n> <stdbool.h> redefines \"true\"/\"false\"; those are close enough.)\".\n>\n\nSounds good. Updated patchset will include that verbiage, along with\nsome regression tests for partitioned tables.\n\nThanks,\n\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 5 Apr 2017 13:22:22 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Wed, Apr 5, 2017 at 1:22 PM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> On Wed, Apr 5, 2017 at 1:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\n>> Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\n>>> On 4/5/17 12:04, Tom Lane wrote:\n>>>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\n>>>> command \"#include <stdbool.h>\"; that does not include the file\n>>>> you'd think it does, it does something magic inside the compiler.\n>>>> The magic evidently includes not complaining about duplicate macro\n>>>> definitions for true and false.\n>>\n>>> Perhaps -Wsystem-headers would change the outcome in your case.\n>>\n>> Hah, you're right: with that,\n>>\n>> In file included from /usr/include/selinux/label.h:9:0,\n>> from label.c:40:\n>> /usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \"true\" redefined\n>> #define true 1\n>>\n>> In file included from ../../src/include/postgres.h:47:0,\n>> from label.c:11:\n>> ../../src/include/c.h:206:0: note: this is the location of the previous definition\n>> #define true ((bool) 1)\n>>\n>> and likewise for \"false\". So that mystery is explained.\n>>\n>> I stand by my previous patch suggestion, except that we can replace\n>> the parenthetical comment with something like \"(We don't care if\n>> <stdbool.h> redefines \"true\"/\"false\"; those are close enough.)\".\n>>\n>\n> Sounds good. Updated patchset will include that verbiage, along with\n> some regression tests for partitioned tables.\n\nLooks like the label regression test is failing even without these changes.\n\nWeirdly enough, this only happens in one place:\n\n 84 SELECT objtype, objname, label FROM pg_seclabels\n 85 WHERE provider = 'selinux' AND objtype = 'column' AND (objname\nlike 't3.%' OR objname like 't4.%');\n\nI haven't figured out why this may be (yet), but it seems like we\nshouldn't assume the order of output from a SELECT. As I understand\nit, the order of the output is subject to change with changes to the\nplanner/configuration.\n\nI'm going to hold the partition table regression changes for a\nseparate patch and include some ORDER BY fixes. Will post tomorrow\n\nIn the meantime, attached are the latest and greatest patches.\n\nThanks,\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 5 Apr 2017 17:29:10 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/05/2017 02:29 PM, Mike Palmiotto wrote:\n> I'm going to hold the partition table regression changes for a\n> separate patch and include some ORDER BY fixes. Will post tomorrow\n> \n> In the meantime, attached are the latest and greatest patches.\n\nI'm going to push the attached in a few hours unless there is any\nadditional discussion. As stated above we'll do the regression changes\nin a separate patch once that is sorted. I used Tom's approach and\ncomment wording for 0001a.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Thu, 6 Apr 2017 08:08:53 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Joe Conway <mail@joeconway.com> writes:\n> I'm going to push the attached in a few hours unless there is any\n> additional discussion. As stated above we'll do the regression changes\n> in a separate patch once that is sorted. I used Tom's approach and\n> comment wording for 0001a.\n\nLooks generally sane, but I noticed a grammatical nitpick:\n\n-\t * Only attributes within regular relation or partition relations have\n+\t * Only attributes within regular relations or partition relations have\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 06 Apr 2017 11:29:58 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/06/2017 08:29 AM, Tom Lane wrote:\n> Joe Conway <mail@joeconway.com> writes:\n>> I'm going to push the attached in a few hours unless there is any\n>> additional discussion. As stated above we'll do the regression changes\n>> in a separate patch once that is sorted. I used Tom's approach and\n>> comment wording for 0001a.\n> \n> Looks generally sane, but I noticed a grammatical nitpick:\n> \n> -\t * Only attributes within regular relation or partition relations have\n> +\t * Only attributes within regular relations or partition relations have\n\nGood call -- thanks!\n\nAny thoughts on whether 0001a and 0001b ought to be backpatched? I'm\nthinking not given the lack of past complaints but it might make sense\nto do.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Thu, 6 Apr 2017 11:19:05 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Joe Conway <mail@joeconway.com> writes:\n> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\n> thinking not given the lack of past complaints but it might make sense\n> to do.\n\nI think 0001a absolutely needs to be, because it is fixing what is really\nan ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\nof bool, but as things stand it's returning _Bool (which is why the\ncompiler is complaining). Now we might get away with that on most\nhardware, but on platforms where those are different widths, it's possible\nto imagine function-return conventions that would make it fail.\n\n0001b seems to only be needed for compilers that aren't smart enough\nto see that tclass won't be referenced for RELKIND_INDEX, so it's\njust cosmetic.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 06 Apr 2017 15:35:40 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/06/2017 12:35 PM, Tom Lane wrote:\n> Joe Conway <mail@joeconway.com> writes:\n>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\n>> thinking not given the lack of past complaints but it might make sense\n>> to do.\n> \n> I think 0001a absolutely needs to be, because it is fixing what is really\n> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\n> of bool, but as things stand it's returning _Bool (which is why the\n> compiler is complaining). Now we might get away with that on most\n> hardware, but on platforms where those are different widths, it's possible\n> to imagine function-return conventions that would make it fail.\n> \n> 0001b seems to only be needed for compilers that aren't smart enough\n> to see that tclass won't be referenced for RELKIND_INDEX, so it's\n> just cosmetic.\n\nOk, committed/pushed that way.\n\nI found some missing bits in the 0002 patch -- new version attached.\nWill wait on new regression tests before committing, but I expect we'll\nhave those by end of today and be able to commit the rest tomorrow.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Thu, 6 Apr 2017 14:52:24 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Thu, Apr 6, 2017 at 5:52 PM, Joe Conway <mail@joeconway.com> wrote:\n> On 04/06/2017 12:35 PM, Tom Lane wrote:\n>> Joe Conway <mail@joeconway.com> writes:\n>>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\n>>> thinking not given the lack of past complaints but it might make sense\n>>> to do.\n>>\n>> I think 0001a absolutely needs to be, because it is fixing what is really\n>> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\n>> of bool, but as things stand it's returning _Bool (which is why the\n>> compiler is complaining). Now we might get away with that on most\n>> hardware, but on platforms where those are different widths, it's possible\n>> to imagine function-return conventions that would make it fail.\n>>\n>> 0001b seems to only be needed for compilers that aren't smart enough\n>> to see that tclass won't be referenced for RELKIND_INDEX, so it's\n>> just cosmetic.\n>\n> Ok, committed/pushed that way.\n>\n> I found some missing bits in the 0002 patch -- new version attached.\n> Will wait on new regression tests before committing, but I expect we'll\n> have those by end of today and be able to commit the rest tomorrow.\n\nAttached are the regression test updates for partitioned tables.\n\nThanks,\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 7 Apr 2017 14:36:29 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Fri, Apr 7, 2017 at 2:36 PM, Mike Palmiotto\n<mike.palmiotto@crunchydata.com> wrote:\n> On Thu, Apr 6, 2017 at 5:52 PM, Joe Conway <mail@joeconway.com> wrote:\n>> On 04/06/2017 12:35 PM, Tom Lane wrote:\n>>> Joe Conway <mail@joeconway.com> writes:\n>>>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\n>>>> thinking not given the lack of past complaints but it might make sense\n>>>> to do.\n>>>\n>>> I think 0001a absolutely needs to be, because it is fixing what is really\n>>> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\n>>> of bool, but as things stand it's returning _Bool (which is why the\n>>> compiler is complaining). Now we might get away with that on most\n>>> hardware, but on platforms where those are different widths, it's possible\n>>> to imagine function-return conventions that would make it fail.\n>>>\n>>> 0001b seems to only be needed for compilers that aren't smart enough\n>>> to see that tclass won't be referenced for RELKIND_INDEX, so it's\n>>> just cosmetic.\n>>\n>> Ok, committed/pushed that way.\n>>\n>> I found some missing bits in the 0002 patch -- new version attached.\n>> Will wait on new regression tests before committing, but I expect we'll\n>> have those by end of today and be able to commit the rest tomorrow.\n>\n> Attached are the regression test updates for partitioned tables.\n\nActually attached this time.\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Fri, 7 Apr 2017 14:37:06 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/07/2017 11:37 AM, Mike Palmiotto wrote:\n>>> I found some missing bits in the 0002 patch -- new version attached.\n>>> Will wait on new regression tests before committing, but I expect we'll\n>>> have those by end of today and be able to commit the rest tomorrow.\n>>\n>> Attached are the regression test updates for partitioned tables.\n> \n> Actually attached this time.\n\nBased on my review and testing of the 0002 patch I believe it is\ncorrect. However Mike and I just went through the regression test patch\nline by line and there are issues he needs to address -- there is no way\nthat is happening by tonight as the output is very verbose and we need\nto be sure we are both testing the correct things and getting the\ncorrect behaviors.\n\nBased on that I can:\n\n1) commit the 0002 patch now before the feature freeze and follow up\n with the regression test patch when ready in a couple of days\n2) hold off on both patches until ready\n3) push both patches to the next commitfest/pg11\n\nSome argue this is an open issue against the new partitioning feature in\npg10 and therefore should be addressed now, and others do not. I can see\nboth sides of that argument.\n\nIn any case, thoughts on what to do?\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Fri, 7 Apr 2017 14:22:19 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n> On 04/07/2017 11:37 AM, Mike Palmiotto wrote:\n>>>> I found some missing bits in the 0002 patch -- new version attached.\n>>>> Will wait on new regression tests before committing, but I expect we'll\n>>>> have those by end of today and be able to commit the rest tomorrow.\n>>>\n>>> Attached are the regression test updates for partitioned tables.\n>>\n>> Actually attached this time.\n>\n> Based on my review and testing of the 0002 patch I believe it is\n> correct. However Mike and I just went through the regression test patch\n> line by line and there are issues he needs to address -- there is no way\n> that is happening by tonight as the output is very verbose and we need\n> to be sure we are both testing the correct things and getting the\n> correct behaviors.\n>\n> Based on that I can:\n>\n> 1) commit the 0002 patch now before the feature freeze and follow up\n> with the regression test patch when ready in a couple of days\n> 2) hold off on both patches until ready\n> 3) push both patches to the next commitfest/pg11\n>\n> Some argue this is an open issue against the new partitioning feature in\n> pg10 and therefore should be addressed now, and others do not. I can see\n> both sides of that argument.\n>\n> In any case, thoughts on what to do?\n\nSpeaking only for myself, I'm OK with any of those options, provided\nthat that \"a couple\" means what my dictionary says it means.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 7 Apr 2017 20:36:51 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/07/2017 05:36 PM, Robert Haas wrote:\n> On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n>> 1) commit the 0002 patch now before the feature freeze and follow up\n>> with the regression test patch when ready in a couple of days\n>> 2) hold off on both patches until ready\n>> 3) push both patches to the next commitfest/pg11\n>>\n>> Some argue this is an open issue against the new partitioning feature in\n>> pg10 and therefore should be addressed now, and others do not. I can see\n>> both sides of that argument.\n>>\n>> In any case, thoughts on what to do?\n> \n> Speaking only for myself, I'm OK with any of those options, provided\n> that that \"a couple\" means what my dictionary says it means.\n\nThanks. I'd prefer not to do #1 actually, and I think we can adhere to\nthe dictionary meaning of \"a couple\" (i.e. by EOD Sunday). Assuming we\nare ready by Sunday I will push both together (#2) or else I will move\nthem both together to the next CF (#3).\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Fri, 7 Apr 2017 19:36:49 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Joe,\n\n* Joe Conway (mail@joeconway.com) wrote:\n> On 04/07/2017 05:36 PM, Robert Haas wrote:\n> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n> >> 1) commit the 0002 patch now before the feature freeze and follow up\n> >> with the regression test patch when ready in a couple of days\n> >> 2) hold off on both patches until ready\n> >> 3) push both patches to the next commitfest/pg11\n> >>\n> >> Some argue this is an open issue against the new partitioning feature in\n> >> pg10 and therefore should be addressed now, and others do not. I can see\n> >> both sides of that argument.\n> >>\n> >> In any case, thoughts on what to do?\n> > \n> > Speaking only for myself, I'm OK with any of those options, provided\n> > that that \"a couple\" means what my dictionary says it means.\n> \n> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\n> the dictionary meaning of \"a couple\" (i.e. by EOD Sunday). Assuming we\n> are ready by Sunday I will push both together (#2) or else I will move\n> them both together to the next CF (#3).\n\nSounds good to me.\n\nThanks!\n\nStephen","msg_date":"Sat, 8 Apr 2017 10:29:21 -0400","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On Sat, Apr 8, 2017 at 10:29 AM, Stephen Frost <sfrost@snowman.net> wrote:\n> Joe,\n>\n> * Joe Conway (mail@joeconway.com) wrote:\n>> On 04/07/2017 05:36 PM, Robert Haas wrote:\n>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n>> >> 1) commit the 0002 patch now before the feature freeze and follow up\n>> >> with the regression test patch when ready in a couple of days\n>> >> 2) hold off on both patches until ready\n>> >> 3) push both patches to the next commitfest/pg11\n>> >>\n>> >> Some argue this is an open issue against the new partitioning feature in\n>> >> pg10 and therefore should be addressed now, and others do not. I can see\n>> >> both sides of that argument.\n>> >>\n>> >> In any case, thoughts on what to do?\n>> >\n>> > Speaking only for myself, I'm OK with any of those options, provided\n>> > that that \"a couple\" means what my dictionary says it means.\n>>\n>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\n>> the dictionary meaning of \"a couple\" (i.e. by EOD Sunday). Assuming we\n>> are ready by Sunday I will push both together (#2) or else I will move\n>> them both together to the next CF (#3).\n\nWithout further ado, the updated patch is attached.\n\nI added some comment updates, making it clearer where partitioned\ntables are used and modified the test-cases to be as close to\nlock-step with regular table regression testing as possible.\n\nLet me know if you find any further issues -- I made sure to review\npretty thoroughly, so I'd be surprised, but it happens.\n\nThanks,\n-- \nMike Palmiotto\nSoftware Engineer\nCrunchy Data Solutions\nhttps://crunchydata.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Sun, 9 Apr 2017 13:18:11 -0400","msg_from":"Mike Palmiotto <mike.palmiotto@crunchydata.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/08/2017 07:29 AM, Stephen Frost wrote:\n> * Joe Conway (mail@joeconway.com) wrote:\n>> On 04/07/2017 05:36 PM, Robert Haas wrote:\n>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n>> >> 1) commit the 0002 patch now before the feature freeze and follow up\n>> >> with the regression test patch when ready in a couple of days\n>> >> 2) hold off on both patches until ready\n>> >> 3) push both patches to the next commitfest/pg11\n>> >>\n>> >> Some argue this is an open issue against the new partitioning feature in\n>> >> pg10 and therefore should be addressed now, and others do not. I can see\n>> >> both sides of that argument.\n>> >>\n>> >> In any case, thoughts on what to do?\n>> > \n>> > Speaking only for myself, I'm OK with any of those options, provided\n>> > that that \"a couple\" means what my dictionary says it means.\n>> \n>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\n>> the dictionary meaning of \"a couple\" (i.e. by EOD Sunday). Assuming we\n>> are ready by Sunday I will push both together (#2) or else I will move\n>> them both together to the next CF (#3).\n> \n> Sounds good to me.\n\nHaving heard no objections, committed and pushed as per option #2.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sun, 9 Apr 2017 14:04:45 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/09/2017 02:04 PM, Joe Conway wrote:\n> On 04/08/2017 07:29 AM, Stephen Frost wrote:\n>> * Joe Conway (mail@joeconway.com) wrote:\n>>> On 04/07/2017 05:36 PM, Robert Haas wrote:\n>>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\n>>> >> 1) commit the 0002 patch now before the feature freeze and follow up\n>>> >> with the regression test patch when ready in a couple of days\n>>> >> 2) hold off on both patches until ready\n>>> >> 3) push both patches to the next commitfest/pg11\n>>> >>\n>>> >> Some argue this is an open issue against the new partitioning feature in\n>>> >> pg10 and therefore should be addressed now, and others do not. I can see\n>>> >> both sides of that argument.\n>>> >>\n>>> >> In any case, thoughts on what to do?\n>>> > \n>>> > Speaking only for myself, I'm OK with any of those options, provided\n>>> > that that \"a couple\" means what my dictionary says it means.\n>>> \n>>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\n>>> the dictionary meaning of \"a couple\" (i.e. by EOD Sunday). Assuming we\n>>> are ready by Sunday I will push both together (#2) or else I will move\n>>> them both together to the next CF (#3).\n>> \n>> Sounds good to me.\n> \n> Having heard no objections, committed and pushed as per option #2.\n\nInteresting -- rhino is now failing. I tested a minute ago manually on\nthe same buildfarm animal and it passed. I'm on it.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sun, 9 Apr 2017 14:28:48 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\n> Interesting -- rhino is now failing. I tested a minute ago manually on\n> the same buildfarm animal and it passed. I'm on it.\n\nThe module for segpsql really needs to be improved so it logs\nregression.diffs - iirc several other modules do.\n\nGreetings,\n\nAndres Freund\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Sun, 9 Apr 2017 14:49:45 -0700","msg_from":"Andres Freund <andres@anarazel.de>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/09/2017 02:49 PM, Andres Freund wrote:\n> On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\n>> Interesting -- rhino is now failing. I tested a minute ago manually on\n>> the same buildfarm animal and it passed. I'm on it.\n> \n> The module for segpsql really needs to be improved so it logs\n> regression.diffs - iirc several other modules do.\n\nSure, but for another day I think.\n\nI turned on the buildfarm \"keep\" setting and looked at the diffs. The\nissue is that in there are a few places that do \"SELECT ... FROM\npg_seclabels ... ORDER BY ...\" and when manually testing I get default\ndatabase encoding \"UTF8\" but with the buildfarm I get \"SQL_ASCII\", hence\na different sorting.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sun, 9 Apr 2017 15:01:42 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/09/2017 03:01 PM, Joe Conway wrote:\n> On 04/09/2017 02:49 PM, Andres Freund wrote:\n>> On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\n>>> Interesting -- rhino is now failing. I tested a minute ago manually on\n>>> the same buildfarm animal and it passed. I'm on it.\n>> \n>> The module for segpsql really needs to be improved so it logs\n>> regression.diffs - iirc several other modules do.\n> \n> Sure, but for another day I think.\n> \n> I turned on the buildfarm \"keep\" setting and looked at the diffs. The\n> issue is that in there are a few places that do \"SELECT ... FROM\n> pg_seclabels ... ORDER BY ...\" and when manually testing I get default\n> database encoding \"UTF8\" but with the buildfarm I get \"SQL_ASCII\", hence\n> a different sorting.\n\nLooks like adding COLLATE \"C\" in the ORDER BY clauses does the trick. Is\nthat the preferred method for this kind of issue?\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sun, 9 Apr 2017 15:45:14 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"Joe Conway <mail@joeconway.com> writes:\n>> I turned on the buildfarm \"keep\" setting and looked at the diffs. The\n>> issue is that in there are a few places that do \"SELECT ... FROM\n>> pg_seclabels ... ORDER BY ...\" and when manually testing I get default\n>> database encoding \"UTF8\" but with the buildfarm I get \"SQL_ASCII\", hence\n>> a different sorting.\n\n> Looks like adding COLLATE \"C\" in the ORDER BY clauses does the trick. Is\n> that the preferred method for this kind of issue?\n\nYeah, that's pretty much the standard fix nowadays.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Sun, 09 Apr 2017 19:14:41 -0400","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"},{"msg_contents":"On 04/09/2017 04:14 PM, Tom Lane wrote:\n> Joe Conway <mail@joeconway.com> writes:\n>>> I turned on the buildfarm \"keep\" setting and looked at the diffs. The\n>>> issue is that in there are a few places that do \"SELECT ... FROM\n>>> pg_seclabels ... ORDER BY ...\" and when manually testing I get default\n>>> database encoding \"UTF8\" but with the buildfarm I get \"SQL_ASCII\", hence\n>>> a different sorting.\n> \n>> Looks like adding COLLATE \"C\" in the ORDER BY clauses does the trick. Is\n>> that the preferred method for this kind of issue?\n> \n> Yeah, that's pretty much the standard fix nowadays.\n\n\nGood to hear, thanks -- rhino is back to green.\n\nJoe\n\n-- \nCrunchy Data - http://crunchydata.com\nPostgreSQL Support for Secure Enterprises\nConsulting, Training, & Open Source Development","msg_date":"Sun, 9 Apr 2017 16:39:55 -0700","msg_from":"Joe Conway <mail@joeconway.com>","msg_from_op":false,"msg_subject":"Re: partitioned tables and contrib/sepgsql"}],"string":"[\n {\n \"msg_contents\": \"Greetings,\\n\\nWhile going over the contrib modules, I noticed that sepgsql was not\\nupdated for partitioned tables. What that appears to mean is that it's\\nnot possible to define labels on partitioned tables. As I recall,\\naccessing the parent of a table will, similar to the GRANT system, not\\nperform checkes against the child tables, meaning that there's no way to\\nhave SELinux checks properly enforced when partitioned tables are being\\nused.\\n\\nThis is an issue which should be resolved for PG10, so I'll add it to\\nthe open items list.\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 09:47:18 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n> Greetings,\\n>\\n> While going over the contrib modules, I noticed that sepgsql was not\\n> updated for partitioned tables. What that appears to mean is that it's\\n> not possible to define labels on partitioned tables. As I recall,\\n> accessing the parent of a table will, similar to the GRANT system, not\\n> perform checkes against the child tables, meaning that there's no way to\\n> have SELinux checks properly enforced when partitioned tables are being\\n> used.\\n\\nI'll start taking a look at this. Presumably we'd just extend existing\\nobject_access_hooks to cover partitioned tables?\\n\\n>\\n> This is an issue which should be resolved for PG10, so I'll add it to\\n> the open items list.\\n\\nI'll grab it. Thanks.\\n\\n--Mike\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 09:59:14 -0500\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Mike,\\n\\n* Mike Palmiotto (mike.palmiotto@crunchydata.com) wrote:\\n> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n> > While going over the contrib modules, I noticed that sepgsql was not\\n> > updated for partitioned tables. What that appears to mean is that it's\\n> > not possible to define labels on partitioned tables. As I recall,\\n> > accessing the parent of a table will, similar to the GRANT system, not\\n> > perform checkes against the child tables, meaning that there's no way to\\n> > have SELinux checks properly enforced when partitioned tables are being\\n> > used.\\n> \\n> I'll start taking a look at this. Presumably we'd just extend existing\\n> object_access_hooks to cover partitioned tables?\\n\\nAt least on first blush that seems like the right approach. We'll need\\nto make sure that the SECURITY LABEL system will properly work with\\npartitioned tables too, of course, and that the checks are called when a\\nuser queries a partitioned table. Then we'll need regression tests to\\nmake sure we get it all correct and don't screw it up in the future. ;)\\n\\n> > This is an issue which should be resolved for PG10, so I'll add it to\\n> > the open items list.\\n> \\n> I'll grab it. Thanks.\\n\\nExcellent, thanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 10:05:00 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n> While going over the contrib modules, I noticed that sepgsql was not\\n> updated for partitioned tables. What that appears to mean is that it's\\n> not possible to define labels on partitioned tables.\\n\\nIt works for me:\\n\\nrhaas=# load 'dummy_seclabel';\\nLOAD\\nrhaas=# create table foo (a int, b text) partition by range (a);\\nCREATE TABLE\\nrhaas=# security label on table foo is 'classified';\\nSECURITY LABEL\\n\\nWhat exactly is the problem you're seeing?\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 27 Mar 2017 10:47:31 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n>> While going over the contrib modules, I noticed that sepgsql was not\\n>> updated for partitioned tables. What that appears to mean is that it's\\n>> not possible to define labels on partitioned tables.\\n>\\n> It works for me:\\n>\\n> rhaas=# load 'dummy_seclabel';\\n> LOAD\\n> rhaas=# create table foo (a int, b text) partition by range (a);\\n> CREATE TABLE\\n> rhaas=# security label on table foo is 'classified';\\n> SECURITY LABEL\\n>\\n> What exactly is the problem you're seeing?\\n\\nIIRC the initial concern was that contrib/sepgsql was not updated for\\nRELKIND_PARTITIONED_TABLE, so the post-create hook would not work\\nproperly for partitioned tables.\\n\\nI've looked into it a bit and saw a post-alter hook in\\nStoreCatalogInheritance1. It seems like it may just be an issue of\\nadding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\\n\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 27 Mar 2017 11:22:36 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Mon, Mar 27, 2017 at 11:22 AM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n>>> While going over the contrib modules, I noticed that sepgsql was not\\n>>> updated for partitioned tables. What that appears to mean is that it's\\n>>> not possible to define labels on partitioned tables.