text stringlengths 0 1.82k |
|---|
**Ben Johnson:** Yeah, and they work great for a lot of use cases. They tend to be much more complex, especially operationally, but they get much better write speed, on something like a B+ tree typically, for random writes. You do stuff in things like range scans, you need to go over an ordered set of keys... They're t... |
**Erik St. Martin:** So now the largest user of Bolt is InfluxDB, right? |
**Ben Johnson:** No. Influx ended up… They have their own format for time series, that's more efficient. The Bolt piece is more of a stop gap. There are a lot of pieces going on as far as distribution of the data, query language, and all that. We moved off of Bolt as actual main storage. We have something called TSM-1 ... |
**Erik St. Martin:** Okay. Okay, I remember it having swappable backends to begin with. I think you guys decided that it was too much overhead to support all of them. |
**Ben Johnson:** At the time there weren't many people, so it was way too much overhead. There were some issues with some of the other libraries we were using, more from an operational standpoint, not so much from the performance standpoint. As far as the largest user, which I'm not sure… Actually, I don't know how muc... |
**Brian Ketelsen:** Wow. So how does one operate, how does an operator manage a large BoltDB database that big? How do you take backups of that file? What are the operational concerns when it comes to using BoltDB? |
**Ben Johnson:** It is a B+ tree, so as it gets larger, depending on how you structure your buckets inside, the accesses will get, I think login slower. Don't quote me on that. One of those O(n), big O notations. So it will start getting a little slower over time. As far as operationally, Bolt, the actual transaction i... |
**Brian Ketelsen:** \[32:10\] Oh, nice. |
**Ben Johnson:** Yeah. Fully serializable, ACID, all that jazz. |
**Brian Ketelsen:** That's beautiful. |
**Ben Johnson:** And again, that's another tradeoff, too. If you're using an LSM tree, it's much more difficult to snapshot the whole database, because you have multiple files as well other operational issues there. It's not a good or bad, it's again, tradeoffs. |
**Erik St. Martin:** I think with almost every tech decision, you have to look at your problems base and determine which decisions gives you the best amount of benefit for the least amount of drawbacks, right? |
**Ben Johnson:** Yeah, totally. |
**Erik St. Martin:** I think that's kind of the whole thing, how we were talking about the appeal of things like Rails. If you're a company that's bootstrapping yourself, and trying to get a product to market, get a proof of concept out - there's huge appeal in how fast you can deploy a proof of concept website in Rail... |
**Ben Johnson:** Oh, sure. Yeah. |
**Erik St. Martin:** Especially if the majority of it is CRUD-based operations. I mean, you can throw together a decent site in a weekend, just hammering it out. |
**Ben Johnson:** It's pretty much what hackathons are, or the startup weekends - it's just like quick Rails apps in a day or two. You can really get some great stuff, some concepts going. |
**Erik St. Martin:** What were those called, the Rails events, Brian? They held a couple locally here, too, where they kind of did it across the country, and people got together. They formed teams and built stuff. It wasn't a hackathon, they called it something else… |
**Carlisia Thompson:** I know what you're talking about, it was at Gopher Gala. I did it once... Rails Rumble. |
**Erik St. Martin:** Rails Rumble. That's what I'm talking about. There were some cool things that came out of that that I really wish had gone somewhere. And some of them were completely humorous. Like, who came up with this idea? |
It actually reminds me too, because there's a new database I saw come out too called noms, and it reminds me of a Rails Rumble project that was here local in Tampa that was called the Omnominator \[laughs\]. |
**Brian Ketelsen:** Omnominator, that's right. I remember that. |
**Erik St. Martin:** I can't remember exactly what it did, but it was something along the lines of finding a local restaurant, but I loved the name \[laughs\]. |
**Brian Ketelsen:** Yeah, like a food truck search engine, or something silly like that. |
**Erik St. Martin:** For the life of me, I can't remember... I haven't seen a Rails Rumble; I haven't been completely connected to the Rails world in a while, either. |
**Brian Ketelsen:** No comment. |
**Erik St. Martin:** So Adam Stacoviak from Changelog just posted a link to it. The site is not up, but the GitHub still exists for Omnominator. |
**Carlisia Thompson:** I want to make sure we get Ben to answer the question that Erik asked. |
**Erik St. Martin:** Answer the question, Ben! \[laughter\] |
**Carlisia Thompson:** Now that you've written a couple of really thorough blog posts on doing the walkthrough with Go Standard libraries, are you up to doing more of it? |
**Ben Johnson:** Oh, yeah. I have 27 more posts lined up, that I'm working on. |
**Carlisia Thompson:** Wow. |
**Ben Johnson:** I'm trying to start on the lowest level of the stack, which is why I did IO and bytes first, because I really don't depend on much Unicode stuff. But I kind of want to move up into the encoding ones, and things that lay on top of that, and so on and so forth. |
**Erik St. Martin:** Nice. |
**Ben Johnson:** Yeah. The next one that's coming is the encoding package. Most people have never actually looked at the encoding package because there's only four interfaces, but I started breaking it out, and going to do some overviews of the other packages inside of there, and what encoding means, and it should work... |
**Erik St. Martin:** \[35:48\] I think working with bytes and streams in the IO package is a really good place to start. Especially if people come from dynamic languages, they're highly used to just working with strings, so they're going to favor those a lot and create a lot of copies, and they're going to buffer a lot... |
**Ben Johnson:** Yeah. Yeah, thanks. |
**Erik St. Martin:** I'd like to see you answer to Bill's challenge. It wasn't really a challenge, but he was looking for a better way to… Basically replacing a string in a stream, in a continuous stream. I started on a solution for it, but I just haven't had the time to put into doing it. |
**Ben Johnson:** Replacing a stream? What was the example? |
**Erik St. Martin:** He had a piece of code that he had put up on Play, the playground, and it was basically just taking bytes, and when it saw bytes in the stream, it would do it. I was trying to do a continuous stream, so basically