\\n>>\\n>> It works for me:\\n>>\\n>> rhaas=# load 'dummy_seclabel';\\n>> LOAD\\n>> rhaas=# create table foo (a int, b text) partition by range (a);\\n>> CREATE TABLE\\n>> rhaas=# security label on table foo is 'classified';\\n>> SECURITY LABEL\\n>>\\n>> What exactly is the problem you're seeing?\\n>\\n> IIRC the initial concern was that contrib/sepgsql was not updated for\\n> RELKIND_PARTITIONED_TABLE, so the post-create hook would not work\\n> properly for partitioned tables.\\n>\\n> I've looked into it a bit and saw a post-alter hook in\\n> StoreCatalogInheritance1. It seems like it may just be an issue of\\n> adding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\\n\\nI thought that kind of thing might be the issue, but it didn't seem to\\nmatch Stephen's description. Adding RELKIND_PARTITIONED_TABLE to that\\nfunction seems like it would probably be sufficient to make that hook\\nwork, although that would need to be tested, but there are numerous\\nother references to RELKIND_RELATION in contrib/sepgsql, some of which\\nprobably also need to be updated to consider\\nRELKIND_PARTITIONED_TABLE.\\n\\nIn my view, this is really an enhancement to sepgsql to handle a new\\nPostgreSQL feature rather than a defect in the partitioning commit, so\\nI don't think it falls on Amit Langote (as the author) or me (as the\\ncommitter) to fix. There might similarly be updates to sepgsql to do\\nsomething with permissions on logical replication's new publication\\nand subscription objects, but we should similarly regard those as\\npossible new features, not things that Petr or Peter need to work on.\\n Note that sepgsql hasn't been updated to work with RLS yet, either,\\nbut we didn't regard that as an open item for RLS, or if we did the\\nresolution was just to document it. I am not opposed to giving a\\nlittle more time to get this straightened out, but if a patch doesn't\\nshow up fairly soon then I think we should just document that sepgsql\\ndoesn't support partitioned tables in v10. sepgsql has a fairly\\nlengthy list of implementation restrictions already, so one more is\\nnot going to kill anybody -- or if it will then that person should\\nproduce a patch soon.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 27 Mar 2017 11:46:42 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Mon, Mar 27, 2017 at 11:22 AM, Mike Palmiotto\\n> <mike.palmiotto@crunchydata.com> wrote:\\n>> On Mon, Mar 27, 2017 at 10:47 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>>> On Thu, Mar 9, 2017 at 9:47 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n>>>> While going over the contrib modules, I noticed that sepgsql was not\\n>>>> updated for partitioned tables. What that appears to mean is that it's\\n>>>> not possible to define labels on partitioned tables.\\n>>>\\n>>> It works for me:\\n>>>\\n>>> rhaas=# load 'dummy_seclabel';\\n>>> LOAD\\n>>> rhaas=# create table foo (a int, b text) partition by range (a);\\n>>> CREATE TABLE\\n>>> rhaas=# security label on table foo is 'classified';\\n>>> SECURITY LABEL\\n>>>\\n>>> What exactly is the problem you're seeing?\\n>>\\n>> IIRC the initial concern was that contrib/sepgsql was not updated for\\n>> RELKIND_PARTITIONED_TABLE, so the post-create hook would not work\\n>> properly for partitioned tables.\\n>>\\n>> I've looked into it a bit and saw a post-alter hook in\\n>> StoreCatalogInheritance1. It seems like it may just be an issue of\\n>> adding the RELKIND_PARTITIONED_TABLE to sepgsql_relation_post_create.\\n>\\n> I thought that kind of thing might be the issue, but it didn't seem to\\n> match Stephen's description. Adding RELKIND_PARTITIONED_TABLE to that\\n> function seems like it would probably be sufficient to make that hook\\n> work, although that would need to be tested, but there are numerous\\n> other references to RELKIND_RELATION in contrib/sepgsql, some of which\\n> probably also need to be updated to consider\\n> RELKIND_PARTITIONED_TABLE.\\n\\nAgreed.\\n\\n>\\n> In my view, this is really an enhancement to sepgsql to handle a new\\n> PostgreSQL feature rather than a defect in the partitioning commit, so\\n> I don't think it falls on Amit Langote (as the author) or me (as the\\n> committer) to fix. There might similarly be updates to sepgsql to do\\n> something with permissions on logical replication's new publication\\n> and subscription objects, but we should similarly regard those as\\n> possible new features, not things that Petr or Peter need to work on.\\n\\nI'll keep these items in mind for future reference. Thanks.\\n\\n> Note that sepgsql hasn't been updated to work with RLS yet, either,\\n> but we didn't regard that as an open item for RLS, or if we did the\\n> resolution was just to document it. I am not opposed to giving a\\n> little more time to get this straightened out, but if a patch doesn't\\n> show up fairly soon then I think we should just document that sepgsql\\n> doesn't support partitioned tables in v10. sepgsql has a fairly\\n> lengthy list of implementation restrictions already, so one more is\\n> not going to kill anybody -- or if it will then that person should\\n> produce a patch soon.\\n\\nOkay, I'll make sure I get something fleshed out today or tomorrow.\\n\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 27 Mar 2017 12:09:08 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Mon, Mar 27, 2017 at 12:09 PM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>> <snip>\\n>> Note that sepgsql hasn't been updated to work with RLS yet, either,\\n>> but we didn't regard that as an open item for RLS, or if we did the\\n>> resolution was just to document it. I am not opposed to giving a\\n>> little more time to get this straightened out, but if a patch doesn't\\n>> show up fairly soon then I think we should just document that sepgsql\\n>> doesn't support partitioned tables in v10. sepgsql has a fairly\\n>> lengthy list of implementation restrictions already, so one more is\\n>> not going to kill anybody -- or if it will then that person should\\n>> produce a patch soon.\\n>\\n> Okay, I'll make sure I get something fleshed out today or tomorrow.\\n\\nApologies for the delay. I was waffling over whether to reference\\nPartitionedRelationId in sepgsql, but ended up deciding to just handle\\nRELKIND_PARTITIONED_TABLE and treat the classOid as\\nRelationRelationId. Seeing as there is a relid in pg_class which\\ncorresponds to the partitioned table, this chosen route seemed\\nacceptable.\\n\\nHere is a demonstration of the partitioned table working with sepgsql hooks:\\nhttps://gist.github.com/anonymous/b10f476a95ae9cdd39b83ef872d4b1e6\\n\\nAttached you will find two patches, which were rebased on master as of\\n156d388 (applied with `git am --revert [patch file]`). The first gets\\nrid of some pesky compiler warnings and the second implements the\\nsepgsql handling of partitioned tables.\\n\\nThanks,\\n\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 31 Mar 2017 14:14:59 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> On Mon, Mar 27, 2017 at 12:09 PM, Mike Palmiotto\\n> <mike.palmiotto@crunchydata.com> wrote:\\n>> On Mon, Mar 27, 2017 at 11:46 AM, Robert Haas <robertmhaas@gmail.com> wrote:\\n>>> <snip>\\n>>> Note that sepgsql hasn't been updated to work with RLS yet, either,\\n>>> but we didn't regard that as an open item for RLS, or if we did the\\n>>> resolution was just to document it. I am not opposed to giving a\\n>>> little more time to get this straightened out, but if a patch doesn't\\n>>> show up fairly soon then I think we should just document that sepgsql\\n>>> doesn't support partitioned tables in v10. sepgsql has a fairly\\n>>> lengthy list of implementation restrictions already, so one more is\\n>>> not going to kill anybody -- or if it will then that person should\\n>>> produce a patch soon.\\n>>\\n>> Okay, I'll make sure I get something fleshed out today or tomorrow.\\n>\\n> Apologies for the delay. I was waffling over whether to reference\\n> PartitionedRelationId in sepgsql, but ended up deciding to just handle\\n> RELKIND_PARTITIONED_TABLE and treat the classOid as\\n> RelationRelationId. Seeing as there is a relid in pg_class which\\n> corresponds to the partitioned table, this chosen route seemed\\n> acceptable.\\n\\nNewest patches remove cruft from said waffling. No need to include\\npg_partitioned_table.h if we're not referencing PartitionedRelationId.\\n\\n>\\n> Here is a demonstration of the partitioned table working with sepgsql hooks:\\n> https://gist.github.com/anonymous/b10f476a95ae9cdd39b83ef872d4b1e6\\n>\\n> Attached you will find two patches, which were rebased on master as of\\n> 156d388 (applied with `git am --revert [patch file]`). The first gets\\n> rid of some pesky compiler warnings and the second implements the\\n> sepgsql handling of partitioned tables.\\n\\nThat should have read `git am --reject [patch file]`. Apologies for\\nthe inaccuracy.\\n\\nThanks,\\n\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 31 Mar 2017 14:42:00 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> Attached you will find two patches, which were rebased on master as of\\n> 156d388 (applied with `git am --revert [patch file]`). The first gets\\n> rid of some pesky compiler warnings and the second implements the\\n> sepgsql handling of partitioned tables.\\n\\n0001 has the problem that we have a firm rule against putting any\\n#includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\nup before, though, and I fear we're going to need to do something\\nabout it.\\n\\n0002 looks extremely straightforward, but I wonder if we could get one\\nof the people on this list who knows about sepgsql to have a look?\\n(Stephen Frost, Joe Conway, KaiGai Kohei...)\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 31 Mar 2017 20:23:39 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Fri, Mar 31, 2017 at 8:23 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\\n> <mike.palmiotto@crunchydata.com> wrote:\\n>> Attached you will find two patches, which were rebased on master as of\\n>> 156d388 (applied with `git am --revert [patch file]`). The first gets\\n>> rid of some pesky compiler warnings and the second implements the\\n>> sepgsql handling of partitioned tables.\\n>\\n> 0001 has the problem that we have a firm rule against putting any\\n> #includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\n> up before, though, and I fear we're going to need to do something\\n> about it.\\n\\nYeah, I recall this rule. The only things I can really think of to\\nsolve the problem are:\\n1) #define _STDBOOL_H in postgres's c.h once bool and the like have\\nbeen defined, so we can avoid re-definition.\\n2) Enforce that any code utilizing the stdbool header manage the\\nre-definition with some combination of #undef/#define/#typedef and\\ndocument the issue somewhere.\\n\\nI'd be more inclined to believe 2) is the correct solution, since 1)\\nis more of a hack than anything.\\n\\nThoughts?\\n\\n>\\n> 0002 looks extremely straightforward, but I wonder if we could get one\\n> of the people on this list who knows about sepgsql to have a look?\\n> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n\\nI welcome any and all feedback.\\n\\nThanks,\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Mon, 3 Apr 2017 11:38:59 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 03/31/2017 05:23 PM, Robert Haas wrote:\\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\\n> <mike.palmiotto@crunchydata.com> wrote:\\n>> Attached you will find two patches, which were rebased on master as of\\n>> 156d388 (applied with `git am --revert [patch file]`). The first gets\\n>> rid of some pesky compiler warnings and the second implements the\\n>> sepgsql handling of partitioned tables.\\n> \\n> 0001 has the problem that we have a firm rule against putting any\\n> #includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\n> up before, though, and I fear we're going to need to do something\\n> about it.\\n> \\n> 0002 looks extremely straightforward, but I wonder if we could get one\\n> of the people on this list who knows about sepgsql to have a look?\\n> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n\\nWill have a look later today.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Mon, 3 Apr 2017 09:02:57 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> 0002 looks extremely straightforward, but I wonder if we could get one\\n>> of the people on this list who knows about sepgsql to have a look?\\n>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n>\\n> Will have a look later today.\\n\\nI think it is now tomorrow, unless your time zone is someplace very\\nfar away. :-)\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Tue, 4 Apr 2017 09:45:26 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/04/2017 06:45 AM, Robert Haas wrote:\\n> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\\n>>> 0002 looks extremely straightforward, but I wonder if we could get one\\n>>> of the people on this list who knows about sepgsql to have a look?\\n>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n>>\\n>> Will have a look later today.\\n> \\n> I think it is now tomorrow, unless your time zone is someplace very\\n> far away. :-)\\n\\nOBE -- I have scheduled time in 30 minutes from now, after I have gotten\\nmy first cup of coffee ;-)\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Tue, 4 Apr 2017 07:19:43 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Tue, Apr 4, 2017 at 10:19 AM, Joe Conway <mail@joeconway.com> wrote:\\n> On 04/04/2017 06:45 AM, Robert Haas wrote:\\n>> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\\n>>>> 0002 looks extremely straightforward, but I wonder if we could get one\\n>>>> of the people on this list who knows about sepgsql to have a look?\\n>>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n>>>\\n>>> Will have a look later today.\\n>>\\n>> I think it is now tomorrow, unless your time zone is someplace very\\n>> far away. :-)\\n>\\n> OBE -- I have scheduled time in 30 minutes from now, after I have gotten\\n> my first cup of coffee ;-)\\n\\nAfter some discussion off-list, I've rebased and udpated the patches.\\nPlease see attached for further review.\\n\\nThanks,\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Tue, 4 Apr 2017 12:55:11 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\\n> On Tue, Apr 4, 2017 at 10:19 AM, Joe Conway <mail@joeconway.com> wrote:\\n>> On 04/04/2017 06:45 AM, Robert Haas wrote:\\n>>> On Mon, Apr 3, 2017 at 12:02 PM, Joe Conway <mail@joeconway.com> wrote:\\n>>>>> 0002 looks extremely straightforward, but I wonder if we could get one\\n>>>>> of the people on this list who knows about sepgsql to have a look?\\n>>>>> (Stephen Frost, Joe Conway, KaiGai Kohei...)\\n>>>>\\n>>>> Will have a look later today.\\n>>>\\n>>> I think it is now tomorrow, unless your time zone is someplace very\\n>>> far away. :-)\\n>>\\n>> OBE -- I have scheduled time in 30 minutes from now, after I have gotten\\n>> my first cup of coffee ;-)\\n> \\n> After some discussion off-list, I've rebased and udpated the patches.\\n> Please see attached for further review.\\n\\nThanks -- will have another look and test on a machine with selinux\\nsetup. Robert, did you want me to take responsibility to commit on this\\nor just provide review/feedback?\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Tue, 4 Apr 2017 10:02:11 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/04/2017 10:02 AM, Joe Conway wrote:\\n> On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\\n>> After some discussion off-list, I've rebased and udpated the patches.\\n>> Please see attached for further review.\\n> \\n> Thanks -- will have another look and test on a machine with selinux\\n> setup. Robert, did you want me to take responsibility to commit on this\\n> or just provide review/feedback?\\n\\nI did some editorializing on these.\\n\\nIn particular I did not like the approach to fixing \\\"warning: ‘tclass’\\nmay be used uninitialized\\\" and ended up just doing it the same as was\\ndone elsewhere in relation.c already (set tclass = 0 in the variable\\ndeclaration). Along the way I also changed one instance of tclass from\\nuint16 to uint16_t for the sake of consistency.\\n\\nInterestingly we figured out that the warning was present with -Og, but\\nnot present with -O0, -O2, or -O3.\\n\\nIf you want to test, apply 0001a and 0001b before 0002.\\n\\nAny objections?\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Tue, 4 Apr 2017 15:56:28 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\\n> On 04/04/2017 10:02 AM, Joe Conway wrote:\\n>> On 04/04/2017 09:55 AM, Mike Palmiotto wrote:\\n>>> After some discussion off-list, I've rebased and udpated the patches.\\n>>> Please see attached for further review.\\n>>\\n>> Thanks -- will have another look and test on a machine with selinux\\n>> setup. Robert, did you want me to take responsibility to commit on this\\n>> or just provide review/feedback?\\n>\\n> I did some editorializing on these.\\n>\\n> In particular I did not like the approach to fixing \\\"warning: ‘tclass’\\n> may be used uninitialized\\\" and ended up just doing it the same as was\\n> done elsewhere in relation.c already (set tclass = 0 in the variable\\n> declaration). Along the way I also changed one instance of tclass from\\n> uint16 to uint16_t for the sake of consistency.\\n>\\n> Interestingly we figured out that the warning was present with -Og, but\\n> not present with -O0, -O2, or -O3.\\n>\\n> If you want to test, apply 0001a and 0001b before 0002.\\n>\\n> Any objections?\\n\\nI'm guessing Tom's going to have a strong feeling about whether 0001a\\nis the right way to address the stdbool issue, but I don't personally\\nknow what the right thing to do is so I will avoid opining on that\\ntopic.\\n\\nSo, nope, no objections here to you committing those.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Tue, 4 Apr 2017 21:29:19 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Robert Haas <robertmhaas@gmail.com> writes:\\n> On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> Any objections?\\n\\n> I'm guessing Tom's going to have a strong feeling about whether 0001a\\n> is the right way to address the stdbool issue,\\n\\nI will? [ looks ... ] Yup, you're right.\\n\\nI doubt that works at all, TBH. What I'd expect to happen with a\\ntypical compiler is a complaint about redefinition of typedef bool,\\nbecause c.h already declared it and here this fragment is doing\\nso again. It'd make sense to me to do\\n\\n+ #ifdef bool\\n+ #undef bool\\n+ #endif\\n\\nto get rid of the macro definition of bool that stdbool.h is\\nsupposed to provide. But there should be no reason to declare\\nour typedef a second time.\\n\\nAnother issue is whether you won't get compiler complaints about\\nredefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\nlikely only be warnings, not flat-out errors.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 00:58:07 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 4/5/17 00:58, Tom Lane wrote:\\n> Another issue is whether you won't get compiler complaints about\\n> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n> likely only be warnings, not flat-out errors.\\n\\nThe complaint about bool is also just a warning.\\n\\n-- \\nPeter Eisentraut http://www.2ndQuadrant.com/\\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 01:17:01 -0400\",\n \"msg_from\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\\n> On 4/5/17 00:58, Tom Lane wrote:\\n>> Another issue is whether you won't get compiler complaints about\\n>> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n>> likely only be warnings, not flat-out errors.\\n\\n> The complaint about bool is also just a warning.\\n\\nReally?\\n\\n$ cat test.c\\ntypedef char bool;\\ntypedef char bool;\\n$ gcc -c test.c\\ntest.c:2: error: redefinition of typedef 'bool'\\ntest.c:1: note: previous declaration of 'bool' was here\\n\\nThis is with gcc 4.4.7.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 01:20:20 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-04-05 00:58:07 -0400, Tom Lane wrote:\\n> Robert Haas <robertmhaas@gmail.com> writes:\\n> > On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\\n> >> Any objections?\\n> \\n> > I'm guessing Tom's going to have a strong feeling about whether 0001a\\n> > is the right way to address the stdbool issue,\\n> \\n> I will? [ looks ... ] Yup, you're right.\\n> \\n> I doubt that works at all, TBH. What I'd expect to happen with a\\n> typical compiler is a complaint about redefinition of typedef bool,\\n> because c.h already declared it and here this fragment is doing\\n> so again. It'd make sense to me to do\\n> \\n> + #ifdef bool\\n> + #undef bool\\n> + #endif\\n> \\n> to get rid of the macro definition of bool that stdbool.h is\\n> supposed to provide. But there should be no reason to declare\\n> our typedef a second time.\\n\\n> Another issue is whether you won't get compiler complaints about\\n> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n> likely only be warnings, not flat-out errors.\\n\\nI argued before that we should migrate to stdbool.h by default, because\\nit's only going to get more common. We already do so in a way for c++\\ncompilers...\\n\\n- Andres\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 00:02:49 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> I argued before that we should migrate to stdbool.h by default, because\\n> it's only going to get more common. We already do so in a way for c++\\n> compilers...\\n\\nYeah, I was just thinking about that. The core problem though is that\\nwe need the \\\"bool\\\" fields in the system catalog structs (or anyplace\\nelse that it represents an on-disk bool datum) to be understood as\\nbeing 1 byte wide. I do not think we can assume that that's true of\\nevery compiler's _Bool type. So we'd need some workaround for that.\\nThere are probably other places such as isnull arrays where it'd be\\nwise to force the width to be 1 byte.\\n\\nIn any case, that's a research project that's not getting done for v10.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 09:43:53 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Wed, Apr 5, 2017 at 12:58 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\\n> Robert Haas <robertmhaas@gmail.com> writes:\\n>> On Tue, Apr 4, 2017 at 6:56 PM, Joe Conway <mail@joeconway.com> wrote:\\n>>> Any objections?\\n>\\n>> I'm guessing Tom's going to have a strong feeling about whether 0001a\\n>> is the right way to address the stdbool issue,\\n>\\n> I will? [ looks ... ] Yup, you're right.\\n\\nIt happens occasionally. :-)\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 09:48:20 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\\n> Andres Freund <andres@anarazel.de> writes:\\n> > I argued before that we should migrate to stdbool.h by default, because\\n> > it's only going to get more common. We already do so in a way for c++\\n> > compilers...\\n> \\n> Yeah, I was just thinking about that. The core problem though is that\\n> we need the \\\"bool\\\" fields in the system catalog structs (or anyplace\\n> else that it represents an on-disk bool datum) to be understood as\\n> being 1 byte wide. I do not think we can assume that that's true of\\n> every compiler's _Bool type. So we'd need some workaround for that.\\n> There are probably other places such as isnull arrays where it'd be\\n> wise to force the width to be 1 byte.\\n\\nI wonder if there's any compiler that has _Bool/stdbool.h where it's not\\n1 byte sized. It's definitely not guaranteed by the standard.\\n\\nHaving a seperate booldatum type or such shouldn't be too bad, either\\nway.\\n\\n\\n> In any case, that's a research project that's not getting done for v10.\\n\\nAgreed.\\n\\n\\n- Andres\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 07:28:46 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\\n> On Fri, Mar 31, 2017 at 2:14 PM, Mike Palmiotto\\n> <mike.palmiotto@crunchydata.com> wrote:\\n> > Attached you will find two patches, which were rebased on master as of\\n> > 156d388 (applied with `git am --revert [patch file]`). The first gets\\n> > rid of some pesky compiler warnings and the second implements the\\n> > sepgsql handling of partitioned tables.\\n> \\n> 0001 has the problem that we have a firm rule against putting any\\n> #includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\n> up before, though, and I fear we're going to need to do something\\n> about it.\\n\\nI think in this case it makes sense to deviate from that rule,\\ntemporarily and locally.\\n\\n- Andres\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 07:30:42 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\\n>> Yeah, I was just thinking about that. The core problem though is that\\n>> we need the \\\"bool\\\" fields in the system catalog structs (or anyplace\\n>> else that it represents an on-disk bool datum) to be understood as\\n>> being 1 byte wide. I do not think we can assume that that's true of\\n>> every compiler's _Bool type. So we'd need some workaround for that.\\n>> There are probably other places such as isnull arrays where it'd be\\n>> wise to force the width to be 1 byte.\\n\\n> I wonder if there's any compiler that has _Bool/stdbool.h where it's not\\n> 1 byte sized. It's definitely not guaranteed by the standard.\\n\\nHm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\\n\\nIf it is char-sized almost everywhere, we could create a policy of\\nusing stdbool.h unless configure sees that _Bool is not char-sized.\\nOTOH, that doesn't improve our existing situation that we have\\nplatform-dependent semantics around \\\"bool\\\" (eg, what happens when\\na non-char-sized value is assigned). It would just change which\\none is the majority case, and not in a very safe direction :-(\\n\\n>> In any case, that's a research project that's not getting done for v10.\\n\\n> Agreed.\\n\\nYeah, it's off-topic for right now.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 10:45:06 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\\n>> 0001 has the problem that we have a firm rule against putting any\\n>> #includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\n>> up before, though, and I fear we're going to need to do something\\n>> about it.\\n\\n> I think in this case it makes sense to deviate from that rule,\\n> temporarily and locally.\\n\\nI'd really rather not. It might be safe here, because this code\\nonly works on Linux anyway, but it's still a dangerous precedent.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 10:48:27 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-04-05 10:45:06 -0400, Tom Lane wrote:\\n> Andres Freund <andres@anarazel.de> writes:\\n> > On 2017-04-05 09:43:53 -0400, Tom Lane wrote:\\n> >> Yeah, I was just thinking about that. The core problem though is that\\n> >> we need the \\\"bool\\\" fields in the system catalog structs (or anyplace\\n> >> else that it represents an on-disk bool datum) to be understood as\\n> >> being 1 byte wide. I do not think we can assume that that's true of\\n> >> every compiler's _Bool type. So we'd need some workaround for that.\\n> >> There are probably other places such as isnull arrays where it'd be\\n> >> wise to force the width to be 1 byte.\\n> \\n> > I wonder if there's any compiler that has _Bool/stdbool.h where it's not\\n> > 1 byte sized. It's definitely not guaranteed by the standard.\\n> \\n> Hm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\\n\\nI think nearly all x86 compilers use 1byte, but I assume there's some\\narchitectures where that'd be expensive.\\n\\n\\n> If it is char-sized almost everywhere, we could create a policy of\\n> using stdbool.h unless configure sees that _Bool is not char-sized.\\n> OTOH, that doesn't improve our existing situation that we have\\n> platform-dependent semantics around \\\"bool\\\" (eg, what happens when\\n> a non-char-sized value is assigned). It would just change which\\n> one is the majority case, and not in a very safe direction :-(\\n\\nYea, no, that doesn't like fun :(.\\n\\n\\nMight make sense to temporarily add a configure test checking for\\n_Bool/stdbool size, so we have some idea what we're talking about.\\n\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 07:52:00 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-04-05 10:48:27 -0400, Tom Lane wrote:\\n> Andres Freund <andres@anarazel.de> writes:\\n> > On 2017-03-31 20:23:39 -0400, Robert Haas wrote:\\n> >> 0001 has the problem that we have a firm rule against putting any\\n> >> #includes whatsoever before \\\"postgres.h\\\". This stdbool issue has come\\n> >> up before, though, and I fear we're going to need to do something\\n> >> about it.\\n> \\n> > I think in this case it makes sense to deviate from that rule,\\n> > temporarily and locally.\\n> \\n> I'd really rather not. It might be safe here, because this code\\n> only works on Linux anyway, but it's still a dangerous precedent.\\n\\nWell, what's the alternative for v10? There's already precedent btw.,\\ncf plperl.h undefining bool.\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 07:53:54 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> On 2017-04-05 10:48:27 -0400, Tom Lane wrote:\\n>> I'd really rather not. It might be safe here, because this code\\n>> only works on Linux anyway, but it's still a dangerous precedent.\\n\\n> Well, what's the alternative for v10? There's already precedent btw.,\\n> cf plperl.h undefining bool.\\n\\nWell, the 0001a patch amounted to undefining bool, once you drop\\nthe redundant typedef redeclaration. So that's a direction we have\\nsome track record with.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 11:06:32 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> On 2017-04-05 10:45:06 -0400, Tom Lane wrote:\\n>> Andres Freund <andres@anarazel.de> writes:\\n>>> I wonder if there's any compiler that has _Bool/stdbool.h where it's not\\n>>> 1 byte sized. It's definitely not guaranteed by the standard.\\n\\n>> Hm. I'd supposed that it'd be pretty common to make _Bool be int-sized.\\n\\n> I think nearly all x86 compilers use 1byte, but I assume there's some\\n> architectures where that'd be expensive.\\n\\n[ pokes around... ] prairiedog's gcc (Apple PPC) thinks sizeof(_Bool)\\nis 4.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 11:25:46 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/04/2017 09:58 PM, Tom Lane wrote:\\n> I doubt that works at all, TBH. What I'd expect to happen with a\\n> typical compiler is a complaint about redefinition of typedef bool,\\n> because c.h already declared it and here this fragment is doing\\n> so again. It'd make sense to me to do\\n> \\n> + #ifdef bool\\n> + #undef bool\\n> + #endif\\n> \\n> to get rid of the macro definition of bool that stdbool.h is\\n> supposed to provide. But there should be no reason to declare\\n> our typedef a second time.\\n\\nmakes sense\\n\\n> Another issue is whether you won't get compiler complaints about\\n> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n> likely only be warnings, not flat-out errors.\\n\\nI have not been able to generate warnings or errors around \\\"true\\\" and\\n\\\"false\\\".\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Wed, 5 Apr 2017 08:31:43 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 4/5/17 01:20, Tom Lane wrote:\\n>> The complaint about bool is also just a warning.\\n> \\n> Really?\\n> \\n> $ cat test.c\\n> typedef char bool;\\n> typedef char bool;\\n> $ gcc -c test.c\\n> test.c:2: error: redefinition of typedef 'bool'\\n> test.c:1: note: previous declaration of 'bool' was here\\n> \\n> This is with gcc 4.4.7.\\n\\nBut the above is not how the current code looks.\\n\\nstdbool.h does\\n\\n#define bool _Bool\\n\\nc.h does\\n\\n#ifndef bool\\ntypedef char bool;\\n#endif\\n\\nSo if you get stdbool.h first, then c.h does nothing. If you get c.h\\nfirst, then the macro from stdbool.h will mask the c.h typedef.\\n\\nWhere this gets really fun is when you include other header files\\nbetween, say, postgres.h and selinux/label.h, because then the function\\ndefinitions from those header files will be incompatible before and\\nafter the stdbool.h inclusion. Which indeed currently leads to the\\nfollowing compilation error:\\n\\nlabel.c: In function �sepgsql_init_client_label�:\\nlabel.c:437:18: error: assignment from incompatible pointer type\\n[-Werror=incompatible-pointer-types]\\n needs_fmgr_hook = sepgsql_needs_fmgr_hook;\\n ^\\n\\nSo I can't reproduce the original complaint, but this is potentially worse.\\n\\n(The above is on Fedora 25. You need a fairly new selinux.)\\n\\n-- \\nPeter Eisentraut http://www.2ndQuadrant.com/\\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 11:56:31 -0400\",\n \"msg_from\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Joe Conway <mail@joeconway.com> writes:\\n> On 04/04/2017 09:58 PM, Tom Lane wrote:\\n>> Another issue is whether you won't get compiler complaints about\\n>> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n>> likely only be warnings, not flat-out errors.\\n\\n> I have not been able to generate warnings or errors around \\\"true\\\" and\\n> \\\"false\\\".\\n\\nInteresting. Poking at it on a Fedora 25 machine, I also see a\\nbool-type-related warning in sepgsql/label.c, but nothing around macro\\nredefinitions. In particular, I find that\\n\\n#include \\\"postgres.h\\\"\\n\\n#include <stdbool.h>\\n\\nis completely silent. On the other hand,\\n\\n#include \\\"postgres.h\\\"\\n\\n#define bool _Bool\\n#define true 1\\n#define false 0\\n\\ngenerates the warnings I expected about \\\"true\\\" and \\\"false\\\" being\\nredefined. Which is damn odd, because I copied-and-pasted those\\nlines out of\\n/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h\\n\\nConclusion: Fedora's gcc is playing fast and loose somehow with the\\ncommand \\\"#include <stdbool.h>\\\"; that does not include the file\\nyou'd think it does, it does something magic inside the compiler.\\nThe magic evidently includes not complaining about duplicate macro\\ndefinitions for true and false.\\n\\nAnyway, I'd recommend that we do something like\\n\\n #include <selinux/label.h>\\n+\\n+/*\\n+ * <selinux/label.h> includes <stdbool.h>, which creates an incompatible\\n+ * #define for bool. Get rid of that so we can use our own typedef.\\n+ * (For obscure reasons, the \\\"true\\\" and \\\"false\\\" macros don't cause issues.)\\n+ */\\n+ #undef bool\\n\\nand call it good.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 12:04:00 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"\\n\\nOn April 5, 2017 9:04:00 AM PDT, Tom Lane <tgl@sss.pgh.pa.us> wrote:\\n>Joe Conway <mail@joeconway.com> writes:\\n>> On 04/04/2017 09:58 PM, Tom Lane wrote:\\n>>> Another issue is whether you won't get compiler complaints about\\n>>> redefinition of the \\\"true\\\" and \\\"false\\\" macros. But those would\\n>>> likely only be warnings, not flat-out errors.\\n>\\n>> I have not been able to generate warnings or errors around \\\"true\\\" and\\n>> \\\"false\\\".\\n>\\n>Interesting. Poking at it on a Fedora 25 machine, I also see a\\n>bool-type-related warning in sepgsql/label.c, but nothing around macro\\n>redefinitions. In particular, I find that\\n>\\n>#include \\\"postgres.h\\\"\\n>\\n>#include <stdbool.h>\\n>\\n>is completely silent. On the other hand,\\n>\\n>#include \\\"postgres.h\\\"\\n>\\n>#define bool _Bool\\n>#define true 1\\n>#define false 0\\n>\\n>generates the warnings I expected about \\\"true\\\" and \\\"false\\\" being\\n>redefined. Which is damn odd, because I copied-and-pasted those\\n>lines out of\\n>/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h\\n>\\n>Conclusion: Fedora's gcc is playing fast and loose somehow with the\\n>command \\\"#include <stdbool.h>\\\"; that does not include the file\\n>you'd think it does, it does something magic inside the compiler.\\n>The magic evidently includes not complaining about duplicate macro\\n>definitions for true and false.\\n>\\n>Anyway, I'd recommend that we do something like\\n>\\n> #include <selinux/label.h>\\n>+\\n>+/*\\n>+ * <selinux/label.h> includes <stdbool.h>, which creates an\\n>incompatible\\n>+ * #define for bool. Get rid of that so we can use our own typedef.\\n>+ * (For obscure reasons, the \\\"true\\\" and \\\"false\\\" macros don't cause\\n>issues.)\\n>+ */\\n\\nGCC generally doesn't warn about macro redefinitions, if both definitions are equivalent. IIRC we have some examples of that in the tree. There's a flag to warn regardless.\\n\\nAndres\\n-- \\nSent from my Android device with K-9 Mail. Please excuse my brevity.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 09:06:52 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\\n> On 4/5/17 01:20, Tom Lane wrote:\\n>> $ cat test.c\\n>> typedef char bool;\\n>> typedef char bool;\\n>> $ gcc -c test.c\\n>> test.c:2: error: redefinition of typedef 'bool'\\n>> test.c:1: note: previous declaration of 'bool' was here\\n\\n> But the above is not how the current code looks.\\n\\nNo, but it is what the proposed 0001a patch would result in.\\n\\nThe experiments I just did prove that F25's gcc is cheating like mad in\\nits implementation of \\\"#include <stdbool.h>\\\". I suspect that it's somehow\\nturning off the warning that you ought to get about redefining typedef\\nbool, as well, though I'm very unclear on how. In any case we don't need\\nthe extra typedef and should leave it out.\\n\\n> So if you get stdbool.h first, then c.h does nothing. If you get c.h\\n> first, then the macro from stdbool.h will mask the c.h typedef.\\n\\nRight, and *either* of those is disastrous if sizeof(_Bool) isn't 1.\\n\\n> Where this gets really fun is when you include other header files\\n> between, say, postgres.h and selinux/label.h, because then the function\\n> definitions from those header files will be incompatible before and\\n> after the stdbool.h inclusion. Which indeed currently leads to the\\n> following compilation error:\\n\\nThe compilation complaint comes from the fact that we have\\n\\nstatic bool\\nsepgsql_needs_fmgr_hook(Oid functionId)\\n\\nappearing after the inclusion of <stdbool.h>, causing it to mean\\nsomething else than what a Postgres programmer would expect.\\nThe other headers aren't at issue really.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 12:14:05 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Andres Freund <andres@anarazel.de> writes:\\n> GCC generally doesn't warn about macro redefinitions, if both\\n> definitions are equivalent.\\n\\nBut they're *not* equivalent. c.h has\\n\\n#define true\\t((bool) 1)\\n\\nwhereas so far as I can see the <stdbool.h> definition is just\\n\\n#define true\\t1\\n\\nAnd if you put those two definitions together you will get a warning.\\nThe lack of one with c.h + <stdbool.h> means somebody is cheating.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 12:26:25 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 4/5/17 12:04, Tom Lane wrote:\\n> Conclusion: Fedora's gcc is playing fast and loose somehow with the\\n> command \\\"#include <stdbool.h>\\\"; that does not include the file\\n> you'd think it does, it does something magic inside the compiler.\\n> The magic evidently includes not complaining about duplicate macro\\n> definitions for true and false.\\n\\nPerhaps -Wsystem-headers would change the outcome in your case.\\n\\n-- \\nPeter Eisentraut http://www.2ndQuadrant.com/\\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 12:53:58 -0400\",\n \"msg_from\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\\n> On 4/5/17 12:04, Tom Lane wrote:\\n>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\\n>> command \\\"#include <stdbool.h>\\\"; that does not include the file\\n>> you'd think it does, it does something magic inside the compiler.\\n>> The magic evidently includes not complaining about duplicate macro\\n>> definitions for true and false.\\n\\n> Perhaps -Wsystem-headers would change the outcome in your case.\\n\\nHah, you're right: with that,\\n\\nIn file included from /usr/include/selinux/label.h:9:0,\\n from label.c:40:\\n/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \\\"true\\\" redefined\\n #define true 1\\n \\nIn file included from ../../src/include/postgres.h:47:0,\\n from label.c:11:\\n../../src/include/c.h:206:0: note: this is the location of the previous definition\\n #define true ((bool) 1)\\n\\nand likewise for \\\"false\\\". So that mystery is explained.\\n\\nI stand by my previous patch suggestion, except that we can replace\\nthe parenthetical comment with something like \\\"(We don't care if\\n<stdbool.h> redefines \\\"true\\\"/\\\"false\\\"; those are close enough.)\\\".\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 05 Apr 2017 13:19:10 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Wed, Apr 5, 2017 at 1:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\\n> Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\\n>> On 4/5/17 12:04, Tom Lane wrote:\\n>>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\\n>>> command \\\"#include <stdbool.h>\\\"; that does not include the file\\n>>> you'd think it does, it does something magic inside the compiler.\\n>>> The magic evidently includes not complaining about duplicate macro\\n>>> definitions for true and false.\\n>\\n>> Perhaps -Wsystem-headers would change the outcome in your case.\\n>\\n> Hah, you're right: with that,\\n>\\n> In file included from /usr/include/selinux/label.h:9:0,\\n> from label.c:40:\\n> /usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \\\"true\\\" redefined\\n> #define true 1\\n>\\n> In file included from ../../src/include/postgres.h:47:0,\\n> from label.c:11:\\n> ../../src/include/c.h:206:0: note: this is the location of the previous definition\\n> #define true ((bool) 1)\\n>\\n> and likewise for \\\"false\\\". So that mystery is explained.\\n>\\n> I stand by my previous patch suggestion, except that we can replace\\n> the parenthetical comment with something like \\\"(We don't care if\\n> <stdbool.h> redefines \\\"true\\\"/\\\"false\\\"; those are close enough.)\\\".\\n>\\n\\nSounds good. Updated patchset will include that verbiage, along with\\nsome regression tests for partitioned tables.\\n\\nThanks,\\n\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 5 Apr 2017 13:22:22 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Wed, Apr 5, 2017 at 1:22 PM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> On Wed, Apr 5, 2017 at 1:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:\\n>> Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:\\n>>> On 4/5/17 12:04, Tom Lane wrote:\\n>>>> Conclusion: Fedora's gcc is playing fast and loose somehow with the\\n>>>> command \\\"#include <stdbool.h>\\\"; that does not include the file\\n>>>> you'd think it does, it does something magic inside the compiler.\\n>>>> The magic evidently includes not complaining about duplicate macro\\n>>>> definitions for true and false.\\n>>\\n>>> Perhaps -Wsystem-headers would change the outcome in your case.\\n>>\\n>> Hah, you're right: with that,\\n>>\\n>> In file included from /usr/include/selinux/label.h:9:0,\\n>> from label.c:40:\\n>> /usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stdbool.h:34:0: warning: \\\"true\\\" redefined\\n>> #define true 1\\n>>\\n>> In file included from ../../src/include/postgres.h:47:0,\\n>> from label.c:11:\\n>> ../../src/include/c.h:206:0: note: this is the location of the previous definition\\n>> #define true ((bool) 1)\\n>>\\n>> and likewise for \\\"false\\\". So that mystery is explained.\\n>>\\n>> I stand by my previous patch suggestion, except that we can replace\\n>> the parenthetical comment with something like \\\"(We don't care if\\n>> <stdbool.h> redefines \\\"true\\\"/\\\"false\\\"; those are close enough.)\\\".\\n>>\\n>\\n> Sounds good. Updated patchset will include that verbiage, along with\\n> some regression tests for partitioned tables.\\n\\nLooks like the label regression test is failing even without these changes.\\n\\nWeirdly enough, this only happens in one place:\\n\\n 84 SELECT objtype, objname, label FROM pg_seclabels\\n 85 WHERE provider = 'selinux' AND objtype = 'column' AND (objname\\nlike 't3.%' OR objname like 't4.%');\\n\\nI haven't figured out why this may be (yet), but it seems like we\\nshouldn't assume the order of output from a SELECT. As I understand\\nit, the order of the output is subject to change with changes to the\\nplanner/configuration.\\n\\nI'm going to hold the partition table regression changes for a\\nseparate patch and include some ORDER BY fixes. Will post tomorrow\\n\\nIn the meantime, attached are the latest and greatest patches.\\n\\nThanks,\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 5 Apr 2017 17:29:10 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/05/2017 02:29 PM, Mike Palmiotto wrote:\\n> I'm going to hold the partition table regression changes for a\\n> separate patch and include some ORDER BY fixes. Will post tomorrow\\n> \\n> In the meantime, attached are the latest and greatest patches.\\n\\nI'm going to push the attached in a few hours unless there is any\\nadditional discussion. As stated above we'll do the regression changes\\nin a separate patch once that is sorted. I used Tom's approach and\\ncomment wording for 0001a.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Thu, 6 Apr 2017 08:08:53 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Joe Conway <mail@joeconway.com> writes:\\n> I'm going to push the attached in a few hours unless there is any\\n> additional discussion. As stated above we'll do the regression changes\\n> in a separate patch once that is sorted. I used Tom's approach and\\n> comment wording for 0001a.\\n\\nLooks generally sane, but I noticed a grammatical nitpick:\\n\\n-\\t * Only attributes within regular relation or partition relations have\\n+\\t * Only attributes within regular relations or partition relations have\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 06 Apr 2017 11:29:58 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/06/2017 08:29 AM, Tom Lane wrote:\\n> Joe Conway <mail@joeconway.com> writes:\\n>> I'm going to push the attached in a few hours unless there is any\\n>> additional discussion. As stated above we'll do the regression changes\\n>> in a separate patch once that is sorted. I used Tom's approach and\\n>> comment wording for 0001a.\\n> \\n> Looks generally sane, but I noticed a grammatical nitpick:\\n> \\n> -\\t * Only attributes within regular relation or partition relations have\\n> +\\t * Only attributes within regular relations or partition relations have\\n\\nGood call -- thanks!\\n\\nAny thoughts on whether 0001a and 0001b ought to be backpatched? I'm\\nthinking not given the lack of past complaints but it might make sense\\nto do.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Thu, 6 Apr 2017 11:19:05 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Joe Conway <mail@joeconway.com> writes:\\n> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\\n> thinking not given the lack of past complaints but it might make sense\\n> to do.\\n\\nI think 0001a absolutely needs to be, because it is fixing what is really\\nan ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\\nof bool, but as things stand it's returning _Bool (which is why the\\ncompiler is complaining). Now we might get away with that on most\\nhardware, but on platforms where those are different widths, it's possible\\nto imagine function-return conventions that would make it fail.\\n\\n0001b seems to only be needed for compilers that aren't smart enough\\nto see that tclass won't be referenced for RELKIND_INDEX, so it's\\njust cosmetic.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 06 Apr 2017 15:35:40 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/06/2017 12:35 PM, Tom Lane wrote:\\n> Joe Conway <mail@joeconway.com> writes:\\n>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\\n>> thinking not given the lack of past complaints but it might make sense\\n>> to do.\\n> \\n> I think 0001a absolutely needs to be, because it is fixing what is really\\n> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\\n> of bool, but as things stand it's returning _Bool (which is why the\\n> compiler is complaining). Now we might get away with that on most\\n> hardware, but on platforms where those are different widths, it's possible\\n> to imagine function-return conventions that would make it fail.\\n> \\n> 0001b seems to only be needed for compilers that aren't smart enough\\n> to see that tclass won't be referenced for RELKIND_INDEX, so it's\\n> just cosmetic.\\n\\nOk, committed/pushed that way.\\n\\nI found some missing bits in the 0002 patch -- new version attached.\\nWill wait on new regression tests before committing, but I expect we'll\\nhave those by end of today and be able to commit the rest tomorrow.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Thu, 6 Apr 2017 14:52:24 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Thu, Apr 6, 2017 at 5:52 PM, Joe Conway <mail@joeconway.com> wrote:\\n> On 04/06/2017 12:35 PM, Tom Lane wrote:\\n>> Joe Conway <mail@joeconway.com> writes:\\n>>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\\n>>> thinking not given the lack of past complaints but it might make sense\\n>>> to do.\\n>>\\n>> I think 0001a absolutely needs to be, because it is fixing what is really\\n>> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\\n>> of bool, but as things stand it's returning _Bool (which is why the\\n>> compiler is complaining). Now we might get away with that on most\\n>> hardware, but on platforms where those are different widths, it's possible\\n>> to imagine function-return conventions that would make it fail.\\n>>\\n>> 0001b seems to only be needed for compilers that aren't smart enough\\n>> to see that tclass won't be referenced for RELKIND_INDEX, so it's\\n>> just cosmetic.\\n>\\n> Ok, committed/pushed that way.\\n>\\n> I found some missing bits in the 0002 patch -- new version attached.\\n> Will wait on new regression tests before committing, but I expect we'll\\n> have those by end of today and be able to commit the rest tomorrow.\\n\\nAttached are the regression test updates for partitioned tables.\\n\\nThanks,\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 7 Apr 2017 14:36:29 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Fri, Apr 7, 2017 at 2:36 PM, Mike Palmiotto\\n<mike.palmiotto@crunchydata.com> wrote:\\n> On Thu, Apr 6, 2017 at 5:52 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> On 04/06/2017 12:35 PM, Tom Lane wrote:\\n>>> Joe Conway <mail@joeconway.com> writes:\\n>>>> Any thoughts on whether 0001a and 0001b ought to be backpatched? I'm\\n>>>> thinking not given the lack of past complaints but it might make sense\\n>>>> to do.\\n>>>\\n>>> I think 0001a absolutely needs to be, because it is fixing what is really\\n>>> an ABI violation: sepgsql_needs_fmgr_hook is supposed to return our notion\\n>>> of bool, but as things stand it's returning _Bool (which is why the\\n>>> compiler is complaining). Now we might get away with that on most\\n>>> hardware, but on platforms where those are different widths, it's possible\\n>>> to imagine function-return conventions that would make it fail.\\n>>>\\n>>> 0001b seems to only be needed for compilers that aren't smart enough\\n>>> to see that tclass won't be referenced for RELKIND_INDEX, so it's\\n>>> just cosmetic.\\n>>\\n>> Ok, committed/pushed that way.\\n>>\\n>> I found some missing bits in the 0002 patch -- new version attached.\\n>> Will wait on new regression tests before committing, but I expect we'll\\n>> have those by end of today and be able to commit the rest tomorrow.\\n>\\n> Attached are the regression test updates for partitioned tables.\\n\\nActually attached this time.\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Fri, 7 Apr 2017 14:37:06 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/07/2017 11:37 AM, Mike Palmiotto wrote:\\n>>> I found some missing bits in the 0002 patch -- new version attached.\\n>>> Will wait on new regression tests before committing, but I expect we'll\\n>>> have those by end of today and be able to commit the rest tomorrow.\\n>>\\n>> Attached are the regression test updates for partitioned tables.\\n> \\n> Actually attached this time.\\n\\nBased on my review and testing of the 0002 patch I believe it is\\ncorrect. However Mike and I just went through the regression test patch\\nline by line and there are issues he needs to address -- there is no way\\nthat is happening by tonight as the output is very verbose and we need\\nto be sure we are both testing the correct things and getting the\\ncorrect behaviors.\\n\\nBased on that I can:\\n\\n1) commit the 0002 patch now before the feature freeze and follow up\\n with the regression test patch when ready in a couple of days\\n2) hold off on both patches until ready\\n3) push both patches to the next commitfest/pg11\\n\\nSome argue this is an open issue against the new partitioning feature in\\npg10 and therefore should be addressed now, and others do not. I can see\\nboth sides of that argument.\\n\\nIn any case, thoughts on what to do?\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Fri, 7 Apr 2017 14:22:19 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n> On 04/07/2017 11:37 AM, Mike Palmiotto wrote:\\n>>>> I found some missing bits in the 0002 patch -- new version attached.\\n>>>> Will wait on new regression tests before committing, but I expect we'll\\n>>>> have those by end of today and be able to commit the rest tomorrow.\\n>>>\\n>>> Attached are the regression test updates for partitioned tables.\\n>>\\n>> Actually attached this time.\\n>\\n> Based on my review and testing of the 0002 patch I believe it is\\n> correct. However Mike and I just went through the regression test patch\\n> line by line and there are issues he needs to address -- there is no way\\n> that is happening by tonight as the output is very verbose and we need\\n> to be sure we are both testing the correct things and getting the\\n> correct behaviors.\\n>\\n> Based on that I can:\\n>\\n> 1) commit the 0002 patch now before the feature freeze and follow up\\n> with the regression test patch when ready in a couple of days\\n> 2) hold off on both patches until ready\\n> 3) push both patches to the next commitfest/pg11\\n>\\n> Some argue this is an open issue against the new partitioning feature in\\n> pg10 and therefore should be addressed now, and others do not. I can see\\n> both sides of that argument.\\n>\\n> In any case, thoughts on what to do?\\n\\nSpeaking only for myself, I'm OK with any of those options, provided\\nthat that \\\"a couple\\\" means what my dictionary says it means.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 7 Apr 2017 20:36:51 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/07/2017 05:36 PM, Robert Haas wrote:\\n> On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> 1) commit the 0002 patch now before the feature freeze and follow up\\n>> with the regression test patch when ready in a couple of days\\n>> 2) hold off on both patches until ready\\n>> 3) push both patches to the next commitfest/pg11\\n>>\\n>> Some argue this is an open issue against the new partitioning feature in\\n>> pg10 and therefore should be addressed now, and others do not. I can see\\n>> both sides of that argument.\\n>>\\n>> In any case, thoughts on what to do?\\n> \\n> Speaking only for myself, I'm OK with any of those options, provided\\n> that that \\\"a couple\\\" means what my dictionary says it means.\\n\\nThanks. I'd prefer not to do #1 actually, and I think we can adhere to\\nthe dictionary meaning of \\\"a couple\\\" (i.e. by EOD Sunday). Assuming we\\nare ready by Sunday I will push both together (#2) or else I will move\\nthem both together to the next CF (#3).\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Fri, 7 Apr 2017 19:36:49 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Joe,\\n\\n* Joe Conway (mail@joeconway.com) wrote:\\n> On 04/07/2017 05:36 PM, Robert Haas wrote:\\n> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n> >> 1) commit the 0002 patch now before the feature freeze and follow up\\n> >> with the regression test patch when ready in a couple of days\\n> >> 2) hold off on both patches until ready\\n> >> 3) push both patches to the next commitfest/pg11\\n> >>\\n> >> Some argue this is an open issue against the new partitioning feature in\\n> >> pg10 and therefore should be addressed now, and others do not. I can see\\n> >> both sides of that argument.\\n> >>\\n> >> In any case, thoughts on what to do?\\n> > \\n> > Speaking only for myself, I'm OK with any of those options, provided\\n> > that that \\\"a couple\\\" means what my dictionary says it means.\\n> \\n> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\\n> the dictionary meaning of \\\"a couple\\\" (i.e. by EOD Sunday). Assuming we\\n> are ready by Sunday I will push both together (#2) or else I will move\\n> them both together to the next CF (#3).\\n\\nSounds good to me.\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Sat, 8 Apr 2017 10:29:21 -0400\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On Sat, Apr 8, 2017 at 10:29 AM, Stephen Frost <sfrost@snowman.net> wrote:\\n> Joe,\\n>\\n> * Joe Conway (mail@joeconway.com) wrote:\\n>> On 04/07/2017 05:36 PM, Robert Haas wrote:\\n>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> >> 1) commit the 0002 patch now before the feature freeze and follow up\\n>> >> with the regression test patch when ready in a couple of days\\n>> >> 2) hold off on both patches until ready\\n>> >> 3) push both patches to the next commitfest/pg11\\n>> >>\\n>> >> Some argue this is an open issue against the new partitioning feature in\\n>> >> pg10 and therefore should be addressed now, and others do not. I can see\\n>> >> both sides of that argument.\\n>> >>\\n>> >> In any case, thoughts on what to do?\\n>> >\\n>> > Speaking only for myself, I'm OK with any of those options, provided\\n>> > that that \\\"a couple\\\" means what my dictionary says it means.\\n>>\\n>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\\n>> the dictionary meaning of \\\"a couple\\\" (i.e. by EOD Sunday). Assuming we\\n>> are ready by Sunday I will push both together (#2) or else I will move\\n>> them both together to the next CF (#3).\\n\\nWithout further ado, the updated patch is attached.\\n\\nI added some comment updates, making it clearer where partitioned\\ntables are used and modified the test-cases to be as close to\\nlock-step with regular table regression testing as possible.\\n\\nLet me know if you find any further issues -- I made sure to review\\npretty thoroughly, so I'd be surprised, but it happens.\\n\\nThanks,\\n-- \\nMike Palmiotto\\nSoftware Engineer\\nCrunchy Data Solutions\\nhttps://crunchydata.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Sun, 9 Apr 2017 13:18:11 -0400\",\n \"msg_from\": \"Mike Palmiotto <mike.palmiotto@crunchydata.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/08/2017 07:29 AM, Stephen Frost wrote:\\n> * Joe Conway (mail@joeconway.com) wrote:\\n>> On 04/07/2017 05:36 PM, Robert Haas wrote:\\n>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n>> >> 1) commit the 0002 patch now before the feature freeze and follow up\\n>> >> with the regression test patch when ready in a couple of days\\n>> >> 2) hold off on both patches until ready\\n>> >> 3) push both patches to the next commitfest/pg11\\n>> >>\\n>> >> Some argue this is an open issue against the new partitioning feature in\\n>> >> pg10 and therefore should be addressed now, and others do not. I can see\\n>> >> both sides of that argument.\\n>> >>\\n>> >> In any case, thoughts on what to do?\\n>> > \\n>> > Speaking only for myself, I'm OK with any of those options, provided\\n>> > that that \\\"a couple\\\" means what my dictionary says it means.\\n>> \\n>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\\n>> the dictionary meaning of \\\"a couple\\\" (i.e. by EOD Sunday). Assuming we\\n>> are ready by Sunday I will push both together (#2) or else I will move\\n>> them both together to the next CF (#3).\\n> \\n> Sounds good to me.\\n\\nHaving heard no objections, committed and pushed as per option #2.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sun, 9 Apr 2017 14:04:45 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/09/2017 02:04 PM, Joe Conway wrote:\\n> On 04/08/2017 07:29 AM, Stephen Frost wrote:\\n>> * Joe Conway (mail@joeconway.com) wrote:\\n>>> On 04/07/2017 05:36 PM, Robert Haas wrote:\\n>>> > On Fri, Apr 7, 2017 at 5:22 PM, Joe Conway <mail@joeconway.com> wrote:\\n>>> >> 1) commit the 0002 patch now before the feature freeze and follow up\\n>>> >> with the regression test patch when ready in a couple of days\\n>>> >> 2) hold off on both patches until ready\\n>>> >> 3) push both patches to the next commitfest/pg11\\n>>> >>\\n>>> >> Some argue this is an open issue against the new partitioning feature in\\n>>> >> pg10 and therefore should be addressed now, and others do not. I can see\\n>>> >> both sides of that argument.\\n>>> >>\\n>>> >> In any case, thoughts on what to do?\\n>>> > \\n>>> > Speaking only for myself, I'm OK with any of those options, provided\\n>>> > that that \\\"a couple\\\" means what my dictionary says it means.\\n>>> \\n>>> Thanks. I'd prefer not to do #1 actually, and I think we can adhere to\\n>>> the dictionary meaning of \\\"a couple\\\" (i.e. by EOD Sunday). Assuming we\\n>>> are ready by Sunday I will push both together (#2) or else I will move\\n>>> them both together to the next CF (#3).\\n>> \\n>> Sounds good to me.\\n> \\n> Having heard no objections, committed and pushed as per option #2.\\n\\nInteresting -- rhino is now failing. I tested a minute ago manually on\\nthe same buildfarm animal and it passed. I'm on it.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sun, 9 Apr 2017 14:28:48 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\\n> Interesting -- rhino is now failing. I tested a minute ago manually on\\n> the same buildfarm animal and it passed. I'm on it.\\n\\nThe module for segpsql really needs to be improved so it logs\\nregression.diffs - iirc several other modules do.\\n\\nGreetings,\\n\\nAndres Freund\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Sun, 9 Apr 2017 14:49:45 -0700\",\n \"msg_from\": \"Andres Freund <andres@anarazel.de>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/09/2017 02:49 PM, Andres Freund wrote:\\n> On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\\n>> Interesting -- rhino is now failing. I tested a minute ago manually on\\n>> the same buildfarm animal and it passed. I'm on it.\\n> \\n> The module for segpsql really needs to be improved so it logs\\n> regression.diffs - iirc several other modules do.\\n\\nSure, but for another day I think.\\n\\nI turned on the buildfarm \\\"keep\\\" setting and looked at the diffs. The\\nissue is that in there are a few places that do \\\"SELECT ... FROM\\npg_seclabels ... ORDER BY ...\\\" and when manually testing I get default\\ndatabase encoding \\\"UTF8\\\" but with the buildfarm I get \\\"SQL_ASCII\\\", hence\\na different sorting.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sun, 9 Apr 2017 15:01:42 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/09/2017 03:01 PM, Joe Conway wrote:\\n> On 04/09/2017 02:49 PM, Andres Freund wrote:\\n>> On 2017-04-09 14:28:48 -0700, Joe Conway wrote:\\n>>> Interesting -- rhino is now failing. I tested a minute ago manually on\\n>>> the same buildfarm animal and it passed. I'm on it.\\n>> \\n>> The module for segpsql really needs to be improved so it logs\\n>> regression.diffs - iirc several other modules do.\\n> \\n> Sure, but for another day I think.\\n> \\n> I turned on the buildfarm \\\"keep\\\" setting and looked at the diffs. The\\n> issue is that in there are a few places that do \\\"SELECT ... FROM\\n> pg_seclabels ... ORDER BY ...\\\" and when manually testing I get default\\n> database encoding \\\"UTF8\\\" but with the buildfarm I get \\\"SQL_ASCII\\\", hence\\n> a different sorting.\\n\\nLooks like adding COLLATE \\\"C\\\" in the ORDER BY clauses does the trick. Is\\nthat the preferred method for this kind of issue?\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sun, 9 Apr 2017 15:45:14 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"Joe Conway <mail@joeconway.com> writes:\\n>> I turned on the buildfarm \\\"keep\\\" setting and looked at the diffs. The\\n>> issue is that in there are a few places that do \\\"SELECT ... FROM\\n>> pg_seclabels ... ORDER BY ...\\\" and when manually testing I get default\\n>> database encoding \\\"UTF8\\\" but with the buildfarm I get \\\"SQL_ASCII\\\", hence\\n>> a different sorting.\\n\\n> Looks like adding COLLATE \\\"C\\\" in the ORDER BY clauses does the trick. Is\\n> that the preferred method for this kind of issue?\\n\\nYeah, that's pretty much the standard fix nowadays.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Sun, 09 Apr 2017 19:14:41 -0400\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n },\n {\n \"msg_contents\": \"On 04/09/2017 04:14 PM, Tom Lane wrote:\\n> Joe Conway <mail@joeconway.com> writes:\\n>>> I turned on the buildfarm \\\"keep\\\" setting and looked at the diffs. The\\n>>> issue is that in there are a few places that do \\\"SELECT ... FROM\\n>>> pg_seclabels ... ORDER BY ...\\\" and when manually testing I get default\\n>>> database encoding \\\"UTF8\\\" but with the buildfarm I get \\\"SQL_ASCII\\\", hence\\n>>> a different sorting.\\n> \\n>> Looks like adding COLLATE \\\"C\\\" in the ORDER BY clauses does the trick. Is\\n>> that the preferred method for this kind of issue?\\n> \\n> Yeah, that's pretty much the standard fix nowadays.\\n\\n\\nGood to hear, thanks -- rhino is back to green.\\n\\nJoe\\n\\n-- \\nCrunchy Data - http://crunchydata.com\\nPostgreSQL Support for Secure Enterprises\\nConsulting, Training, & Open Source Development\",\n \"msg_date\": \"Sun, 9 Apr 2017 16:39:55 -0700\",\n \"msg_from\": \"Joe Conway <mail@joeconway.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partitioned tables and contrib/sepgsql\"\n }\n]","isConversational":false}}},{"rowIdx":44097,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"I was just doing some testing on [1] when I noticed that there's a problem\nwith parallel bitmap index scans scans.\n\nTest case:\n\npatch with [1]\n\n=# create table r1(value int);\nCREATE TABLE\n=# insert into r1 select (random()*1000)::int from\ngenerate_Series(1,1000000);\nINSERT 0 1000000\n=# create index on r1 using brin(value);\nCREATE INDEX\n=# set enable_seqscan=0;\nSET\n=# explain select * from r1 where value=555;\n QUERY PLAN\n\n-----------------------------------------------------------------------------------------\n Gather (cost=3623.52..11267.45 rows=5000 width=4)\n Workers Planned: 2\n -> Parallel Bitmap Heap Scan on r1 (cost=2623.52..9767.45 rows=2083\nwidth=4)\n Recheck Cond: (value = 555)\n -> Bitmap Index Scan on r1_value_idx (cost=0.00..2622.27\nrows=522036 width=0)\n Index Cond: (value = 555)\n(6 rows)\n\n=# explain analyze select * from r1 where value=555;\nserver closed the connection unexpectedly\nThis probably means the server terminated abnormally\nbefore or while processing the request.\nThe connection to the server was lost. Attempting reset: Succeeded.\n\nThe crash occurs in tbm_shared_iterate() at:\n\nPagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\n\n\nI see in tbm_prepare_shared_iterate() tbm->npages is zero. I'm unsure if\nbringetbitmap() does something different with npages than btgetbitmap()\naround setting npages?\n\nBut anyway, due to the npages being 0 the tbm->ptpages is not allocated\nin tbm_prepare_shared_iterate()\n\nif (tbm->npages)\n{\ntbm->ptpages = dsa_allocate(tbm->dsa, sizeof(PTIterationArray) +\ntbm->npages * sizeof(int));\n\nso when tbm_shared_iterate runs this code;\n\n/*\n* If both chunk and per-page data remain, must output the numerically\n* earlier page.\n*/\nif (istate->schunkptr < istate->nchunks)\n{\nPagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\nPagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\nBlockNumber chunk_blockno;\n\nchunk_blockno = chunk->blockno + istate->schunkbit;\n\nif (istate->spageptr >= istate->npages ||\nchunk_blockno < page->blockno)\n{\n/* Return a lossy page indicator from the chunk */\noutput->blockno = chunk_blockno;\noutput->ntuples = -1;\noutput->recheck = true;\nistate->schunkbit++;\n\nLWLockRelease(&istate->lock);\nreturn output;\n}\n}\n\nit fails, due to idxpages pointing to random memory\n\nProbably this is a simple fix for the authors, so passing it along. I'm a\nbit unable to see how the part above is meant to work.\n\n[1]\nhttps://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n\n-- \n David Rowley http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\nI was just doing some testing on [1] when I noticed that there's a problem with parallel bitmap index scans scans.Test case:patch with [1]=# create table r1(value int);CREATE TABLE=# insert into r1 select (random()*1000)::int from generate_Series(1,1000000);INSERT 0 1000000=# create index on r1 using brin(value);CREATE INDEX=# set enable_seqscan=0;SET=# explain select * from r1 where value=555; QUERY PLAN ----------------------------------------------------------------------------------------- Gather (cost=3623.52..11267.45 rows=5000 width=4) Workers Planned: 2 -> Parallel Bitmap Heap Scan on r1 (cost=2623.52..9767.45 rows=2083 width=4) Recheck Cond: (value = 555) -> Bitmap Index Scan on r1_value_idx (cost=0.00..2622.27 rows=522036 width=0) Index Cond: (value = 555)(6 rows)=# explain analyze select * from r1 where value=555;server closed the connection unexpectedlyThis probably means the server terminated abnormallybefore or while processing the request.The connection to the server was lost. Attempting reset: Succeeded.The crash occurs in tbm_shared_iterate() at:PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];I see in tbm_prepare_shared_iterate() tbm->npages is zero. I'm unsure if bringetbitmap() does something different with npages than btgetbitmap() around setting npages?But anyway, due to the npages being 0 the tbm->ptpages is not allocated in tbm_prepare_shared_iterate() if (tbm->npages) { tbm->ptpages = dsa_allocate(tbm->dsa, sizeof(PTIterationArray) + tbm->npages * sizeof(int));so when tbm_shared_iterate runs this code; /* * If both chunk and per-page data remain, must output the numerically * earlier page. */ if (istate->schunkptr < istate->nchunks) { PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]]; PagetableEntry *page = &ptbase[idxpages[istate->spageptr]]; BlockNumber chunk_blockno; chunk_blockno = chunk->blockno + istate->schunkbit; if (istate->spageptr >= istate->npages || chunk_blockno < page->blockno) { /* Return a lossy page indicator from the chunk */ output->blockno = chunk_blockno; output->ntuples = -1; output->recheck = true; istate->schunkbit++; LWLockRelease(&istate->lock); return output; } }it fails, due to idxpages pointing to random memoryProbably this is a simple fix for the authors, so passing it along. I'm a bit unable to see how the part above is meant to work.[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch-- David Rowley http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services","msg_date":"Fri, 10 Mar 2017 04:47:37 +1300","msg_from":"David Rowley <david.rowley@2ndquadrant.com>","msg_from_op":true,"msg_subject":"Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 9, 2017 at 9:17 PM, David Rowley\n<david.rowley@2ndquadrant.com> wrote:\n> patch with [1]\n>\n> =# create table r1(value int);\n> CREATE TABLE\n> =# insert into r1 select (random()*1000)::int from\n> generate_Series(1,1000000);\n> INSERT 0 1000000\n> =# create index on r1 using brin(value);\n> CREATE INDEX\n> =# set enable_seqscan=0;\n> SET\n> =# explain select * from r1 where value=555;\n\nI am looking into the issue, I have already reproduced it. I will\nupdate on this soon.\n\nThanks for reporting.\n\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 21:37:25 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 9, 2017 at 9:37 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n>> =# create table r1(value int);\n>> CREATE TABLE\n>> =# insert into r1 select (random()*1000)::int from\n>> generate_Series(1,1000000);\n>> INSERT 0 1000000\n>> =# create index on r1 using brin(value);\n>> CREATE INDEX\n>> =# set enable_seqscan=0;\n>> SET\n>> =# explain select * from r1 where value=555;\n>\n> I am looking into the issue, I have already reproduced it. I will\n> update on this soon.\n>\n> Thanks for reporting.\n\nI slightly modified your query to reproduce this issue.\n\nexplain analyze select * from r1 where value<555;\n\nPatch is attached to fix the problem.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 9 Mar 2017 22:02:12 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> I slightly modified your query to reproduce this issue.\n>\n> explain analyze select * from r1 where value<555;\n>\n> Patch is attached to fix the problem.\n\nI forgot to mention the cause of the problem.\n\nif (istate->schunkptr < istate->nchunks)\n{\n PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\n PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\n BlockNumber chunk_blockno;\n\nIn above if condition we have only checked istate->schunkptr <\nistate->nchunks that means we have some chunk left so we are safe to\naccess idxchunks, But just after that we are accessing\nptbase[idxpages[istate->spageptr]] without checking that accessing\nidxpages is safe or not.\n\ntbm_iterator already handling this case, I broke it in tbm_shared_iterator.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 22:20:56 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n>> I slightly modified your query to reproduce this issue.\n>>\n>> explain analyze select * from r1 where value<555;\n>>\n>> Patch is attached to fix the problem.\n>\n> I forgot to mention the cause of the problem.\n>\n> if (istate->schunkptr < istate->nchunks)\n> {\n> PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\n> PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\n> BlockNumber chunk_blockno;\n>\n> In above if condition we have only checked istate->schunkptr <\n> istate->nchunks that means we have some chunk left so we are safe to\n> access idxchunks, But just after that we are accessing\n> ptbase[idxpages[istate->spageptr]] without checking that accessing\n> idxpages is safe or not.\n>\n> tbm_iterator already handling this case, I broke it in tbm_shared_iterator.\n\nI don't know if this is the only problem -- it would be good if David\ncould retest -- but it's certainly *a* problem, so committed.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 9 Mar 2017 12:17:09 -0500","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On 10 March 2017 at 06:17, Robert Haas <robertmhaas@gmail.com> wrote:\n\n> On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com>\n> wrote:\n> > On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com>\n> wrote:\n> >> I slightly modified your query to reproduce this issue.\n> >>\n> >> explain analyze select * from r1 where value<555;\n> >>\n> >> Patch is attached to fix the problem.\n> >\n> > I forgot to mention the cause of the problem.\n> >\n> > if (istate->schunkptr < istate->nchunks)\n> > {\n> > PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\n> > PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\n> > BlockNumber chunk_blockno;\n> >\n> > In above if condition we have only checked istate->schunkptr <\n> > istate->nchunks that means we have some chunk left so we are safe to\n> > access idxchunks, But just after that we are accessing\n> > ptbase[idxpages[istate->spageptr]] without checking that accessing\n> > idxpages is safe or not.\n> >\n> > tbm_iterator already handling this case, I broke it in\n> tbm_shared_iterator.\n>\n> I don't know if this is the only problem -- it would be good if David\n> could retest -- but it's certainly *a* problem, so committed.\n>\n\nThanks for committing, and generally parallelising more stuff.\n\nI confirm that my test case is now working again.\n\nI'll be in this general area today, so will mention if I stumble over\nanything that looks broken.\n\n-- \n David Rowley http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\nOn 10 March 2017 at 06:17, Robert Haas <robertmhaas@gmail.com> wrote:On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n>> I slightly modified your query to reproduce this issue.\n>>\n>> explain analyze select * from r1 where value<555;\n>>\n>> Patch is attached to fix the problem.\n>\n> I forgot to mention the cause of the problem.\n>\n> if (istate->schunkptr < istate->nchunks)\n> {\n> PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\n> PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\n> BlockNumber chunk_blockno;\n>\n> In above if condition we have only checked istate->schunkptr <\n> istate->nchunks that means we have some chunk left so we are safe to\n> access idxchunks, But just after that we are accessing\n> ptbase[idxpages[istate->spageptr]] without checking that accessing\n> idxpages is safe or not.\n>\n> tbm_iterator already handling this case, I broke it in tbm_shared_iterator.\n\nI don't know if this is the only problem -- it would be good if David\ncould retest -- but it's certainly *a* problem, so committed.Thanks for committing, and generally parallelising more stuff.I confirm that my test case is now working again.I'll be in this general area today, so will mention if I stumble over anything that looks broken. -- David Rowley http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services","msg_date":"Fri, 10 Mar 2017 11:31:23 +1300","msg_from":"David Rowley <david.rowley@2ndquadrant.com>","msg_from_op":true,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> I don't know if this is the only problem\n\n> I'll be in this general area today, so will mention if I stumble over\n> anything that looks broken.\n\nI was testing the same patch with a large dataset and got a different segfault:\n\n> hasegeli=# explain select * from only mp_notification_20170225 where server_id = 7;\n> QUERY PLAN\n> ----------------------------------------------------------------------------------------------------------\n> Gather (cost=26682.94..476995.88 rows=1 width=215)\n> Workers Planned: 2\n> -> Parallel Bitmap Heap Scan on mp_notification_20170225 (cost=25682.94..475995.78 rows=1 width=215)\n> Recheck Cond: (server_id = 7)\n> -> Bitmap Index Scan on mp_notification_block_idx (cost=0.00..25682.94 rows=4557665 width=0)\n> Index Cond: (server_id = 7)\n> (6 rows)\n>\n> hasegeli=# select * from only mp_notification_20170225 where server_id = 7;\n> server closed the connection unexpectedly\n> This probably means the server terminated abnormally\n> before or while processing the request.\n> The connection to the server was lost. Attempting reset: Failed.\n\n> * thread #1: tid = 0x5045a8f, 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414, queue = 'com.apple.main-thread', stop reason = signal SIGUSR1\n> * frame #0: 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414 [opt]\n> frame #1: 0x000000010ae4000c postgres`bringetbitmap(scan=0x00007fea3c875c20, tbm=<unavailable>) + 428 at brin.c:398 [opt]\n> frame #2: 0x000000010ae9b451 postgres`index_getbitmap(scan=0x00007fea3c875c20, bitmap=<unavailable>) + 65 at indexam.c:726 [opt]\n> frame #3: 0x000000010b0035a9 postgres`MultiExecBitmapIndexScan(node=<unavailable>) + 233 at nodeBitmapIndexscan.c:91 [opt]\n> frame #4: 0x000000010b002840 postgres`BitmapHeapNext(node=<unavailable>) + 400 at nodeBitmapHeapscan.c:143 [opt]\n> frame #5: 0x000000010afef5d0 postgres`ExecProcNode(node=0x00007fea3c873948) + 224 at execProcnode.c:459 [opt]\n> frame #6: 0x000000010b004cc9 postgres`ExecGather [inlined] gather_getnext(gatherstate=<unavailable>) + 520 at nodeGather.c:276 [opt]\n> frame #7: 0x000000010b004ac1 postgres`ExecGather(node=<unavailable>) + 497 at nodeGather.c:212 [opt]\n> frame #8: 0x000000010afef6b2 postgres`ExecProcNode(node=0x00007fea3c872f58) + 450 at execProcnode.c:541 [opt]\n> frame #9: 0x000000010afeaf90 postgres`standard_ExecutorRun [inlined] ExecutePlan(estate=<unavailable>, planstate=<unavailable>, use_parallel_mode=<unavailable>, operation=<unavailable>, numberTuples=0, direction=<unavailable>, dest=<unavailable>) + 29 at execMain.c:1616 [opt]\n> frame #10: 0x000000010afeaf73 postgres`standard_ExecutorRun(queryDesc=<unavailable>, direction=<unavailable>, count=0) + 291 at execMain.c:348 [opt]\n> frame #11: 0x000000010af8b108 postgres`ExplainOnePlan(plannedstmt=0x00007fea3c871040, into=0x0000000000000000, es=0x00007fea3c805360, queryString=<unavailable>, params=<unavailable>, planduration=<unavailable>) + 328 at explain.c:533 [opt]\n> frame #12: 0x000000010af8ab98 postgres`ExplainOneQuery(query=0x00007fea3c805890, cursorOptions=<unavailable>, into=0x0000000000000000, es=0x00007fea3c805360, queryString=<unavailable>,params=0x0000000000000000) + 280 at explain.c:369 [opt]\n> frame #13: 0x000000010af8a773 postgres`ExplainQuery(pstate=<unavailable>, stmt=0x00007fea3d005450, queryString=\"explain analyze select * from only mp_notification_20170225 where server_id > 6;\",params=0x0000000000000000, dest=0x00007fea3c8052c8) + 819 at explain.c:254 [opt]\n> frame #14: 0x000000010b13b660 postgres`standard_ProcessUtility(pstmt=0x00007fea3d005fa8, queryString=\"explain analyze select * from only mp_notification_20170225 where server_id > 6;\",context=PROCESS_UTILITY_TOPLEVEL, params=0x0000000000000000, dest=0x00007fea3c8052c8, completionTag=<unavailable>) + 1104 at utility.c:675 [opt]\n> frame #15: 0x000000010b13ad2a postgres`PortalRunUtility(portal=0x00007fea3c837640, pstmt=0x00007fea3d005fa8, isTopLevel='\\x01', setHoldSnapshot=<unavailable>, dest=0x00007fea3c8052c8, completionTag=<unavailable>) + 90 at pquery.c:1165 [opt]\n> frame #16: 0x000000010b139f56 postgres`FillPortalStore(portal=0x00007fea3c837640, isTopLevel='\\x01') + 182 at pquery.c:1025 [opt]\n> frame #17: 0x000000010b139c22 postgres`PortalRun(portal=0x00007fea3c837640, count=<unavailable>, isTopLevel='\\x01', dest=<unavailable>, altdest=<unavailable>, completionTag=<unavailable>) + 402 at pquery.c:757 [opt]\n> frame #18: 0x000000010b13789b postgres`PostgresMain + 44 at postgres.c:1101 [opt]\n> frame #19: 0x000000010b13786f postgres`PostgresMain(argc=<unavailable>, argv=<unavailable>, dbname=<unavailable>, username=<unavailable>) + 8927 at postgres.c:4066 [opt]\n> frame #20: 0x000000010b0ba113 postgres`PostmasterMain [inlined] BackendRun + 7587 at postmaster.c:4317 [opt]\n> frame #21: 0x000000010b0ba0e8 postgres`PostmasterMain [inlined] BackendStartup at postmaster.c:3989 [opt]\n> frame #22: 0x000000010b0ba0e8 postgres`PostmasterMain at postmaster.c:1729 [opt]\n> frame #23: 0x000000010b0ba0e8 postgres`PostmasterMain(argc=<unavailable>, argv=<unavailable>) + 7544 at postmaster.c:1337 [opt]\n> frame #24: 0x000000010b0332af postgres`main(argc=<unavailable>, argv=<unavailable>) + 1567 at main.c:228 [opt]\n> frame #25: 0x00007fffb4e28255 libdyld.dylib`start + 1\n\nI can try to provide a test case, if that wouldn't be enough to spot\nthe problem.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 15:41:31 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Wed, Mar 15, 2017 at 8:11 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>\n> I can try to provide a test case, if that wouldn't be enough to spot\n> the problem.\n\nThanks for reporting, I am looking into this. Meanwhile, if you can\nprovide the reproducible test case then locating the issue will be\nfaster.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 20:51:22 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Wed, Mar 15, 2017 at 8:51 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n>> I can try to provide a test case, if that wouldn't be enough to spot\n>> the problem.\n>\n> Thanks for reporting, I am looking into this. Meanwhile, if you can\n> provide the reproducible test case then locating the issue will be\n> faster.\n\nAfter trying multiple attempts with different datasets I am unable to\nreproduce the issue.\n\nI tried with below test case:\ncreate table t(a int, b varchar);\n insert into t values(generate_series(1,10000000), repeat('x', 100));\n insert into t values(generate_series(1,100000000), repeat('x', 100));\ncreate index idx on t using brin(a);\npostgres=# analyze t;\nANALYZE\npostgres=# explain analyze select * from t where a>6;\n\nQUERY PLAN\n--------------------------------------------------------------------------------------------------------------------------------------------------\n Gather (cost=580794.52..3059826.52 rows=110414922 width=105) (actual\ntime=92.324..91853.716 rows=110425971 loops=1)\n Workers Planned: 2\n Workers Launched: 2\n -> Parallel Bitmap Heap Scan on t (cost=579794.52..3058826.52\nrows=46006218 width=105) (actual time=65.651..62023.020 rows=36808657\nloops=3)\n Recheck Cond: (a > 6)\n Rows Removed by Index Recheck: 4\n Heap Blocks: lossy=204401\n -> Bitmap Index Scan on idx (cost=0.00..552190.79\nrows=110425920 width=0) (actual time=88.215..88.215 rows=19040000\nloops=1)\n Index Cond: (a > 6)\n Planning time: 1.116 ms\n Execution time: 96176.881 ms\n(11 rows)\n\nIs it possible for you to provide a reproducible test case? I also\napplied the patch given up thread[1] but still could not reproduce.\n\n\n[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 21:35:52 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Wed, Mar 15, 2017 at 8:11 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> * thread #1: tid = 0x5045a8f, 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414, queue = 'com.apple.main-thread', stop reason = signal SIGUSR1\n>> * frame #0: 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414 [opt]\n>> frame #1: 0x000000010ae4000c postgres`bringetbitmap(scan=0x00007fea3c875c20, tbm=<unavailable>) + 428 at brin.c:398 [opt]\n>> frame #2: 0x000000010ae9b451 postgres`index_getbitmap(scan=0x00007fea3c875c20, bitmap=<unavailable>) + 65 at indexam.c:726 [opt]\n>> frame #3: 0x000000010b0035a9 postgres`MultiExecBitmapIndexScan(node=<unavailable>) + 233 at nodeBitmapIndexscan.c:91 [opt]\n>> frame #4: 0x000000010b002840 postgres`BitmapHeapNext(node=<unavailable>) + 400 at nodeBitmapHeapscan.c:143 [opt]\n\nFurther analyzing the call stack, seems like this is not exact call\nstack where it crashed. Because, if you notice the code in the\nbrin_deform_tuple (line 414)\n\nbrin_deform_tuple(BrinDesc *brdesc, BrinTuple *tuple)\n\n{\n dtup = brin_new_memtuple(brdesc);\n\n if (BrinTupleIsPlaceholder(tuple))\n dtup->bt_placeholder = true;\n dtup->bt_blkno = tuple->bt_blkno; --> line 414\n\nThis can crash at line:414, if either tuple is invalid memory(but I\nthink it's not because we have already accessed this memory in above\nif check) or dtup is invalid (this is also not possible because\nbrin_new_memtuple has already accessed this).\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 21:53:08 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> This can crash at line:414, if either tuple is invalid memory(but I\n> think it's not because we have already accessed this memory in above\n> if check) or dtup is invalid (this is also not possible because\n> brin_new_memtuple has already accessed this).\n\nI was testing with the brin correlation patch [1] applied. I cannot\ncrash it without the patch either. I am sorry for not testing it\nbefore. The patch make BRIN selectivity estimation function access\nmore information.\n\n[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 17:32:58 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Wed, Mar 15, 2017 at 10:02 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n> I was testing with the brin correlation patch [1] applied. I cannot\n> crash it without the patch either. I am sorry for not testing it\n> before. The patch make BRIN selectivity estimation function access\n> more information.\n>\n> [1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n\nWith my test case, I could not crash even with this patch applied.\nCan you provide your test case?\n(table, index, data, query)\n\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 22:05:59 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> With my test case, I could not crash even with this patch applied.\n> Can you provide your test case?\n\nYes:\n\n> hasegeli=# create table r2 as select (random() * 3)::int as i from generate_series(1, 1000000);\n> SELECT 1000000\n> hasegeli=# create index on r2 using brin (i);\n> CREATE INDEX\n> hasegeli=# analyze r2;\n> ANALYZE\n> hasegeli=# explain select * from only r2 where i = 10;\n> QUERY PLAN\n> -------------------------------------------------------------------------------------\n> Gather (cost=2867.50..9225.32 rows=1 width=4)\n> Workers Planned: 2\n> -> Parallel Bitmap Heap Scan on r2 (cost=1867.50..8225.22 rows=1 width=4)\n> Recheck Cond: (i = 10)\n> -> Bitmap Index Scan on r2_i_idx (cost=0.00..1867.50 rows=371082 width=0)\n> Index Cond: (i = 10)\n> (6 rows)\n>\n> hasegeli=# select * from only r2 where i = 10;\n> server closed the connection unexpectedly\n> This probably means the server terminated abnormally\n> before or while processing the request.\n> The connection to the server was lost. Attempting reset: Failed.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 17:51:23 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Wed, Mar 15, 2017 at 10:21 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> hasegeli=# create table r2 as select (random() * 3)::int as i from generate_series(1, 1000000);\n>> SELECT 1000000\n>> hasegeli=# create index on r2 using brin (i);\n>> CREATE INDEX\n>> hasegeli=# analyze r2;\n>> ANALYZE\n>> hasegeli=# explain select * from only r2 where i = 10;\n>> QUERY PLAN\n>> -------------------------------------------------------------------------------------\n>> Gather (cost=2867.50..9225.32 rows=1 width=4)\n>> Workers Planned: 2\n>> -> Parallel Bitmap Heap Scan on r2 (cost=1867.50..8225.22 rows=1 width=4)\n>> Recheck Cond: (i = 10)\n>> -> Bitmap Index Scan on r2_i_idx (cost=0.00..1867.50 rows=371082 width=0)\n>> Index Cond: (i = 10)\n>> (6 rows)\n>>\n>> hasegeli=# select * from only r2 where i = 10;\n\nI am able to reproduce the bug, and attached patch fixes the same.\nProblem is that I am not handling TBM_EMPTY state properly. I\nremember that while reviewing the patch Robert mentioned that we might\nneed to handle the TBM_EMPTY and I told that since we are not handling\nin non-parallel mode so we don't need to handle here as well. But, I\nwas wrong. So the problem is that if state is not TBM_HASH then it's\ndirectly assuming TBM_ONE_PAGE which is completely wrong. I have\nfixed that and also fixed in other similar locations.\n\nPlease verify the fix.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Wed, 15 Mar 2017 23:01:27 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> Please verify the fix.\n\nThe same test with both of the patches applied still crashes for me.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Wed, 15 Mar 2017 20:26:36 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 12:56 AM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> Please verify the fix.\n>\n> The same test with both of the patches applied still crashes for me.\nAfter above fix, I am not able to reproduce. Can you give me the\nbacktrace of the crash location or the dump?\n\nI am trying on the below commit\n\ncommit c5832346625af4193b1242e57e7d13e66a220b38\nAuthor: Stephen Frost <sfrost@snowman.net>\nDate: Wed Mar 15 11:19:39 2017 -0400\n\n+ https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n+ fix_tbm_empty.patch\n\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 05:02:42 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 5:02 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> After above fix, I am not able to reproduce. Can you give me the\n> backtrace of the crash location or the dump?\n>\n> I am trying on the below commit\n>\n> commit c5832346625af4193b1242e57e7d13e66a220b38\n> Author: Stephen Frost <sfrost@snowman.net>\n> Date: Wed Mar 15 11:19:39 2017 -0400\n>\n> + https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\n> + fix_tbm_empty.patch\n\nForgot to mention after fix I am seeing this output.\n\npostgres=# explain analyze select * from only r2 where i = 10;\n QUERY PLAN\n-------------------------------------------------------------------------------------------------------------------------------\n Gather (cost=2880.56..9251.98 rows=1 width=4) (actual\ntime=3.857..3.857 rows=0 loops=1)\n Workers Planned: 2\n Workers Launched: 2\n -> Parallel Bitmap Heap Scan on r2 (cost=1880.56..8251.88 rows=1\nwidth=4) (actual time=0.043..0.043 rows=0 loops=3)\n Recheck Cond: (i = 10)\n -> Bitmap Index Scan on r2_i_idx (cost=0.00..1880.56\nrows=373694 width=0) (actual time=0.052..0.052 rows=0 loops=1)\n Index Cond: (i = 10)\n Planning time: 0.111 ms\n Execution time: 4.449 ms\n(9 rows)\n\npostgres=# select * from only r2 where i = 10;\n i\n---\n(0 rows)\n\nAre you getting the crash with the same test case?\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 05:11:17 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> Are you getting the crash with the same test case?\n\nYes. Here is the new backtrace:\n\n> * thread #1: tid = 0x51828fd, 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\n> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\n> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\n> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\n> frame #3: 0x0000000100caf303 postgres`tbm_prepare_shared_iterate(tbm=<unavailable>) + 723 at tidbitmap.c:875 [opt]\n> frame #4: 0x0000000100c74844 postgres`BitmapHeapNext(node=<unavailable>) + 436 at nodeBitmapHeapscan.c:154 [opt]\n> frame #5: 0x0000000100c615b0 postgres`ExecProcNode(node=0x00007fdabf8189f0) + 224 at execProcnode.c:459 [opt]\n> frame #6: 0x0000000100c76ca9 postgres`ExecGather [inlined] gather_getnext(gatherstate=<unavailable>) + 520 at nodeGather.c:276 [opt]\n> frame #7: 0x0000000100c76aa1 postgres`ExecGather(node=<unavailable>) + 497 at nodeGather.c:212 [opt]\n> frame #8: 0x0000000100c61692 postgres`ExecProcNode(node=0x00007fdabf818558) + 450 at execProcnode.c:541 [opt]\n> frame #9: 0x0000000100c5cf70 postgres`standard_ExecutorRun [inlined] ExecutePlan(estate=<unavailable>, planstate=<unavailable>, use_parallel_mode=<unavailable>, operation=<unavailable>, numberTuples=0, direction=<unavailable>, dest=<unavailable>) + 29 at execMain.c:1616 [opt]\n> frame #10: 0x0000000100c5cf53 postgres`standard_ExecutorRun(queryDesc=<unavailable>, direction=<unavailable>, count=0) + 291 at execMain.c:348 [opt]\n> frame #11: 0x0000000100dac0df postgres`PortalRunSelect(portal=0x00007fdac000b240, forward=<unavailable>, count=0, dest=<unavailable>) + 255 at pquery.c:921 [opt]\n> frame #12: 0x0000000100dabc84 postgres`PortalRun(portal=0x00007fdac000b240, count=<unavailable>, isTopLevel='\\x01', dest=<unavailable>, altdest=<unavailable>, completionTag=<unavailable>) + 500 at pquery.c:762 [opt]\n> frame #13: 0x0000000100da989b postgres`PostgresMain + 44 at postgres.c:1101 [opt]\n> frame #14: 0x0000000100da986f postgres`PostgresMain(argc=<unavailable>, argv=<unavailable>, dbname=<unavailable>, username=<unavailable>) + 8927 at postgres.c:4066 [opt]\n> frame #15: 0x0000000100d2c113 postgres`PostmasterMain [inlined] BackendRun + 7587 at postmaster.c:4317 [opt]\n> frame #16: 0x0000000100d2c0e8 postgres`PostmasterMain [inlined] BackendStartup at postmaster.c:3989 [opt]\n> frame #17: 0x0000000100d2c0e8 postgres`PostmasterMain at postmaster.c:1729 [opt]\n> frame #18: 0x0000000100d2c0e8 postgres`PostmasterMain(argc=<unavailable>, argv=<unavailable>) + 7544 at postmaster.c:1337 [opt]\n> frame #19: 0x0000000100ca528f postgres`main(argc=<unavailable>, argv=<unavailable>) + 1567 at main.c:228 [opt]\n> frame #20: 0x00007fffb4e28255 libdyld.dylib`start + 1\n> frame #21: 0x00007fffb4e28255 libdyld.dylib`start + 1\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 11:22:02 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 3:52 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> * thread #1: tid = 0x51828fd, 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\n>> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\n>> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\n>> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\n\nBy looking at the call stack I got the problem location. I am\nreviewing other parts of the code if there are the similar mistake at\nother places. Soon I will post the patch. Thanks for the help.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 17:14:46 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 5:14 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\n>>> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\n>>> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\n>>> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\n>\n> By looking at the call stack I got the problem location. I am\n> reviewing other parts of the code if there are the similar mistake at\n> other places. Soon I will post the patch. Thanks for the help.\n\nBased on the call stack I have tried to fix the issue. The problem is\nthere was some uninitialized pointer access (in some special cases\ni.e. TBM_EMPTY when pagetable is not created at all).\n\n fix_tbm_empty.patch have fixed some of them but induced one which you\nare seeing in your call stack.\n\nHopefully, this time I got it correct. Since I am unable to reproduce\nthe issue so I will again need your help in verifying the fix.\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 16 Mar 2017 17:56:46 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"> Hopefully, this time I got it correct. Since I am unable to reproduce\n> the issue so I will again need your help in verifying the fix.\n\nIt is not crashing with the new patch. Thank you.\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 15:56:27 +0100","msg_from":"Emre Hasegeli <emre@hasegeli.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 10:56 AM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> Hopefully, this time I got it correct. Since I am unable to reproduce\n>> the issue so I will again need your help in verifying the fix.\n>\n> It is not crashing with the new patch. Thank you.\n\nThanks for confirming. Some review comments on v2:\n\n+ if (istate->pagetable)\n\nPlease compare explicitly to InvalidDsaPointer.\n\n+ if (iterator->ptbase)\n+ ptbase = iterator->ptbase->ptentry;\n+ if (iterator->ptpages)\n+ idxpages = iterator->ptpages->index;\n+ if (iterator->ptchunks)\n+ idxchunks = iterator->ptchunks->index;\n\nSimilarly.\n\nDilip, please also provide a proposed commit message describing what\nthis is fixing. Is it just the TBM_EMPTY case, or is there anything\nelse?\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 11:12:36 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 8:42 PM, Robert Haas <robertmhaas@gmail.com> wrote:\n> Thanks for confirming. Some review comments on v2:\n>\n> + if (istate->pagetable)\nfixed\n>\n> Please compare explicitly to InvalidDsaPointer.\n>\n> + if (iterator->ptbase)\n> + ptbase = iterator->ptbase->ptentry;\n> + if (iterator->ptpages)\n> + idxpages = iterator->ptpages->index;\n> + if (iterator->ptchunks)\n> + idxchunks = iterator->ptchunks->index;\n>\n> Similarly.\nfixed\n\nAlso fixed at\n+ if (ptbase)\n+ pg_atomic_init_u32(&ptbase->refcount, 0);\n\n>\n> Dilip, please also provide a proposed commit message describing what\n> this is fixing. Is it just the TBM_EMPTY case, or is there anything\n> else?\n\nOkay, I have added the commit message in the patch.\n\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 16 Mar 2017 23:20:15 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 8:26 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\n>> Hopefully, this time I got it correct. Since I am unable to reproduce\n>> the issue so I will again need your help in verifying the fix.\n>\n> It is not crashing with the new patch. Thank you.\n\nThanks for verifying.\n\n\n-- \nRegards,\nDilip Kumar\nEnterpriseDB: http://www.enterprisedb.com\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 23:20:42 +0530","msg_from":"Dilip Kumar <dilipbalaut@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"},{"msg_contents":"On Thu, Mar 16, 2017 at 1:50 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\n> fixed\n\nCommitted.\n\n-- \nRobert Haas\nEnterpriseDB: http://www.enterprisedb.com\nThe Enterprise PostgreSQL Company\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 16 Mar 2017 15:10:48 -0400","msg_from":"Robert Haas <robertmhaas@gmail.com>","msg_from_op":false,"msg_subject":"Re: Parallel Bitmap scans a bit broken"}],"string":"[\n {\n \"msg_contents\": \"I was just doing some testing on [1] when I noticed that there's a problem\\nwith parallel bitmap index scans scans.\\n\\nTest case:\\n\\npatch with [1]\\n\\n=# create table r1(value int);\\nCREATE TABLE\\n=# insert into r1 select (random()*1000)::int from\\ngenerate_Series(1,1000000);\\nINSERT 0 1000000\\n=# create index on r1 using brin(value);\\nCREATE INDEX\\n=# set enable_seqscan=0;\\nSET\\n=# explain select * from r1 where value=555;\\n QUERY PLAN\\n\\n-----------------------------------------------------------------------------------------\\n Gather (cost=3623.52..11267.45 rows=5000 width=4)\\n Workers Planned: 2\\n -> Parallel Bitmap Heap Scan on r1 (cost=2623.52..9767.45 rows=2083\\nwidth=4)\\n Recheck Cond: (value = 555)\\n -> Bitmap Index Scan on r1_value_idx (cost=0.00..2622.27\\nrows=522036 width=0)\\n Index Cond: (value = 555)\\n(6 rows)\\n\\n=# explain analyze select * from r1 where value=555;\\nserver closed the connection unexpectedly\\nThis probably means the server terminated abnormally\\nbefore or while processing the request.\\nThe connection to the server was lost. Attempting reset: Succeeded.\\n\\nThe crash occurs in tbm_shared_iterate() at:\\n\\nPagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\n\\n\\nI see in tbm_prepare_shared_iterate() tbm->npages is zero. I'm unsure if\\nbringetbitmap() does something different with npages than btgetbitmap()\\naround setting npages?\\n\\nBut anyway, due to the npages being 0 the tbm->ptpages is not allocated\\nin tbm_prepare_shared_iterate()\\n\\nif (tbm->npages)\\n{\\ntbm->ptpages = dsa_allocate(tbm->dsa, sizeof(PTIterationArray) +\\ntbm->npages * sizeof(int));\\n\\nso when tbm_shared_iterate runs this code;\\n\\n/*\\n* If both chunk and per-page data remain, must output the numerically\\n* earlier page.\\n*/\\nif (istate->schunkptr < istate->nchunks)\\n{\\nPagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\\nPagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\nBlockNumber chunk_blockno;\\n\\nchunk_blockno = chunk->blockno + istate->schunkbit;\\n\\nif (istate->spageptr >= istate->npages ||\\nchunk_blockno < page->blockno)\\n{\\n/* Return a lossy page indicator from the chunk */\\noutput->blockno = chunk_blockno;\\noutput->ntuples = -1;\\noutput->recheck = true;\\nistate->schunkbit++;\\n\\nLWLockRelease(&istate->lock);\\nreturn output;\\n}\\n}\\n\\nit fails, due to idxpages pointing to random memory\\n\\nProbably this is a simple fix for the authors, so passing it along. I'm a\\nbit unable to see how the part above is meant to work.\\n\\n[1]\\nhttps://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n\\n-- \\n David Rowley http://www.2ndQuadrant.com/\\n PostgreSQL Development, 24x7 Support, Training & Services\\n\\nI was just doing some testing on [1] when I noticed that there's a problem with parallel bitmap index scans scans.Test case:patch with [1]=# create table r1(value int);CREATE TABLE=# insert into r1 select (random()*1000)::int from generate_Series(1,1000000);INSERT 0 1000000=# create index on r1 using brin(value);CREATE INDEX=# set enable_seqscan=0;SET=# explain select * from r1 where value=555; QUERY PLAN ----------------------------------------------------------------------------------------- Gather (cost=3623.52..11267.45 rows=5000 width=4) Workers Planned: 2 -> Parallel Bitmap Heap Scan on r1 (cost=2623.52..9767.45 rows=2083 width=4) Recheck Cond: (value = 555) -> Bitmap Index Scan on r1_value_idx (cost=0.00..2622.27 rows=522036 width=0) Index Cond: (value = 555)(6 rows)=# explain analyze select * from r1 where value=555;server closed the connection unexpectedlyThis probably means the server terminated abnormallybefore or while processing the request.The connection to the server was lost. Attempting reset: Succeeded.The crash occurs in tbm_shared_iterate() at:PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];I see in tbm_prepare_shared_iterate() tbm->npages is zero. I'm unsure if bringetbitmap() does something different with npages than btgetbitmap() around setting npages?But anyway, due to the npages being 0 the tbm->ptpages is not allocated in tbm_prepare_shared_iterate() if (tbm->npages) { tbm->ptpages = dsa_allocate(tbm->dsa, sizeof(PTIterationArray) + tbm->npages * sizeof(int));so when tbm_shared_iterate runs this code; /* * If both chunk and per-page data remain, must output the numerically * earlier page. */ if (istate->schunkptr < istate->nchunks) { PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]]; PagetableEntry *page = &ptbase[idxpages[istate->spageptr]]; BlockNumber chunk_blockno; chunk_blockno = chunk->blockno + istate->schunkbit; if (istate->spageptr >= istate->npages || chunk_blockno < page->blockno) { /* Return a lossy page indicator from the chunk */ output->blockno = chunk_blockno; output->ntuples = -1; output->recheck = true; istate->schunkbit++; LWLockRelease(&istate->lock); return output; } }it fails, due to idxpages pointing to random memoryProbably this is a simple fix for the authors, so passing it along. I'm a bit unable to see how the part above is meant to work.[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch-- David Rowley http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services\",\n \"msg_date\": \"Fri, 10 Mar 2017 04:47:37 +1300\",\n \"msg_from\": \"David Rowley <david.rowley@2ndquadrant.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 9:17 PM, David Rowley\\n<david.rowley@2ndquadrant.com> wrote:\\n> patch with [1]\\n>\\n> =# create table r1(value int);\\n> CREATE TABLE\\n> =# insert into r1 select (random()*1000)::int from\\n> generate_Series(1,1000000);\\n> INSERT 0 1000000\\n> =# create index on r1 using brin(value);\\n> CREATE INDEX\\n> =# set enable_seqscan=0;\\n> SET\\n> =# explain select * from r1 where value=555;\\n\\nI am looking into the issue, I have already reproduced it. I will\\nupdate on this soon.\\n\\nThanks for reporting.\\n\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 21:37:25 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 9:37 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n>> =# create table r1(value int);\\n>> CREATE TABLE\\n>> =# insert into r1 select (random()*1000)::int from\\n>> generate_Series(1,1000000);\\n>> INSERT 0 1000000\\n>> =# create index on r1 using brin(value);\\n>> CREATE INDEX\\n>> =# set enable_seqscan=0;\\n>> SET\\n>> =# explain select * from r1 where value=555;\\n>\\n> I am looking into the issue, I have already reproduced it. I will\\n> update on this soon.\\n>\\n> Thanks for reporting.\\n\\nI slightly modified your query to reproduce this issue.\\n\\nexplain analyze select * from r1 where value<555;\\n\\nPatch is attached to fix the problem.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 9 Mar 2017 22:02:12 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> I slightly modified your query to reproduce this issue.\\n>\\n> explain analyze select * from r1 where value<555;\\n>\\n> Patch is attached to fix the problem.\\n\\nI forgot to mention the cause of the problem.\\n\\nif (istate->schunkptr < istate->nchunks)\\n{\\n PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\\n PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\n BlockNumber chunk_blockno;\\n\\nIn above if condition we have only checked istate->schunkptr <\\nistate->nchunks that means we have some chunk left so we are safe to\\naccess idxchunks, But just after that we are accessing\\nptbase[idxpages[istate->spageptr]] without checking that accessing\\nidxpages is safe or not.\\n\\ntbm_iterator already handling this case, I broke it in tbm_shared_iterator.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 22:20:56 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n>> I slightly modified your query to reproduce this issue.\\n>>\\n>> explain analyze select * from r1 where value<555;\\n>>\\n>> Patch is attached to fix the problem.\\n>\\n> I forgot to mention the cause of the problem.\\n>\\n> if (istate->schunkptr < istate->nchunks)\\n> {\\n> PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\\n> PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\n> BlockNumber chunk_blockno;\\n>\\n> In above if condition we have only checked istate->schunkptr <\\n> istate->nchunks that means we have some chunk left so we are safe to\\n> access idxchunks, But just after that we are accessing\\n> ptbase[idxpages[istate->spageptr]] without checking that accessing\\n> idxpages is safe or not.\\n>\\n> tbm_iterator already handling this case, I broke it in tbm_shared_iterator.\\n\\nI don't know if this is the only problem -- it would be good if David\\ncould retest -- but it's certainly *a* problem, so committed.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 9 Mar 2017 12:17:09 -0500\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On 10 March 2017 at 06:17, Robert Haas <robertmhaas@gmail.com> wrote:\\n\\n> On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com>\\n> wrote:\\n> > On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com>\\n> wrote:\\n> >> I slightly modified your query to reproduce this issue.\\n> >>\\n> >> explain analyze select * from r1 where value<555;\\n> >>\\n> >> Patch is attached to fix the problem.\\n> >\\n> > I forgot to mention the cause of the problem.\\n> >\\n> > if (istate->schunkptr < istate->nchunks)\\n> > {\\n> > PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\\n> > PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\n> > BlockNumber chunk_blockno;\\n> >\\n> > In above if condition we have only checked istate->schunkptr <\\n> > istate->nchunks that means we have some chunk left so we are safe to\\n> > access idxchunks, But just after that we are accessing\\n> > ptbase[idxpages[istate->spageptr]] without checking that accessing\\n> > idxpages is safe or not.\\n> >\\n> > tbm_iterator already handling this case, I broke it in\\n> tbm_shared_iterator.\\n>\\n> I don't know if this is the only problem -- it would be good if David\\n> could retest -- but it's certainly *a* problem, so committed.\\n>\\n\\nThanks for committing, and generally parallelising more stuff.\\n\\nI confirm that my test case is now working again.\\n\\nI'll be in this general area today, so will mention if I stumble over\\nanything that looks broken.\\n\\n-- \\n David Rowley http://www.2ndQuadrant.com/\\n PostgreSQL Development, 24x7 Support, Training & Services\\n\\nOn 10 March 2017 at 06:17, Robert Haas <robertmhaas@gmail.com> wrote:On Thu, Mar 9, 2017 at 11:50 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> On Thu, Mar 9, 2017 at 10:02 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n>> I slightly modified your query to reproduce this issue.\\n>>\\n>> explain analyze select * from r1 where value<555;\\n>>\\n>> Patch is attached to fix the problem.\\n>\\n> I forgot to mention the cause of the problem.\\n>\\n> if (istate->schunkptr < istate->nchunks)\\n> {\\n> PagetableEntry *chunk = &ptbase[idxchunks[istate->schunkptr]];\\n> PagetableEntry *page = &ptbase[idxpages[istate->spageptr]];\\n> BlockNumber chunk_blockno;\\n>\\n> In above if condition we have only checked istate->schunkptr <\\n> istate->nchunks that means we have some chunk left so we are safe to\\n> access idxchunks, But just after that we are accessing\\n> ptbase[idxpages[istate->spageptr]] without checking that accessing\\n> idxpages is safe or not.\\n>\\n> tbm_iterator already handling this case, I broke it in tbm_shared_iterator.\\n\\nI don't know if this is the only problem -- it would be good if David\\ncould retest -- but it's certainly *a* problem, so committed.Thanks for committing, and generally parallelising more stuff.I confirm that my test case is now working again.I'll be in this general area today, so will mention if I stumble over anything that looks broken. -- David Rowley http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services\",\n \"msg_date\": \"Fri, 10 Mar 2017 11:31:23 +1300\",\n \"msg_from\": \"David Rowley <david.rowley@2ndquadrant.com>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> I don't know if this is the only problem\\n\\n> I'll be in this general area today, so will mention if I stumble over\\n> anything that looks broken.\\n\\nI was testing the same patch with a large dataset and got a different segfault:\\n\\n> hasegeli=# explain select * from only mp_notification_20170225 where server_id = 7;\\n> QUERY PLAN\\n> ----------------------------------------------------------------------------------------------------------\\n> Gather (cost=26682.94..476995.88 rows=1 width=215)\\n> Workers Planned: 2\\n> -> Parallel Bitmap Heap Scan on mp_notification_20170225 (cost=25682.94..475995.78 rows=1 width=215)\\n> Recheck Cond: (server_id = 7)\\n> -> Bitmap Index Scan on mp_notification_block_idx (cost=0.00..25682.94 rows=4557665 width=0)\\n> Index Cond: (server_id = 7)\\n> (6 rows)\\n>\\n> hasegeli=# select * from only mp_notification_20170225 where server_id = 7;\\n> server closed the connection unexpectedly\\n> This probably means the server terminated abnormally\\n> before or while processing the request.\\n> The connection to the server was lost. Attempting reset: Failed.\\n\\n> * thread #1: tid = 0x5045a8f, 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414, queue = 'com.apple.main-thread', stop reason = signal SIGUSR1\\n> * frame #0: 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414 [opt]\\n> frame #1: 0x000000010ae4000c postgres`bringetbitmap(scan=0x00007fea3c875c20, tbm=<unavailable>) + 428 at brin.c:398 [opt]\\n> frame #2: 0x000000010ae9b451 postgres`index_getbitmap(scan=0x00007fea3c875c20, bitmap=<unavailable>) + 65 at indexam.c:726 [opt]\\n> frame #3: 0x000000010b0035a9 postgres`MultiExecBitmapIndexScan(node=<unavailable>) + 233 at nodeBitmapIndexscan.c:91 [opt]\\n> frame #4: 0x000000010b002840 postgres`BitmapHeapNext(node=<unavailable>) + 400 at nodeBitmapHeapscan.c:143 [opt]\\n> frame #5: 0x000000010afef5d0 postgres`ExecProcNode(node=0x00007fea3c873948) + 224 at execProcnode.c:459 [opt]\\n> frame #6: 0x000000010b004cc9 postgres`ExecGather [inlined] gather_getnext(gatherstate=<unavailable>) + 520 at nodeGather.c:276 [opt]\\n> frame #7: 0x000000010b004ac1 postgres`ExecGather(node=<unavailable>) + 497 at nodeGather.c:212 [opt]\\n> frame #8: 0x000000010afef6b2 postgres`ExecProcNode(node=0x00007fea3c872f58) + 450 at execProcnode.c:541 [opt]\\n> frame #9: 0x000000010afeaf90 postgres`standard_ExecutorRun [inlined] ExecutePlan(estate=<unavailable>, planstate=<unavailable>, use_parallel_mode=<unavailable>, operation=<unavailable>, numberTuples=0, direction=<unavailable>, dest=<unavailable>) + 29 at execMain.c:1616 [opt]\\n> frame #10: 0x000000010afeaf73 postgres`standard_ExecutorRun(queryDesc=<unavailable>, direction=<unavailable>, count=0) + 291 at execMain.c:348 [opt]\\n> frame #11: 0x000000010af8b108 postgres`ExplainOnePlan(plannedstmt=0x00007fea3c871040, into=0x0000000000000000, es=0x00007fea3c805360, queryString=<unavailable>, params=<unavailable>, planduration=<unavailable>) + 328 at explain.c:533 [opt]\\n> frame #12: 0x000000010af8ab98 postgres`ExplainOneQuery(query=0x00007fea3c805890, cursorOptions=<unavailable>, into=0x0000000000000000, es=0x00007fea3c805360, queryString=<unavailable>,params=0x0000000000000000) + 280 at explain.c:369 [opt]\\n> frame #13: 0x000000010af8a773 postgres`ExplainQuery(pstate=<unavailable>, stmt=0x00007fea3d005450, queryString=\\\"explain analyze select * from only mp_notification_20170225 where server_id > 6;\\\",params=0x0000000000000000, dest=0x00007fea3c8052c8) + 819 at explain.c:254 [opt]\\n> frame #14: 0x000000010b13b660 postgres`standard_ProcessUtility(pstmt=0x00007fea3d005fa8, queryString=\\\"explain analyze select * from only mp_notification_20170225 where server_id > 6;\\\",context=PROCESS_UTILITY_TOPLEVEL, params=0x0000000000000000, dest=0x00007fea3c8052c8, completionTag=<unavailable>) + 1104 at utility.c:675 [opt]\\n> frame #15: 0x000000010b13ad2a postgres`PortalRunUtility(portal=0x00007fea3c837640, pstmt=0x00007fea3d005fa8, isTopLevel='\\\\x01', setHoldSnapshot=<unavailable>, dest=0x00007fea3c8052c8, completionTag=<unavailable>) + 90 at pquery.c:1165 [opt]\\n> frame #16: 0x000000010b139f56 postgres`FillPortalStore(portal=0x00007fea3c837640, isTopLevel='\\\\x01') + 182 at pquery.c:1025 [opt]\\n> frame #17: 0x000000010b139c22 postgres`PortalRun(portal=0x00007fea3c837640, count=<unavailable>, isTopLevel='\\\\x01', dest=<unavailable>, altdest=<unavailable>, completionTag=<unavailable>) + 402 at pquery.c:757 [opt]\\n> frame #18: 0x000000010b13789b postgres`PostgresMain + 44 at postgres.c:1101 [opt]\\n> frame #19: 0x000000010b13786f postgres`PostgresMain(argc=<unavailable>, argv=<unavailable>, dbname=<unavailable>, username=<unavailable>) + 8927 at postgres.c:4066 [opt]\\n> frame #20: 0x000000010b0ba113 postgres`PostmasterMain [inlined] BackendRun + 7587 at postmaster.c:4317 [opt]\\n> frame #21: 0x000000010b0ba0e8 postgres`PostmasterMain [inlined] BackendStartup at postmaster.c:3989 [opt]\\n> frame #22: 0x000000010b0ba0e8 postgres`PostmasterMain at postmaster.c:1729 [opt]\\n> frame #23: 0x000000010b0ba0e8 postgres`PostmasterMain(argc=<unavailable>, argv=<unavailable>) + 7544 at postmaster.c:1337 [opt]\\n> frame #24: 0x000000010b0332af postgres`main(argc=<unavailable>, argv=<unavailable>) + 1567 at main.c:228 [opt]\\n> frame #25: 0x00007fffb4e28255 libdyld.dylib`start + 1\\n\\nI can try to provide a test case, if that wouldn't be enough to spot\\nthe problem.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 15:41:31 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 15, 2017 at 8:11 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>\\n> I can try to provide a test case, if that wouldn't be enough to spot\\n> the problem.\\n\\nThanks for reporting, I am looking into this. Meanwhile, if you can\\nprovide the reproducible test case then locating the issue will be\\nfaster.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 20:51:22 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 15, 2017 at 8:51 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n>> I can try to provide a test case, if that wouldn't be enough to spot\\n>> the problem.\\n>\\n> Thanks for reporting, I am looking into this. Meanwhile, if you can\\n> provide the reproducible test case then locating the issue will be\\n> faster.\\n\\nAfter trying multiple attempts with different datasets I am unable to\\nreproduce the issue.\\n\\nI tried with below test case:\\ncreate table t(a int, b varchar);\\n insert into t values(generate_series(1,10000000), repeat('x', 100));\\n insert into t values(generate_series(1,100000000), repeat('x', 100));\\ncreate index idx on t using brin(a);\\npostgres=# analyze t;\\nANALYZE\\npostgres=# explain analyze select * from t where a>6;\\n\\nQUERY PLAN\\n--------------------------------------------------------------------------------------------------------------------------------------------------\\n Gather (cost=580794.52..3059826.52 rows=110414922 width=105) (actual\\ntime=92.324..91853.716 rows=110425971 loops=1)\\n Workers Planned: 2\\n Workers Launched: 2\\n -> Parallel Bitmap Heap Scan on t (cost=579794.52..3058826.52\\nrows=46006218 width=105) (actual time=65.651..62023.020 rows=36808657\\nloops=3)\\n Recheck Cond: (a > 6)\\n Rows Removed by Index Recheck: 4\\n Heap Blocks: lossy=204401\\n -> Bitmap Index Scan on idx (cost=0.00..552190.79\\nrows=110425920 width=0) (actual time=88.215..88.215 rows=19040000\\nloops=1)\\n Index Cond: (a > 6)\\n Planning time: 1.116 ms\\n Execution time: 96176.881 ms\\n(11 rows)\\n\\nIs it possible for you to provide a reproducible test case? I also\\napplied the patch given up thread[1] but still could not reproduce.\\n\\n\\n[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 21:35:52 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 15, 2017 at 8:11 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> * thread #1: tid = 0x5045a8f, 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414, queue = 'com.apple.main-thread', stop reason = signal SIGUSR1\\n>> * frame #0: 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414 [opt]\\n>> frame #1: 0x000000010ae4000c postgres`bringetbitmap(scan=0x00007fea3c875c20, tbm=<unavailable>) + 428 at brin.c:398 [opt]\\n>> frame #2: 0x000000010ae9b451 postgres`index_getbitmap(scan=0x00007fea3c875c20, bitmap=<unavailable>) + 65 at indexam.c:726 [opt]\\n>> frame #3: 0x000000010b0035a9 postgres`MultiExecBitmapIndexScan(node=<unavailable>) + 233 at nodeBitmapIndexscan.c:91 [opt]\\n>> frame #4: 0x000000010b002840 postgres`BitmapHeapNext(node=<unavailable>) + 400 at nodeBitmapHeapscan.c:143 [opt]\\n\\nFurther analyzing the call stack, seems like this is not exact call\\nstack where it crashed. Because, if you notice the code in the\\nbrin_deform_tuple (line 414)\\n\\nbrin_deform_tuple(BrinDesc *brdesc, BrinTuple *tuple)\\n\\n{\\n dtup = brin_new_memtuple(brdesc);\\n\\n if (BrinTupleIsPlaceholder(tuple))\\n dtup->bt_placeholder = true;\\n dtup->bt_blkno = tuple->bt_blkno; --> line 414\\n\\nThis can crash at line:414, if either tuple is invalid memory(but I\\nthink it's not because we have already accessed this memory in above\\nif check) or dtup is invalid (this is also not possible because\\nbrin_new_memtuple has already accessed this).\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 21:53:08 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> This can crash at line:414, if either tuple is invalid memory(but I\\n> think it's not because we have already accessed this memory in above\\n> if check) or dtup is invalid (this is also not possible because\\n> brin_new_memtuple has already accessed this).\\n\\nI was testing with the brin correlation patch [1] applied. I cannot\\ncrash it without the patch either. I am sorry for not testing it\\nbefore. The patch make BRIN selectivity estimation function access\\nmore information.\\n\\n[1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 17:32:58 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 15, 2017 at 10:02 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n> I was testing with the brin correlation patch [1] applied. I cannot\\n> crash it without the patch either. I am sorry for not testing it\\n> before. The patch make BRIN selectivity estimation function access\\n> more information.\\n>\\n> [1] https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n\\nWith my test case, I could not crash even with this patch applied.\\nCan you provide your test case?\\n(table, index, data, query)\\n\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 22:05:59 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> With my test case, I could not crash even with this patch applied.\\n> Can you provide your test case?\\n\\nYes:\\n\\n> hasegeli=# create table r2 as select (random() * 3)::int as i from generate_series(1, 1000000);\\n> SELECT 1000000\\n> hasegeli=# create index on r2 using brin (i);\\n> CREATE INDEX\\n> hasegeli=# analyze r2;\\n> ANALYZE\\n> hasegeli=# explain select * from only r2 where i = 10;\\n> QUERY PLAN\\n> -------------------------------------------------------------------------------------\\n> Gather (cost=2867.50..9225.32 rows=1 width=4)\\n> Workers Planned: 2\\n> -> Parallel Bitmap Heap Scan on r2 (cost=1867.50..8225.22 rows=1 width=4)\\n> Recheck Cond: (i = 10)\\n> -> Bitmap Index Scan on r2_i_idx (cost=0.00..1867.50 rows=371082 width=0)\\n> Index Cond: (i = 10)\\n> (6 rows)\\n>\\n> hasegeli=# select * from only r2 where i = 10;\\n> server closed the connection unexpectedly\\n> This probably means the server terminated abnormally\\n> before or while processing the request.\\n> The connection to the server was lost. Attempting reset: Failed.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 17:51:23 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Wed, Mar 15, 2017 at 10:21 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> hasegeli=# create table r2 as select (random() * 3)::int as i from generate_series(1, 1000000);\\n>> SELECT 1000000\\n>> hasegeli=# create index on r2 using brin (i);\\n>> CREATE INDEX\\n>> hasegeli=# analyze r2;\\n>> ANALYZE\\n>> hasegeli=# explain select * from only r2 where i = 10;\\n>> QUERY PLAN\\n>> -------------------------------------------------------------------------------------\\n>> Gather (cost=2867.50..9225.32 rows=1 width=4)\\n>> Workers Planned: 2\\n>> -> Parallel Bitmap Heap Scan on r2 (cost=1867.50..8225.22 rows=1 width=4)\\n>> Recheck Cond: (i = 10)\\n>> -> Bitmap Index Scan on r2_i_idx (cost=0.00..1867.50 rows=371082 width=0)\\n>> Index Cond: (i = 10)\\n>> (6 rows)\\n>>\\n>> hasegeli=# select * from only r2 where i = 10;\\n\\nI am able to reproduce the bug, and attached patch fixes the same.\\nProblem is that I am not handling TBM_EMPTY state properly. I\\nremember that while reviewing the patch Robert mentioned that we might\\nneed to handle the TBM_EMPTY and I told that since we are not handling\\nin non-parallel mode so we don't need to handle here as well. But, I\\nwas wrong. So the problem is that if state is not TBM_HASH then it's\\ndirectly assuming TBM_ONE_PAGE which is completely wrong. I have\\nfixed that and also fixed in other similar locations.\\n\\nPlease verify the fix.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Wed, 15 Mar 2017 23:01:27 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> Please verify the fix.\\n\\nThe same test with both of the patches applied still crashes for me.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Wed, 15 Mar 2017 20:26:36 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 12:56 AM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> Please verify the fix.\\n>\\n> The same test with both of the patches applied still crashes for me.\\nAfter above fix, I am not able to reproduce. Can you give me the\\nbacktrace of the crash location or the dump?\\n\\nI am trying on the below commit\\n\\ncommit c5832346625af4193b1242e57e7d13e66a220b38\\nAuthor: Stephen Frost <sfrost@snowman.net>\\nDate: Wed Mar 15 11:19:39 2017 -0400\\n\\n+ https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n+ fix_tbm_empty.patch\\n\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 05:02:42 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 5:02 AM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> After above fix, I am not able to reproduce. Can you give me the\\n> backtrace of the crash location or the dump?\\n>\\n> I am trying on the below commit\\n>\\n> commit c5832346625af4193b1242e57e7d13e66a220b38\\n> Author: Stephen Frost <sfrost@snowman.net>\\n> Date: Wed Mar 15 11:19:39 2017 -0400\\n>\\n> + https://www.postgresql.org/message-id/attachment/50164/brin-correlation-v3.patch\\n> + fix_tbm_empty.patch\\n\\nForgot to mention after fix I am seeing this output.\\n\\npostgres=# explain analyze select * from only r2 where i = 10;\\n QUERY PLAN\\n-------------------------------------------------------------------------------------------------------------------------------\\n Gather (cost=2880.56..9251.98 rows=1 width=4) (actual\\ntime=3.857..3.857 rows=0 loops=1)\\n Workers Planned: 2\\n Workers Launched: 2\\n -> Parallel Bitmap Heap Scan on r2 (cost=1880.56..8251.88 rows=1\\nwidth=4) (actual time=0.043..0.043 rows=0 loops=3)\\n Recheck Cond: (i = 10)\\n -> Bitmap Index Scan on r2_i_idx (cost=0.00..1880.56\\nrows=373694 width=0) (actual time=0.052..0.052 rows=0 loops=1)\\n Index Cond: (i = 10)\\n Planning time: 0.111 ms\\n Execution time: 4.449 ms\\n(9 rows)\\n\\npostgres=# select * from only r2 where i = 10;\\n i\\n---\\n(0 rows)\\n\\nAre you getting the crash with the same test case?\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 05:11:17 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> Are you getting the crash with the same test case?\\n\\nYes. Here is the new backtrace:\\n\\n> * thread #1: tid = 0x51828fd, 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\\n> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\\n> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\\n> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\\n> frame #3: 0x0000000100caf303 postgres`tbm_prepare_shared_iterate(tbm=<unavailable>) + 723 at tidbitmap.c:875 [opt]\\n> frame #4: 0x0000000100c74844 postgres`BitmapHeapNext(node=<unavailable>) + 436 at nodeBitmapHeapscan.c:154 [opt]\\n> frame #5: 0x0000000100c615b0 postgres`ExecProcNode(node=0x00007fdabf8189f0) + 224 at execProcnode.c:459 [opt]\\n> frame #6: 0x0000000100c76ca9 postgres`ExecGather [inlined] gather_getnext(gatherstate=<unavailable>) + 520 at nodeGather.c:276 [opt]\\n> frame #7: 0x0000000100c76aa1 postgres`ExecGather(node=<unavailable>) + 497 at nodeGather.c:212 [opt]\\n> frame #8: 0x0000000100c61692 postgres`ExecProcNode(node=0x00007fdabf818558) + 450 at execProcnode.c:541 [opt]\\n> frame #9: 0x0000000100c5cf70 postgres`standard_ExecutorRun [inlined] ExecutePlan(estate=<unavailable>, planstate=<unavailable>, use_parallel_mode=<unavailable>, operation=<unavailable>, numberTuples=0, direction=<unavailable>, dest=<unavailable>) + 29 at execMain.c:1616 [opt]\\n> frame #10: 0x0000000100c5cf53 postgres`standard_ExecutorRun(queryDesc=<unavailable>, direction=<unavailable>, count=0) + 291 at execMain.c:348 [opt]\\n> frame #11: 0x0000000100dac0df postgres`PortalRunSelect(portal=0x00007fdac000b240, forward=<unavailable>, count=0, dest=<unavailable>) + 255 at pquery.c:921 [opt]\\n> frame #12: 0x0000000100dabc84 postgres`PortalRun(portal=0x00007fdac000b240, count=<unavailable>, isTopLevel='\\\\x01', dest=<unavailable>, altdest=<unavailable>, completionTag=<unavailable>) + 500 at pquery.c:762 [opt]\\n> frame #13: 0x0000000100da989b postgres`PostgresMain + 44 at postgres.c:1101 [opt]\\n> frame #14: 0x0000000100da986f postgres`PostgresMain(argc=<unavailable>, argv=<unavailable>, dbname=<unavailable>, username=<unavailable>) + 8927 at postgres.c:4066 [opt]\\n> frame #15: 0x0000000100d2c113 postgres`PostmasterMain [inlined] BackendRun + 7587 at postmaster.c:4317 [opt]\\n> frame #16: 0x0000000100d2c0e8 postgres`PostmasterMain [inlined] BackendStartup at postmaster.c:3989 [opt]\\n> frame #17: 0x0000000100d2c0e8 postgres`PostmasterMain at postmaster.c:1729 [opt]\\n> frame #18: 0x0000000100d2c0e8 postgres`PostmasterMain(argc=<unavailable>, argv=<unavailable>) + 7544 at postmaster.c:1337 [opt]\\n> frame #19: 0x0000000100ca528f postgres`main(argc=<unavailable>, argv=<unavailable>) + 1567 at main.c:228 [opt]\\n> frame #20: 0x00007fffb4e28255 libdyld.dylib`start + 1\\n> frame #21: 0x00007fffb4e28255 libdyld.dylib`start + 1\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 11:22:02 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 3:52 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> * thread #1: tid = 0x51828fd, 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\\n>> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\\n>> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\\n>> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\\n\\nBy looking at the call stack I got the problem location. I am\\nreviewing other parts of the code if there are the similar mistake at\\nother places. Soon I will post the patch. Thanks for the help.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 17:14:46 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 5:14 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> pg_atomic_write_u32_impl(val=0) at generic.h:57, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)\\n>>> * frame #0: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_write_u32_impl(val=0) at generic.h:57 [opt]\\n>>> frame #1: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32_impl(val_=0) at generic.h:163 [opt]\\n>>> frame #2: 0x0000000100caf314 postgres`tbm_prepare_shared_iterate [inlined] pg_atomic_init_u32(val=0) + 17 at atomics.h:237 [opt]\\n>\\n> By looking at the call stack I got the problem location. I am\\n> reviewing other parts of the code if there are the similar mistake at\\n> other places. Soon I will post the patch. Thanks for the help.\\n\\nBased on the call stack I have tried to fix the issue. The problem is\\nthere was some uninitialized pointer access (in some special cases\\ni.e. TBM_EMPTY when pagetable is not created at all).\\n\\n fix_tbm_empty.patch have fixed some of them but induced one which you\\nare seeing in your call stack.\\n\\nHopefully, this time I got it correct. Since I am unable to reproduce\\nthe issue so I will again need your help in verifying the fix.\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 16 Mar 2017 17:56:46 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"> Hopefully, this time I got it correct. Since I am unable to reproduce\\n> the issue so I will again need your help in verifying the fix.\\n\\nIt is not crashing with the new patch. Thank you.\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 15:56:27 +0100\",\n \"msg_from\": \"Emre Hasegeli <emre@hasegeli.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 10:56 AM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> Hopefully, this time I got it correct. Since I am unable to reproduce\\n>> the issue so I will again need your help in verifying the fix.\\n>\\n> It is not crashing with the new patch. Thank you.\\n\\nThanks for confirming. Some review comments on v2:\\n\\n+ if (istate->pagetable)\\n\\nPlease compare explicitly to InvalidDsaPointer.\\n\\n+ if (iterator->ptbase)\\n+ ptbase = iterator->ptbase->ptentry;\\n+ if (iterator->ptpages)\\n+ idxpages = iterator->ptpages->index;\\n+ if (iterator->ptchunks)\\n+ idxchunks = iterator->ptchunks->index;\\n\\nSimilarly.\\n\\nDilip, please also provide a proposed commit message describing what\\nthis is fixing. Is it just the TBM_EMPTY case, or is there anything\\nelse?\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 11:12:36 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 8:42 PM, Robert Haas <robertmhaas@gmail.com> wrote:\\n> Thanks for confirming. Some review comments on v2:\\n>\\n> + if (istate->pagetable)\\nfixed\\n>\\n> Please compare explicitly to InvalidDsaPointer.\\n>\\n> + if (iterator->ptbase)\\n> + ptbase = iterator->ptbase->ptentry;\\n> + if (iterator->ptpages)\\n> + idxpages = iterator->ptpages->index;\\n> + if (iterator->ptchunks)\\n> + idxchunks = iterator->ptchunks->index;\\n>\\n> Similarly.\\nfixed\\n\\nAlso fixed at\\n+ if (ptbase)\\n+ pg_atomic_init_u32(&ptbase->refcount, 0);\\n\\n>\\n> Dilip, please also provide a proposed commit message describing what\\n> this is fixing. Is it just the TBM_EMPTY case, or is there anything\\n> else?\\n\\nOkay, I have added the commit message in the patch.\\n\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 16 Mar 2017 23:20:15 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 8:26 PM, Emre Hasegeli <emre@hasegeli.com> wrote:\\n>> Hopefully, this time I got it correct. Since I am unable to reproduce\\n>> the issue so I will again need your help in verifying the fix.\\n>\\n> It is not crashing with the new patch. Thank you.\\n\\nThanks for verifying.\\n\\n\\n-- \\nRegards,\\nDilip Kumar\\nEnterpriseDB: http://www.enterprisedb.com\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 23:20:42 +0530\",\n \"msg_from\": \"Dilip Kumar <dilipbalaut@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n },\n {\n \"msg_contents\": \"On Thu, Mar 16, 2017 at 1:50 PM, Dilip Kumar <dilipbalaut@gmail.com> wrote:\\n> fixed\\n\\nCommitted.\\n\\n-- \\nRobert Haas\\nEnterpriseDB: http://www.enterprisedb.com\\nThe Enterprise PostgreSQL Company\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 16 Mar 2017 15:10:48 -0400\",\n \"msg_from\": \"Robert Haas <robertmhaas@gmail.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Parallel Bitmap scans a bit broken\"\n }\n]","isConversational":false}}},{"rowIdx":44098,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Greetings,\n\nConsider this:\n\ncreate table t10 (c1 int, c2 int);\ncreate index on t10 (c1) where c2 > 5;\n\n\\d t10\n Table \"sfrost.t10\"\n Column | Type | Modifiers \n--------+---------+-----------\n c1 | integer | \n c2 | integer | \nIndexes:\n \"t10_c1_idx\" btree (c1) WHERE c2 > 5\n\ninsert into t10 select * from generate_series(1,10000) a, generate_series(1,10) b;\n(repeat a bunch of times, if desired)\n\nvacuum analyze t10;\nset work_mem = '64kB';\nset enable_indexscan = false;\nset enable_seqscan = false;\n\n=*> explain analyze select * from t10 where c2 > 6;\n QUERY PLAN \n----------------------------------------------------------------------------------------------------------------------------------\n Bitmap Heap Scan on t10 (cost=6496.49..15037.50 rows=318653 width=8) (actual time=34.682..116.236 rows=320000 loops=1)\n Recheck Cond: (c2 > 5)\n Rows Removed by Index Recheck: 327502\n Filter: (c2 > 6)\n Rows Removed by Filter: 80000\n Heap Blocks: exact=642 lossy=2898\n -> Bitmap Index Scan on t10_c1_idx (cost=0.00..6416.83 rows=400081 width=0) (actual time=34.601..34.601 rows=400000 loops=1)\n Planning time: 0.087 ms\n Execution time: 124.229 ms\n(9 rows)\n\nPerhaps I'm missing something obvious, but isn't it a bit redundant to\nhave both a Recheck condition (which is the predicate of the index) and\na Filter condition (which is the user's predicate) when we've already\ndecided that the user's predicate must result in a subset of the\nindex's, as, otherwise, we wouldn't be able to use the index in the\nfirst place?\n\nIn other words, it seems like we shouldn't need a Filter in the above\nBitmap Heap Scan, instead we should just make the Recheck be (c2 > 6).\n\nI've not looked into the code side of this at all and there may be\nreasons why this is hard to do, but it seems like a worthwhile\nimprovement to consider doing, though perhaps I'm missing some reason\nwhy we need both the Recheck and the Filter in such cases for\ncorrectness.\n\nThanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 11:01:21 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"partial indexes and bitmap scans"},{"msg_contents":"Stephen Frost <sfrost@snowman.net> writes:\n> Perhaps I'm missing something obvious, but isn't it a bit redundant to\n> have both a Recheck condition (which is the predicate of the index) and\n> a Filter condition (which is the user's predicate) when we've already\n> decided that the user's predicate must result in a subset of the\n> index's, as, otherwise, we wouldn't be able to use the index in the\n> first place?\n\nYeah, I think this is just something that the planner doesn't see fit\nto expend cycles on detecting.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 09 Mar 2017 11:44:24 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partial indexes and bitmap scans"},{"msg_contents":"Tom,\n\n* Tom Lane (tgl@sss.pgh.pa.us) wrote:\n> Stephen Frost <sfrost@snowman.net> writes:\n> > Perhaps I'm missing something obvious, but isn't it a bit redundant to\n> > have both a Recheck condition (which is the predicate of the index) and\n> > a Filter condition (which is the user's predicate) when we've already\n> > decided that the user's predicate must result in a subset of the\n> > index's, as, otherwise, we wouldn't be able to use the index in the\n> > first place?\n> \n> Yeah, I think this is just something that the planner doesn't see fit\n> to expend cycles on detecting.\n\nWe have already figured out that the user's predicate results in a\nsubset of the index's or we wouldn't be able to use that index though,\nright? Do we really need to spend cycles re-discovering that? Are\nthere cases where we actually need the index's predicate to ever be\nincluded for correctness..?\n\nThis seems like we're going out of our way to add in an additional check\nfor something that we've already determined must always be true and that\nstrikes me as odd.\n\nThanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 12:01:24 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"Re: partial indexes and bitmap scans"},{"msg_contents":"Stephen Frost <sfrost@snowman.net> writes:\n> We have already figured out that the user's predicate results in a\n> subset of the index's or we wouldn't be able to use that index though,\n> right? Do we really need to spend cycles re-discovering that? Are\n> there cases where we actually need the index's predicate to ever be\n> included for correctness..?\n\nIn a bitmap indexscan, yes, because the entire point of the recheck\ncondition is that we're going to scan a whole page and possibly see\ntuples that don't satisfy the index predicate at all. Another point\nthat's mentioned in the comments in createplan.c is that if you're\nconsidering the result of a BitmapOr or BitmapAnd, there's not necessarily\na single index involved so it's much harder to reason about which part\nof the qual is an index predicate.\n\nI do notice that createplan.c makes some effort to get rid of filter\nconditions that are provably true when the index conditions are.\nDoesn't look like it considers the reverse direction. Not sure if\nthat's missing a bet.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 09 Mar 2017 12:25:25 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partial indexes and bitmap scans"},{"msg_contents":"Tom,\n\n* Tom Lane (tgl@sss.pgh.pa.us) wrote:\n> Stephen Frost <sfrost@snowman.net> writes:\n> > We have already figured out that the user's predicate results in a\n> > subset of the index's or we wouldn't be able to use that index though,\n> > right? Do we really need to spend cycles re-discovering that? Are\n> > there cases where we actually need the index's predicate to ever be\n> > included for correctness..?\n> \n> In a bitmap indexscan, yes, because the entire point of the recheck\n> condition is that we're going to scan a whole page and possibly see\n> tuples that don't satisfy the index predicate at all.\n\nI understand the point of the recheck condition being that we might see\ntuples that don't satisfy either the index predicate or the user's\npredicate, but that isn't the point- to even consider using that index\nwe must have already realized that the user's predicate will only be\nsatisfied in a subset of cases when the index predicate predicate will\nbe satisfied, making any check of the index predicate necessairly always\ntrue.\n\n> Another point\n> that's mentioned in the comments in createplan.c is that if you're\n> considering the result of a BitmapOr or BitmapAnd, there's not necessarily\n> a single index involved so it's much harder to reason about which part\n> of the qual is an index predicate.\n\nWe must be pulling the index's predicate to be able to put it into the\nRecheck condition in the first place. What I'm arguing is that once\nwe've decided that we're able to use an index X, because the values\nwhich the user's predicate will satisfy is a subset of those which\nthe index's predicate will satisfy, then we can entirely forget about\nthe index's predicate as being redundant to the user's.\n\nI don't see anything about a BitmapAnd or BitmapOr being relevant to\nthat. We will always need to check the user's predicate against the\ntuples being returned from the Bitmap Heap Scan, unless by chance it\nmatches the index's predicate exactly *and* we have an entirely exact\nmatch bitmap without any lossy parts.\n\n> I do notice that createplan.c makes some effort to get rid of filter\n> conditions that are provably true when the index conditions are.\n> Doesn't look like it considers the reverse direction. Not sure if\n> that's missing a bet.\n\nThat actually strikes me as a less likely condition to have in the\ngeneral case, isn't it? Wouldn't that only happen if the index\npredicate and the user predicate are identical, because otherwise we\neither can't use the index or we must keep the user's predicate because\nit will only be satisfied in a subset of cases?\n\nThanks!\n\nStephen","msg_date":"Thu, 9 Mar 2017 12:42:13 -0500","msg_from":"Stephen Frost <sfrost@snowman.net>","msg_from_op":true,"msg_subject":"Re: partial indexes and bitmap scans"},{"msg_contents":"Stephen Frost <sfrost@snowman.net> writes:\n> * Tom Lane (tgl@sss.pgh.pa.us) wrote:\n>> I do notice that createplan.c makes some effort to get rid of filter\n>> conditions that are provably true when the index conditions are.\n>> Doesn't look like it considers the reverse direction. Not sure if\n>> that's missing a bet.\n\n> That actually strikes me as a less likely condition to have in the\n> general case, isn't it? Wouldn't that only happen if the index\n> predicate and the user predicate are identical, because otherwise we\n> either can't use the index or we must keep the user's predicate because\n> it will only be satisfied in a subset of cases?\n\nWell, remember that the planner's idea of an ideal situation is to not\nhave any filter conditions, not to not have any index (a/k/a recheck)\nconditions. It's going to try to put as much load as it can on the index\ncondition side of things, and that gives rise to the need for rechecks.\n\nIt seems like there might be some mileage to be gained by reversing the\nproof direction here, and having it get rid of recheck conditions that are\nimplied by filter conditions rather than vice versa. I'm not quite\nconvinced though, and I'm also not sure how hard it would be to mechanize.\nA lot of that code is shared between the bitmap and plain-indexscan cases,\nwhich could make it tricky to not break the plain-indexscan case.\n\n\t\t\tregards, tom lane\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Thu, 09 Mar 2017 13:04:31 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":false,"msg_subject":"Re: partial indexes and bitmap scans"}],"string":"[\n {\n \"msg_contents\": \"Greetings,\\n\\nConsider this:\\n\\ncreate table t10 (c1 int, c2 int);\\ncreate index on t10 (c1) where c2 > 5;\\n\\n\\\\d t10\\n Table \\\"sfrost.t10\\\"\\n Column | Type | Modifiers \\n--------+---------+-----------\\n c1 | integer | \\n c2 | integer | \\nIndexes:\\n \\\"t10_c1_idx\\\" btree (c1) WHERE c2 > 5\\n\\ninsert into t10 select * from generate_series(1,10000) a, generate_series(1,10) b;\\n(repeat a bunch of times, if desired)\\n\\nvacuum analyze t10;\\nset work_mem = '64kB';\\nset enable_indexscan = false;\\nset enable_seqscan = false;\\n\\n=*> explain analyze select * from t10 where c2 > 6;\\n QUERY PLAN \\n----------------------------------------------------------------------------------------------------------------------------------\\n Bitmap Heap Scan on t10 (cost=6496.49..15037.50 rows=318653 width=8) (actual time=34.682..116.236 rows=320000 loops=1)\\n Recheck Cond: (c2 > 5)\\n Rows Removed by Index Recheck: 327502\\n Filter: (c2 > 6)\\n Rows Removed by Filter: 80000\\n Heap Blocks: exact=642 lossy=2898\\n -> Bitmap Index Scan on t10_c1_idx (cost=0.00..6416.83 rows=400081 width=0) (actual time=34.601..34.601 rows=400000 loops=1)\\n Planning time: 0.087 ms\\n Execution time: 124.229 ms\\n(9 rows)\\n\\nPerhaps I'm missing something obvious, but isn't it a bit redundant to\\nhave both a Recheck condition (which is the predicate of the index) and\\na Filter condition (which is the user's predicate) when we've already\\ndecided that the user's predicate must result in a subset of the\\nindex's, as, otherwise, we wouldn't be able to use the index in the\\nfirst place?\\n\\nIn other words, it seems like we shouldn't need a Filter in the above\\nBitmap Heap Scan, instead we should just make the Recheck be (c2 > 6).\\n\\nI've not looked into the code side of this at all and there may be\\nreasons why this is hard to do, but it seems like a worthwhile\\nimprovement to consider doing, though perhaps I'm missing some reason\\nwhy we need both the Recheck and the Filter in such cases for\\ncorrectness.\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 11:01:21 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"partial indexes and bitmap scans\"\n },\n {\n \"msg_contents\": \"Stephen Frost <sfrost@snowman.net> writes:\\n> Perhaps I'm missing something obvious, but isn't it a bit redundant to\\n> have both a Recheck condition (which is the predicate of the index) and\\n> a Filter condition (which is the user's predicate) when we've already\\n> decided that the user's predicate must result in a subset of the\\n> index's, as, otherwise, we wouldn't be able to use the index in the\\n> first place?\\n\\nYeah, I think this is just something that the planner doesn't see fit\\nto expend cycles on detecting.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 09 Mar 2017 11:44:24 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partial indexes and bitmap scans\"\n },\n {\n \"msg_contents\": \"Tom,\\n\\n* Tom Lane (tgl@sss.pgh.pa.us) wrote:\\n> Stephen Frost <sfrost@snowman.net> writes:\\n> > Perhaps I'm missing something obvious, but isn't it a bit redundant to\\n> > have both a Recheck condition (which is the predicate of the index) and\\n> > a Filter condition (which is the user's predicate) when we've already\\n> > decided that the user's predicate must result in a subset of the\\n> > index's, as, otherwise, we wouldn't be able to use the index in the\\n> > first place?\\n> \\n> Yeah, I think this is just something that the planner doesn't see fit\\n> to expend cycles on detecting.\\n\\nWe have already figured out that the user's predicate results in a\\nsubset of the index's or we wouldn't be able to use that index though,\\nright? Do we really need to spend cycles re-discovering that? Are\\nthere cases where we actually need the index's predicate to ever be\\nincluded for correctness..?\\n\\nThis seems like we're going out of our way to add in an additional check\\nfor something that we've already determined must always be true and that\\nstrikes me as odd.\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 12:01:24 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: partial indexes and bitmap scans\"\n },\n {\n \"msg_contents\": \"Stephen Frost <sfrost@snowman.net> writes:\\n> We have already figured out that the user's predicate results in a\\n> subset of the index's or we wouldn't be able to use that index though,\\n> right? Do we really need to spend cycles re-discovering that? Are\\n> there cases where we actually need the index's predicate to ever be\\n> included for correctness..?\\n\\nIn a bitmap indexscan, yes, because the entire point of the recheck\\ncondition is that we're going to scan a whole page and possibly see\\ntuples that don't satisfy the index predicate at all. Another point\\nthat's mentioned in the comments in createplan.c is that if you're\\nconsidering the result of a BitmapOr or BitmapAnd, there's not necessarily\\na single index involved so it's much harder to reason about which part\\nof the qual is an index predicate.\\n\\nI do notice that createplan.c makes some effort to get rid of filter\\nconditions that are provably true when the index conditions are.\\nDoesn't look like it considers the reverse direction. Not sure if\\nthat's missing a bet.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 09 Mar 2017 12:25:25 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partial indexes and bitmap scans\"\n },\n {\n \"msg_contents\": \"Tom,\\n\\n* Tom Lane (tgl@sss.pgh.pa.us) wrote:\\n> Stephen Frost <sfrost@snowman.net> writes:\\n> > We have already figured out that the user's predicate results in a\\n> > subset of the index's or we wouldn't be able to use that index though,\\n> > right? Do we really need to spend cycles re-discovering that? Are\\n> > there cases where we actually need the index's predicate to ever be\\n> > included for correctness..?\\n> \\n> In a bitmap indexscan, yes, because the entire point of the recheck\\n> condition is that we're going to scan a whole page and possibly see\\n> tuples that don't satisfy the index predicate at all.\\n\\nI understand the point of the recheck condition being that we might see\\ntuples that don't satisfy either the index predicate or the user's\\npredicate, but that isn't the point- to even consider using that index\\nwe must have already realized that the user's predicate will only be\\nsatisfied in a subset of cases when the index predicate predicate will\\nbe satisfied, making any check of the index predicate necessairly always\\ntrue.\\n\\n> Another point\\n> that's mentioned in the comments in createplan.c is that if you're\\n> considering the result of a BitmapOr or BitmapAnd, there's not necessarily\\n> a single index involved so it's much harder to reason about which part\\n> of the qual is an index predicate.\\n\\nWe must be pulling the index's predicate to be able to put it into the\\nRecheck condition in the first place. What I'm arguing is that once\\nwe've decided that we're able to use an index X, because the values\\nwhich the user's predicate will satisfy is a subset of those which\\nthe index's predicate will satisfy, then we can entirely forget about\\nthe index's predicate as being redundant to the user's.\\n\\nI don't see anything about a BitmapAnd or BitmapOr being relevant to\\nthat. We will always need to check the user's predicate against the\\ntuples being returned from the Bitmap Heap Scan, unless by chance it\\nmatches the index's predicate exactly *and* we have an entirely exact\\nmatch bitmap without any lossy parts.\\n\\n> I do notice that createplan.c makes some effort to get rid of filter\\n> conditions that are provably true when the index conditions are.\\n> Doesn't look like it considers the reverse direction. Not sure if\\n> that's missing a bet.\\n\\nThat actually strikes me as a less likely condition to have in the\\ngeneral case, isn't it? Wouldn't that only happen if the index\\npredicate and the user predicate are identical, because otherwise we\\neither can't use the index or we must keep the user's predicate because\\nit will only be satisfied in a subset of cases?\\n\\nThanks!\\n\\nStephen\",\n \"msg_date\": \"Thu, 9 Mar 2017 12:42:13 -0500\",\n \"msg_from\": \"Stephen Frost <sfrost@snowman.net>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: partial indexes and bitmap scans\"\n },\n {\n \"msg_contents\": \"Stephen Frost <sfrost@snowman.net> writes:\\n> * Tom Lane (tgl@sss.pgh.pa.us) wrote:\\n>> I do notice that createplan.c makes some effort to get rid of filter\\n>> conditions that are provably true when the index conditions are.\\n>> Doesn't look like it considers the reverse direction. Not sure if\\n>> that's missing a bet.\\n\\n> That actually strikes me as a less likely condition to have in the\\n> general case, isn't it? Wouldn't that only happen if the index\\n> predicate and the user predicate are identical, because otherwise we\\n> either can't use the index or we must keep the user's predicate because\\n> it will only be satisfied in a subset of cases?\\n\\nWell, remember that the planner's idea of an ideal situation is to not\\nhave any filter conditions, not to not have any index (a/k/a recheck)\\nconditions. It's going to try to put as much load as it can on the index\\ncondition side of things, and that gives rise to the need for rechecks.\\n\\nIt seems like there might be some mileage to be gained by reversing the\\nproof direction here, and having it get rid of recheck conditions that are\\nimplied by filter conditions rather than vice versa. I'm not quite\\nconvinced though, and I'm also not sure how hard it would be to mechanize.\\nA lot of that code is shared between the bitmap and plain-indexscan cases,\\nwhich could make it tricky to not break the plain-indexscan case.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Thu, 09 Mar 2017 13:04:31 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: partial indexes and bitmap scans\"\n }\n]","isConversational":false}}},{"rowIdx":44099,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"I typically build with --with-system-tzdata, which means that any time\nRed Hat sees fit to push out a new copy of the tzdata package, that's\nwhat I'm testing against. This morning they updated to tzdata2017a,\nand I'm seeing the attached test failures.\n\nThe reason for this is that the IANA crew have really moved forward\naggressively on their project to remove invented zone abbreviations.\nAs stated at\nhttp://mm.icann.org/pipermail/tz-announce/2017-February/000045.html\nthey've removed text abbreviations for pretty much all of South\nAmerica, which is what's breaking these cases.\n\nFor the cases involving the America/Santiago zone, I'm a bit inclined\nto just switch that to America/New_York, which seems much less likely\nto get fooled with by IANA. But I'm wondering if Alvaro had a specific\nreason for using the Santiago zone in those test cases.\n\nThe diffs involving VET (America/Caracas) are more problematic: we're\nintentionally trying to test a time-varying zone abbreviation, so we can't\njust switch to a stable zone. As far as the lines 2395..2437 hunk goes,\nwe could perhaps make that segment of the test print in ISO datestyle\ninstead of Postgres datestyle, which would force the zone to be printed in\nnumeric form not as an abbreviation. That's slightly annoying, but it\ndoesn't really compromise what that part of the test is trying to test.\nHowever, the very last hunk indicates that VET no longer functions as a\ntime-varying zone abbreviation at all, because it doesn't match any\nabbreviation recorded for America/Caracas in the IANA database, so we fall\nback to treating it as a simple synonym for America/Caracas (cf commit\n39b691f25). So that's now failing altogether to test what it means to.\n\nWhat I propose we do about that is replace the America/Caracas test cases\nwith Europe/Moscow tests, moving the dates as needed to match DST\ntransitions from when Moscow was observing DST (pre 2011). The comments\nin the IANA files indicate that they believe the MSK/MSD abbreviations\nhave or had real-world usage, so they probably won't replace them with\nnumeric offsets. We can hope, anyway.\n\n\t\t\tregards, tom lane\n\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers","msg_date":"Thu, 09 Mar 2017 14:24:30 -0500","msg_from":"Tom Lane <tgl@sss.pgh.pa.us>","msg_from_op":true,"msg_subject":"tzdata2017a breaks timestamptz regression test"},{"msg_contents":"Tom Lane wrote:\n\n> For the cases involving the America/Santiago zone, I'm a bit inclined\n> to just switch that to America/New_York, which seems much less likely\n> to get fooled with by IANA. But I'm wondering if Alvaro had a specific\n> reason for using the Santiago zone in those test cases.\n\nNo, I just used the Santiago zone because it was local. But I agree\nit's a bad one to use, because in the last decade politicians have been\nplaying with it excessively, so +1 for changing to something else.\n\n> What I propose we do about that is replace the America/Caracas test cases\n> with Europe/Moscow tests, moving the dates as needed to match DST\n> transitions from when Moscow was observing DST (pre 2011). The comments\n> in the IANA files indicate that they believe the MSK/MSD abbreviations\n> have or had real-world usage, so they probably won't replace them with\n> numeric offsets. We can hope, anyway.\n\nSounds reasonable to me.\n\n-- \n�lvaro Herrera https://www.2ndQuadrant.com/\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription:\nhttp://www.postgresql.org/mailpref/pgsql-hackers\n","msg_date":"Fri, 10 Mar 2017 10:53:29 -0300","msg_from":"Alvaro Herrera <alvherre@2ndquadrant.com>","msg_from_op":false,"msg_subject":"Re: tzdata2017a breaks timestamptz regression test"}],"string":"[\n {\n \"msg_contents\": \"I typically build with --with-system-tzdata, which means that any time\\nRed Hat sees fit to push out a new copy of the tzdata package, that's\\nwhat I'm testing against. This morning they updated to tzdata2017a,\\nand I'm seeing the attached test failures.\\n\\nThe reason for this is that the IANA crew have really moved forward\\naggressively on their project to remove invented zone abbreviations.\\nAs stated at\\nhttp://mm.icann.org/pipermail/tz-announce/2017-February/000045.html\\nthey've removed text abbreviations for pretty much all of South\\nAmerica, which is what's breaking these cases.\\n\\nFor the cases involving the America/Santiago zone, I'm a bit inclined\\nto just switch that to America/New_York, which seems much less likely\\nto get fooled with by IANA. But I'm wondering if Alvaro had a specific\\nreason for using the Santiago zone in those test cases.\\n\\nThe diffs involving VET (America/Caracas) are more problematic: we're\\nintentionally trying to test a time-varying zone abbreviation, so we can't\\njust switch to a stable zone. As far as the lines 2395..2437 hunk goes,\\nwe could perhaps make that segment of the test print in ISO datestyle\\ninstead of Postgres datestyle, which would force the zone to be printed in\\nnumeric form not as an abbreviation. That's slightly annoying, but it\\ndoesn't really compromise what that part of the test is trying to test.\\nHowever, the very last hunk indicates that VET no longer functions as a\\ntime-varying zone abbreviation at all, because it doesn't match any\\nabbreviation recorded for America/Caracas in the IANA database, so we fall\\nback to treating it as a simple synonym for America/Caracas (cf commit\\n39b691f25). So that's now failing altogether to test what it means to.\\n\\nWhat I propose we do about that is replace the America/Caracas test cases\\nwith Europe/Moscow tests, moving the dates as needed to match DST\\ntransitions from when Moscow was observing DST (pre 2011). The comments\\nin the IANA files indicate that they believe the MSK/MSD abbreviations\\nhave or had real-world usage, so they probably won't replace them with\\nnumeric offsets. We can hope, anyway.\\n\\n\\t\\t\\tregards, tom lane\\n\\n\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\",\n \"msg_date\": \"Thu, 09 Mar 2017 14:24:30 -0500\",\n \"msg_from\": \"Tom Lane <tgl@sss.pgh.pa.us>\",\n \"msg_from_op\": true,\n \"msg_subject\": \"tzdata2017a breaks timestamptz regression test\"\n },\n {\n \"msg_contents\": \"Tom Lane wrote:\\n\\n> For the cases involving the America/Santiago zone, I'm a bit inclined\\n> to just switch that to America/New_York, which seems much less likely\\n> to get fooled with by IANA. But I'm wondering if Alvaro had a specific\\n> reason for using the Santiago zone in those test cases.\\n\\nNo, I just used the Santiago zone because it was local. But I agree\\nit's a bad one to use, because in the last decade politicians have been\\nplaying with it excessively, so +1 for changing to something else.\\n\\n> What I propose we do about that is replace the America/Caracas test cases\\n> with Europe/Moscow tests, moving the dates as needed to match DST\\n> transitions from when Moscow was observing DST (pre 2011). The comments\\n> in the IANA files indicate that they believe the MSK/MSD abbreviations\\n> have or had real-world usage, so they probably won't replace them with\\n> numeric offsets. We can hope, anyway.\\n\\nSounds reasonable to me.\\n\\n-- \\n�lvaro Herrera https://www.2ndQuadrant.com/\\nPostgreSQL Development, 24x7 Support, Remote DBA, Training & Services\\n\\n\\n-- \\nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\\nTo make changes to your subscription:\\nhttp://www.postgresql.org/mailpref/pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 10 Mar 2017 10:53:29 -0300\",\n \"msg_from\": \"Alvaro Herrera <alvherre@2ndquadrant.com>\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: tzdata2017a breaks timestamptz regression test\"\n }\n]","isConversational":false}}}],"truncated":false,"partial":false},"paginationData":{"pageIndex":440,"numItemsPerPage":100,"numTotalItems":58371,"offset":44000,"length":100}},"jwt":"eyJhbGciOiJFZERTQSIsImtpZCI6IjVHZDBvd0g5MTM2eDZjc1FvbE1zcktNWUZoRVFoUm5PVVVybEpjOUhaUEEifQ.eyJyZWFkIjp0cnVlLCJwZXJtaXNzaW9ucyI6eyJyZXBvLmNvbnRlbnQucmVhZCI6dHJ1ZX0sImlhdCI6MTc4NTQ5MjE2OSwianRpIjoiYmFkYmZjMmQtYTRjZi00MTEyLTk5YmItMWJkOTFkYzk1NjNmIiwic3ViIjoiL2RhdGFzZXRzL3dhbnNoZW5sL3Bnc3FsLWhhY2tlcnMiLCJleHAiOjE3ODU0OTU3NjksImlzcyI6Imh0dHBzOi8vaHVnZ2luZ2ZhY2UuY28ifQ.sRWfgw-wMF9SY4HyVtj8vgPowM0Uw31-TjnA4JjrfOy5myJwtXRnSNJ9LlqqIHS-qdKMwQXj_omNTqEO7TZkDg","displayUrls":true,"splitSizeSummaries":[{"config":"default","split":"train","numRows":58371,"numBytesParquet":368254123}]},"discussionsStats":{"closed":0,"open":0,"total":0},"fullWidth":true,"hasGatedAccess":true,"hasFullAccess":true,"isEmbedded":false,"savedQueries":{"community":[],"user":[]}}">
threads listlengths 1 2.99k |
|---|
[
{
"msg_contents": "Hi all,\n\nThe autovacuum reports the number of skipped frozen pages to the\nVACUUM output. But these information is not appeared by manual VACUUM.\nThis information is useful for the user to check efficiency of VACUUM.\n\nAttached patch add this information to VACUUM output.\n\n* Example\n=#... |
[
{
"msg_contents": "Currently bytea_output supports values of 'hex' (the default) and\n'escape' (octal). 'hex' uses two characters per byte, while escape uses\nthree (ignoring the prefix overhead of \\x or \\[0-9].)\n\nIt is my understanding that base64 uses 1.37 characters per byte:\n\n\thttps://en.wikipedia.o... |
[
{
"msg_contents": "When doing a hash join with large work_mem, you can have a large number of\nchunks. Then if ExecHashIncreaseNumBatches gets called, those chunks are\nwalked through, moving the tuples to new chunks (or to disk, if they no\nlonger match the batch's bitmask), and freeing the old chunks.\n\nThe... |
[
{
"msg_contents": "Hi all\n\nWhile writing some tests I noticed that in newer IPC::Run or Perl\nversions (haven't dug extensively to find out which), perl appends the\nlocation to the extension, so 'ne' doesn't match the passed exception\nstring.\n\nPattern-match the exception string to handle this.\n\nBugfix, ... |
[
{
"msg_contents": "Hi Hackers,\n\nHere I attached an implementation patch that allows\nutility statements that have queries underneath such as\nCREATE TABLE AS, CREATE MATERIALIZED VIEW\nand REFRESH commands to benefit from parallel plan.\n\nThese write operations not performed concurrently by the\nparallel wor... |
[
{
"msg_contents": "I've stumbled over an interesting query out in the wild where the\nquery was testing a nullable timestamptz column was earlier than some\npoint in time, and also checking the column IS NOT NULL.\n\nThe use pattern here was that records which required processing had\nthis timestamp set to some... |
[
{
"msg_contents": "Per the discussion at\nhttps://www.postgresql.org/message-id/CA%2BOCxoyYxO%2BJmzv2Micj4uAaQdAi6nq0w25BPQgLLxsrvTmREw%40mail.gmail.com,\nattached is a patch that implements the following:\n\n- Adds a default role called pg_monitor\n- Gives members of the pg_monitor role full access to:\n pg... |
[
{
"msg_contents": "Hi all\n\nIs the current implementation of nested loop join in postgres\n(memory-trace) oblivious? Why or why not?\n\nHi allIs the current implementation of nested loop join in postgres (memory-trace) oblivious? Why or why not?",
"msg_date": "Sat, 25 Feb 2017 01:07:11 +0000",
"msg_fro... |
[
{
"msg_contents": "Patch to reduce lock levels of various relation-level options,\nfollowing on from earlier work by Fabrizio, with additional work and\ndocumentation by me.\n\nSubmitted to Sept CF, requested to start new thread to discuss\nhttps://www.postgresql.org/message-id/CA%2BTgmoYe5eDhjRodo3uOtVFGiDWwO2... |
[
{
"msg_contents": "Over in\nhttps://www.postgresql.org/message-id/alpine.DEB.2.11.1702251701030.3920@Sandal.Woodpecker\nit's pointed out that pltcl_loadmod was never updated for the switch\nto standard_conforming_strings (and the patch proposed there doesn't\nbegin to cover all the places that would need fixed ... |
[
{
"msg_contents": "Something that has been bothering me in PL/Python for a long time is the\nnon-object-oriented way in which plans are prepared and executed:\n\n plan = plpy.prepare(...)\n res = plpy.execute(plan, ...)\n\nwhere plpy.execute() takes either a plan or a query string.\n\nI think a better sty... |
[
{
"msg_contents": "logtape.c stores block numbers on disk. These block numbers are\nrepresented in memory as being of type long. This is why BufFile\nclients that use the block-based interface (currently limited to\nlogtape.c and gistbuildbuffers.c) must live with a specific\nlimitation, as noted within buffile... |
[
{
"msg_contents": "Dear 7b4ac19 authors,\n\nField ps_snapshot_data usually receives four-byte alignment within\nParallelIndexScanDescData, but it contains the eight-byte whenTaken field.\nThe select_parallel test dies with SIGBUS on \"Oracle Solaris 10 1/13\ns10s_u11wos_24a SPARC\", building with gcc 4.9.2. So... |
[
{
"msg_contents": "Hi all,\n\nI bumped into a case where it would have been rather useful to specify\na service file path in a connection string with a service name. In my\ncase, I have finished by setting up PGSERVICEFILE, but now like\nPGPASSFILE I think that being able to define the service file\navailable a... |
[
{
"msg_contents": "Hi all,\n\nAs the next commit fest beginning soon, there will be no place where\nhackers can park future patches for PG11~. So I have created a Future\nCF named 2017-07 where you can do that.\nFeel free to use it to park your future patches. The next development\nschedule is undecided, so the... |
[
{
"msg_contents": "I see that if the table is a inheritance parent, and ONLY is not\nspecified, the child tables are also added to the publication.\n\ncreate table parent (a int);\ncreate table child () inherits (parent);\ncreate publication parent_pub for table parent;\n\\d parent\n Table \"publi... |
[
{
"msg_contents": "Hi All,\n\nBelow are findings for postgres_fdw related to timezone.\n\n-- Remote Sever\n\n\nServer A (timezone = 'US/Eastern')\n\nCreate table public.test_timestamp(\n col1 varchar,\n ins_ts timestamp without time zone not null default\ncurrent_timestamp\n) witho... |
[
{
"msg_contents": "Hi,\n\nI'd like to propose to support parameterized foreign joins. Attached is \na patch for that, which has been created on top of [1].\n\nIn [2], I said that postgres_fdw could create parameterized paths from \njoinable combinations of the cheapest-parameterized paths for the \ninner/ou... |
[
{
"msg_contents": "This assignment is on todo list and has a benefit of providing an\nadditional defense against SQL-injection attacks. Previous mailing list\ndiscussion is here\n<https://www.postgresql.org/message-id/9236.1167968298@sss.pgh.pa.us> and I\nattach a small patch that fix the issue by checking whet... |
[
{
"msg_contents": "How about changing the default for log_directory from 'pg_log' to, say,\n'log'?\n\nWe have been emphasizing that the prefix \"pg_\" is for things reserved to\nPostgreSQL, whereas the pg_log directory is entirely an arbitrary\nuser-space name. Also, with a different name, the directory would ... |
[
{
"msg_contents": "Hi hackers,\n\nWorking on vectorized extension for Postgres (VOPS) I faced with two \nthings in Postgres compiler which break my expectations and force me to \nabandon my original implementation plan. I wonder if it is really \nprinciple and correct that:\n\n1. Moving-aggregate implementation... |
[
{
"msg_contents": "Currently pg_stop_backup() will wait for all WAL to be archived before\nreturning. If there is a problem with the archive command or a large\nbacklog it may not return for a very long time (if ever). Backup\nsoftware is faced with the choice of cancelling the query and hoping the\nstop back... |
[
{
"msg_contents": "I found this issue while working on a pg_stop_backup() patch. If a\nnon-exclusive pg_stop_backup() is cancelled and then attempted again the\nbackend will crash on assertion:\n\n$ test/pg/bin/psql\npsql (10devel)\nType \"help\" for help.\n\npostgres=# select * from pg_start_backup('label', t... |
[
{
"msg_contents": "Hello, I found a variable definition with wrong type\nspecification in KeepLogSeg, which doesn't harm anything.\n\n> static void\n> KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)\n> {\n> ...\n> /* then check whether slots limit removal further */\n> if (max_replication_slots > 0 &... |
[
{
"msg_contents": "Submissions for patches to be included in PostgresSQL 10.0 will be\naccepted until 2017-03-01 00:00 AoE (UTC-12).\n\nGet 'em in while you can!\n\n-- \n-David\ndavid@pgmasters.net\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to your subscription... |
[
{
"msg_contents": "Hi\n\nA novice user asked me about the last example here, which kept producing errors:\n\nhttps://www.postgresql.org/docs/9.6/static/tutorial-agg.html?\n\nIt turned out that the 'callout' was causing confusion because it\nsticks \"(1)\" into the middle of the query in approximately the same\n... |
[
{
"msg_contents": "Hello.\n\nAlthough replication slot is helpful to avoid unwanted WAL\ndeletion, on the other hand it can cause a disastrous situation\nby keeping WAL segments without a limit. Removing the causal\nrepslot will save this situation but it is not doable if the\nstandby is active. We should do a ... |
[
{
"msg_contents": "I have written an extension to manage openstreetmap data. There is a C\nfunction to perform spatial top k query on several tables and return an\narray of int8 type as result. The code skeleton of this function is as\nfollows:\n\nDatum vquery(PG_FUNCTION_ARGS) {\n\nint array_len = PG_GETARG_I... |
[
{
"msg_contents": "Here's a small patch to make a BRIN page range unsummarized. This is\nuseful if data has been deleted, and the heap pages are now used for\ncompletely different data.\n\n-- \n�lvaro Herrera\n\n\n\n-- \nSent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)\nTo make changes to y... |
[
{
"msg_contents": "Here's another small patch, this time from Simon Riggs. Maybe he already\nposted it for this commitfest, but I didn't find it in a quick look so\nhere it is.\n\nThis patch reduces the amount of bloat you get from running CREATE INDEX\nCONCURRENTLY by destroying the snapshot taken in the firs... |
[
{
"msg_contents": "Hello everyone,\n\nI've attached a patch which implements WAL consistency checking for\nhash indexes. This feature is going to be useful for developing and\ntesting of WAL logging for hash index.\n\nNote that, this patch should be applied on top of the following WAL\nlogging for hash index pa... |
[
{
"msg_contents": "Hello,\n\nAs I stated here and at the PGConf.ASIA developer meeting last year, I'd like to propose statement-level rollback feature. To repeat myself, this is requested for users to migrate from other DBMSs to PostgreSQL. They expect that a failure of one SQL statement should not abort the ... |
[
{
"msg_contents": "Hello Hackers,\n\nFollowing is a proposal for reporting the progress of ANALYZE command:\n\nIt seems that the following could be the phases of ANALYZE processing:\n1. Collecting sample rows\n2. Collecting inherited sample rows\n3. Computing heap stats\n4. Computing index stats\n5. Cleaning up... |
[
{
"msg_contents": "Hi all,\n\nI found a wrong explanation about tsquery_phrase. I was slightly confused when I tried to use it.\n\nThe current document explains tsquery_phrase as the followings[1].\n\n- Function: tsquery_phrase(query1 tsquery, query2 tsquery, distance integer)\n- Return Type: tsquery\n- Descrip... |
[
{
"msg_contents": "Hi all\n\nI would like to propose patch with a set of new small functions for fts in\ncase of\njsonb data type:\n\n* to_tsvector(config, jsonb) - make a tsvector from all string values and\n elements of jsonb object. To prevent the situation, when tsquery can find\na\n phrase consisting of ... |
[
{
"msg_contents": "Hi,\n\nHere is a patch for $subject. This is the same as what I proposed in \ncombination with a feature for full joins [1]; this would allow us to \npush down left/right/full joins with PHVs to the remote and improve how \nto deparse whole-row references. Since this is implemented on to... |
[
{
"msg_contents": "In CREATE INDEX CONCURRENTLY it seems possible to release the index\nbuild snapshot early, so we can reset our xmin before we complete the\nsort and write the main part of the index.\n\nPatch to implement this attached.\n\n-- \nSimon Riggs http://www.2ndQuadrant.com/\nPostgreSQ... |
[
{
"msg_contents": "Hello All,\n\nDuring the second heap scan of CREATE INDEX CONCURRENTLY, we're only\ninterested in the tuples which were inserted after the first scan was\nstarted. All such tuples can only exists in pages which have their VM bit\nunset. So I propose the attached patch which consults VM during... |
[
{
"msg_contents": "Hello.\n\nI decided to figure out whether current implementation of declarative\npartitioning has any bottlenecks when there is a lot of partitions. Here\nis what I did [1].\n\n```\n-- init schema\n\n\\timing on\n\nCREATE TABLE part_test (pk int not null, k int, v varchar(128)) PARTITION BY R... |
[
{
"msg_contents": "Hi all,\n\nNow we have a declarative partitioning, but hash partitioning is not\nimplemented yet. Attached is a POC patch to add the hash partitioning\nfeature. I know we will need more discussions about the syntax and other\nspecifications before going ahead the project, but I think this run... |
[
{
"msg_contents": "Hi,\n\ncommit 2f5c9d9c9ce removed CatalogUpdateIndexes and replaced it with \nCatalogTupleInsert/CatalogTupleUpdate, which do both the operation and \nindex update.\n\nBut there remained three places in comments still referencing the \nremoved CatalogUpdateIndexes. Attached is a patch fixing ... |
[
{
"msg_contents": "Hi all\n\nRegarding to the previous conversation [1], here is the patch for generic\ntype\nsubscripting with several improvements. It contains the following changes:\n\n* Subscripting node was separated into `SubscriptingRef` (only for data\n extraction) and `SubscriptingAssignRef` (only for... |
[
{
"msg_contents": "Greetings!\n\nThe PostgreSQL committers would like to welcome Andrew Gierth as a new\ncommitter for the PostgreSQL project.\n\nAndrew - welcome!\n\nThanks!\n\nStephen",
"msg_date": "Tue, 28 Feb 2017 13:22:22 -0500",
"msg_from": "Stephen Frost <sfrost@snowman.net>",
"msg_from_op": ... |
[
{
"msg_contents": "Hi there,\n\n\nAttached patch is an implementation of SQL/JSON data model from SQL-2016\nstandard (ISO/IEC 9075-2:2016(E)), which was published 2016-12-15 and is\navailable only for purchase from ISO web site (\nhttps://www.iso.org/standard/63556.html). Unfortunately I didn't find any\npublic... |
[
{
"msg_contents": "There are a couple of open items for the parallel CREATE INDEX patch\nthat at this point represent blockers to commit, IMV. The first is\naround a deficiency in the shared refcount mechanism, which is well\nunderstood and doesn't need to be rehashed on this thread. The second\nis the cost mod... |
[
{
"msg_contents": "Hello,\n\nCurrently inserting the data into a partitioned table that does not fit into\nany of its partitions is not allowed.\n\nThe attached patch provides a capability to add a default partition to a\nlist\npartitioned table as follows.\n\npostgres=# CREATE TABLE list_partitioned (\n a i... |
[
{
"msg_contents": "Hi,\n\nCurrently pg_stat_statements replaces constant values with ? characters.\nI've seen this be a problem on multiple occasions, in particular since it\nconflicts with the use of ? as an operator.\n\nI'd like to propose changing the replacement character from ? to instead be\na parameter (... |
[
{
"msg_contents": "PostgreSQL currently requires the file mode mask (umask) to be 0077.\nHowever, this precludes the possibility of a user in the postgres group\nperforming a backup (or whatever). Now that\npg_start_backup()/pg_stop_backup() privileges can be delegated to an\nunprivileged user, it makes sense ... |
[
{
"msg_contents": "Subject changed for better context of the patch.\n(was - Re: Question about grant create on database and pg_dump/pg_dumpall)\n\nOn Fri, Sep 30, 2016 at 12:29 AM, Tom Lane\n<tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:\n>\n>1. pg_dump without --create continues to do what it does today, ie it just... |
[
{
"msg_contents": "This is the \"grand finale\" that goes on top of the \"DROP FUNCTION of\nmultiple functions\" patch set. The purpose is to allow referring to\nfunctions without having to spell out the argument list, when the\nfunction name is unique. This is especially useful when having to\noperate on \"b... |
[
{
"msg_contents": "Here is a patch to refactor some macro hell in dblink.\n\nThis patch was discussed in the background sessions thread as a\nprerequisite for some work there, but I figure I'll make a separate\nthread for it to give everyone interested in dblink a chance to respond\nseparate from the other thre... |
[
{
"msg_contents": "The SQL standard defines a separate error code for nextval exhausting\nthe sequence space. I haven't found any discussion of this in the\narchives, so it seems this was just not considered or not yet in\nexistence when the error codes were introduced. Here is a patch to\ncorrect it.\n\n-- \... |
[
{
"msg_contents": "I think one of the most serious issues with BRIN indexes is how they\ndon't get updated automatically as the table is filled. This patch\nattempts to improve on that. During brininsert() time, we check whether\nwe're inserting the first item on the first page in a range. If we are,\nreques... |
[
{
"msg_contents": "Hi hackers!\n\nAs I've understood from thread [1] the main issue of creating local \nindexes for partitions is supporting REINDEX and DROP INDEX operations \non parent partitioned tables. Furthermore Robert Haas mentioned the \nproblem of creating index on key that is represented in partition... |
[
{
"msg_contents": "Hackers,\n\nI've attached a small patch which aims to improve the performance of\nAccessExclusiveLock when there are many AccessExclusiveLock stored.\n\n\n-- \n David Rowley http://www.2ndQuadrant.com/\n PostgreSQL Development, 24x7 Support, Training & Services\n\n\n\n-- \nS... |
[
{
"msg_contents": "The 2017-03 commitfest is now in progress. Here's the breakdown:\n\nNeeds review: 128\nWaiting on Author: 26\nReady for Committer: 26\nTotal: 180\n\nIf you are an author and have a patch in the \"Waiting for Author\" state\nplease update it as soon as you can. Otherwise, jump in and start\n... |
[
{
"msg_contents": "Hi,\n\nWhile reading through the cost_agg() I found that startup cost for the\ngroup aggregate is not correctly assigned. Due to this explain plan is\nnot printing the correct startup cost.\n\nWithout patch:\n\npostgres=# explain select aid, sum(abalance) from pgbench_accounts where\nfiller l... |
[
{
"msg_contents": "Hello,\n\nLooking into the MergeAppendPath generation, I'm a bit surprised by several \nthings:\n\n - When generating the mergeappendpath, we use a dummy path to take the sort \ncost into account for non-sorted subpaths. This is called with the base \nrelation tuples instead of the subpath e... |
[
{
"msg_contents": "Hi all,\n\nI found that ecpg has not been supporting COMMIT PREPARED and ROLLBACK\nPREPARED since version 9.0.2.\nFor example, if the test code does,\n EXEC SQL BEGIN;\n EXEC SQL INSERT INTO t1 VALUES(1);\n EXEC SQL PREPARE TRANSACTION 'gxid';\n EXEC SQL COMMIT PREPARED 'gxid';\nI got err... |
[
{
"msg_contents": "Dear community member(s),\n\nI am João Afonso, a Portuguese MSc student and I'm writing to ask some information about the GSoC projects.\n\nFor the reasons explained below, PostgreSQL was the organisation that I most identify with, so I am trying to introduce myself to the community. This way... |
[
{
"msg_contents": "Hello, some of my collegues found that orafce crashes with\npostgresql compliled with dtrace.\n\n=== The cause\n\nThe immediate cause was I think that it just did\nLWLockNewTrancheId and forget to do LWLockRegisterTranche. But\nthis is hardly detectable by a module developer.\n\n\nTypical tra... |
[
{
"msg_contents": "Allow vacuums to report oldestxmin\n\nAllow VACUUM and Autovacuum to report the oldestxmin value they\nused while cleaning tables, helping to make better sense out of\nthe other statistics we report in various cases.\n\nBranch\n------\nmaster\n\nDetails\n-------\nhttp://git.postgresql.org/pg/... |
[
{
"msg_contents": "Use asynchronous connect API in libpqwalreceiver\n\nThis makes the connection attempt from CREATE SUBSCRIPTION and from\nWalReceiver interruptable by the user in case the libpq connection is\nhanging. The previous coding required immediate shutdown (SIGQUIT) of\nPostgreSQL in that situation.... |
[
{
"msg_contents": "\nI have been setting up a buildfarm member with -DRELCACHE_FORCE_RELEASE\n-DCLOBBER_FREED_MEMORY, settings which Alvaro suggested to me.I got core\ndumps with these stack traces. The platform is Amazon Linux.\n\n\n================== stack trace:\npgsql.build/src/test/regress/tmp_check/data/c... |
[
{
"msg_contents": "Hello everybody,\n\nfollowing up on this thread: \nhttps://www.postgresql.org/message-id/flat/d297e048-ac49-9bed-32e3-9dd4e65d0978%40mail.de\nspecifically on this mail: \nhttps://www.postgresql.org/message-id/baef819f-acf0-a64d-c1eb-d2c5da1e5030%40mail.de\nI hope this idea fulfills the requir... |
[
{
"msg_contents": "Hi All,\n\n From following git commit onwards, parallel seq scan is never getting\nselected for inheritance or partitioned tables.\n\n<git-commit>\ncommit 51ee6f3160d2e1515ed6197594bda67eb99dc2cc\nAuthor: Robert Haas <rhaas@postgresql.org>\nDate: Wed Feb 15 13:37:24 2017 -0500\n\n Replac... |
[
{
"msg_contents": "I've been asked to investigate a case of a foreign join not occurring\non the foreign server as would have been expected.\n\nI've narrowed this down and the problem seems to only occur with outer\ntype joins.\n\nThe problem can be reproduced by the attached test_case.sql\n\nUpon investigation... |
[
{
"msg_contents": "Hello all,\n\nThis is to bring to notice a peculiar instance I found recently while\nrunning TPC-H benchmark queries. Q20 of the benchmark took 19 hours to\ncomplete when run on a machine with 512 GB RAM and 32 cores with\nfollowing parameter settings on the commit id -\n0c2070cefa0e5d097b715... |
[
{
"msg_contents": "Hello!\n\nMy name is Kirill Borozdin. I am a student of Ural Federal University from\nYekaterinburg, Russia.\nI was trying to find some interesting algorithmic problem for GSoC and\nfinally ran into PostgreSQL and\n\"Sorting algorithms benchmark and implementation\" task.\nThis problem looks ... |
[
{
"msg_contents": "Hi,\n\npg_dump segfaults if there are more than one DO_PUBLICATION_REL objects to\ndump.\n\ncreate table foo (a int);\ncreate publication foo_pub;\nalter publication foo_pub add table foo;\n\n$ pg_dump\n<OK>\n\ncreate table bar (a int);\nalter publication foo_pub add table bar;\n\n$ pg_dump -... |
[
{
"msg_contents": "psql currently supports \\di+ to view indexes,\n\n List of relations\n Schema | Name | Type | Owner | Table | Size | Description\n--------+--------------------+-------+-------+---------+--------+-------------\n public | ii | inde... |
[
{
"msg_contents": "Hi hackers,\n\nI've seen that pg_dump execute the dump of an eventual comment of a\nTSDictionary without\nspecifying the namespace where it is defined:\n\nhttps://github.com/postgres/postgres/blob/master/src/bin/pg_dump/pg_dump.c#L13542\n\nThis is actually a problem if a new TSDictionary is c... |
[
{
"msg_contents": "Hello,\n\nI would like to work on push-based executor [1] during GSoC, so I'm\nwriting to introduce myself and start the discussion of the project. I\nthink I should mention beforehand that the subject is my master's\nthesis topic, and I have already started working on it. This letter is\nnot... |
[
{
"msg_contents": "Remove objname/objargs split for referring to objects\n\nIn simpler times, it might have worked to refer to all kinds of objects\nby a list of name components and an optional argument list. But this\ndoesn't work for all objects, which has resulted in a collection of\nhacks to place various ... |
[
{
"msg_contents": "With e434ad39ae7316bcf35fd578dd34ad7e1ff3c25f I did a `make world`,\n`make install-world`, a fresh default initdb, a start with default\nconfig, `make installcheck`, connected to the regression database\nwith psql as the initial superuser, and ran:\n\nregression=# vacuum freeze analyze;\nWARN... |
[
{
"msg_contents": "Hi,\n\nRight now check-world, a sensible thing to run before commits that\naren't very narrow, takes a long time. To the point it seems to impact\ndevelopment velocity enough that it's more sensible to just skip it, and\nrely on the buildfarm.\n\nThat's not a great solution.\n\nA large amoun... |
[
{
"msg_contents": "Recap\n=====\n\nA design goal of parallel tuplesort is that the parallel case be as\nclose to possible as the serial case is already. Very little new code\nis needed in tuplesort.c and logtape.c, most of which is about using a\nnew lower-level facility which is very well encapsulated. And, ev... |
[
{
"msg_contents": "While scanning over postgresql.conf I happened to notice something\nthat didn't ring quite true about max_parallel_workers. The comment\nconfuses worker_processes with parallel workers.\n\nThe attached aims to put this right.\n\n-- \n David Rowley http://www.2ndQuadrant.com/... |
[
{
"msg_contents": "Hi,\n\nI was going through the grammer rule for Character types in gram.y and\nfound an optional sub rule in is \"opt_charset\"\n\nCharacterWithLength: character '(' Iconst ')' opt_charset\n{\n if (($5 != NULL) && (strcmp($5, \"sql_text\") != 0))\n $1 = psprintf... |
[
{
"msg_contents": "Hi,\n\nthere has been discussion at the logical replication initial copy thread\n[1] about making apply work with sync commit off by default for\nperformance reasons and adding option to change that per subscription.\n\nHere I propose patch to implement this - it adds boolean column\nsubssync... |
[
{
"msg_contents": "Hi,\n\nWhen I logged in PostgreSQL as non-superuser and ran\nALTER PUBLICATION command, I got a segmentation fault.\nThe code checking the owner of publication might have a bug.\n\n=# CREATE ROLE foo NOSUPERUSER LOGIN\n=# \\c - foo\n=> \\dRp\n List of publications\n Name | Owne... |
[
{
"msg_contents": "Is there a good reason why RELKIND_PARTITIONED_TABLE is 'P' not 'p'?\nIt looks rather out of place considering that seven of the eight\npre-existing relkind codes are lower case. (And no, I don't especially\napprove of RELKIND_SEQUENCE being 'S' either, but it's far too late to\nchange that.... |
[
{
"msg_contents": "Amit, Rafia,\n\nnodeIndexscan.c, unlike nodeSeqscan.c, thinks that a parallel-aware\nscan will always be executed in parallel mode. But that's not true:\nan Execute message with a non-zero row count could cause us to abandon\nplanned parallelism and execute the plan serially. I believe thi... |
[
{
"msg_contents": "Hi all!\n\nDoes anyone know of a simple, example Custom Scan Provider implementation\nfor 9.6+?\n\nI found pg_strom by searching GitHub. Its gpuscan.c looks like maybe it\nimplements a pattern similar to what I want to do, but there's a lot of\nextraneous (to me) stuff to parse through.\n\nI... |
[
{
"msg_contents": "Hi all,\n\nAs discussed on the thread dedicated to SCRAM\n(https://www.postgresql.org/message-id/243d8c11-6149-a4bb-0909-136992f74b23@iki.fi),\nhere is a separate thread dedicated to the following extension for\nCREATE/ALTER ROLE: PASSWORD ('value' USING 'method').\n\nNow that password_encryp... |
[
{
"msg_contents": "Hi all,\n\nHere is another thread for an issue related to SCRAM\n(https://www.postgresql.org/message-id/243d8c11-6149-a4bb-0909-136992f74b23@iki.fi),\nthat can be discussed independently: SASLprep.\n\nRFC5802 (https://tools.ietf.org/html/rfc5802) regarding the\nimplementation of SCRAM, needs ... |
[
{
"msg_contents": "Hello,\n\ndblink fails to close the unnamed connection as follows when a new unnamed connection is requested. The attached patch fixes this.\n\n\n\npostgres=# select count(*) from pg_stat_activity;\n count\n-------\n 1\n(1 row)\n\npostgres=# select dblink_connect('dbname=postgres');\n db... |
[
{
"msg_contents": "Hi,\n\nOver in another thread it was pointed out that a patch I submitted\nbroke a project rule by including \"postgres.h\" in a header. Here is a\npatch to remove it from dsa.h where I made the same mistake, and also\na case I found in bufmask.h by grepping.\n\nThere are also instances in r... |
[
{
"msg_contents": "On Wed, Mar 8, 2017 at 4:50 PM, Rushabh Lathia <rushabh.lathia@gmail.com>\nwrote:\n\n> I am attaching another version of the patch, as I found stupid mistake\n> in the earlier version of patch, where I missed to initialize initial\n> value to\n> WaitEventIO enum. Also earlier version was not... |
[
{
"msg_contents": "Fix parallel index and index-only scans to fall back to serial.\n\nParallel executor nodes can't assume that parallel execution will\nhappen in every case where the plan calls for it, because it might\nnot work out that way. However, parallel index scan and parallel\nindex-only scan failed t... |
[
{
"msg_contents": "Here's a patch which updates parallel.sgml for the new parallel scan\ntypes which have recently been committed, and also expands the\nexplanation of parallel joins slightly. I hope everyone will find\nthis useful in understanding what new capabilities we now have, and\nwhat remains to be don... |
[
{
"msg_contents": "Hi all\n\nI've found a minor off-by-one error in the resource retention logic\nfor logical slots, where we treat confirmed_flush as meaning \"flushed\nup to and including this LSN\". Seems reasonable, but the rest of the\ncode treats it as \"flushed up to but excluding this LSN\". In\nparticu... |
[
{
"msg_contents": "Hi!\n\nI'd like to offer a forward port from a change I'm contributing\nthe Greenplum code base over here:\n https://github.com/greenplum-db/gpdb/pull/1983\n\nWhile it is true that 64 bit atomics are mostly a no-op\nfor Postgres right now, perhaps this patch could still\nbe useful.\n\nTo m... |
[
{
"msg_contents": "Hi,\n\nWe are having some performance issues after we upgraded to newest\nversion of PostgreSQL, before it everything was fast and smooth.\n\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\nupgraded to 9.6.2 with no improvement.\n\nSome information about our setup: Freebsd,... |
[
{
"msg_contents": "Hi,\n\nWe are having some performance issues after we upgraded to newest\nversion of PostgreSQL, before it everything was fast and smooth.\n\nUpgrade was done by pg_upgrade from 9.4 directly do 9.6.1. Now we\nupgraded to 9.6.2 with no improvement.\n\nSome information about our setup: Freebsd,... |
[
{
"msg_contents": "Greetings,\n\nWhile reviewing Amit Langote's patch to handle partitioned tables\nproperly in various contrib modules (mostly by throwing an error since\nthings like pageinspect aren't going to work on the empty 'parent'\ntable), I went looking through contrib for other modules that do\nsometh... |
[
{
"msg_contents": "Greetings,\n\nWhile going over the contrib modules, I noticed that sepgsql was not\nupdated for partitioned tables. What that appears to mean is that it's\nnot possible to define labels on partitioned tables. As I recall,\naccessing the parent of a table will, similar to the GRANT system, n... |
[
{
"msg_contents": "I was just doing some testing on [1] when I noticed that there's a problem\nwith parallel bitmap index scans scans.\n\nTest case:\n\npatch with [1]\n\n=# create table r1(value int);\nCREATE TABLE\n=# insert into r1 select (random()*1000)::int from\ngenerate_Series(1,1000000);\nINSERT 0 100000... |
[
{
"msg_contents": "Greetings,\n\nConsider this:\n\ncreate table t10 (c1 int, c2 int);\ncreate index on t10 (c1) where c2 > 5;\n\n\\d t10\n Table \"sfrost.t10\"\n Column | Type | Modifiers \n--------+---------+-----------\n c1 | integer | \n c2 | integer | \nIndexes:\n \"t10_c1_idx\" btree (c1... |
[
{
"msg_contents": "I typically build with --with-system-tzdata, which means that any time\nRed Hat sees fit to push out a new copy of the tzdata package, that's\nwhat I'm testing against. This morning they updated to tzdata2017a,\nand I'm seeing the attached test failures.\n\nThe reason for this is that the IA... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.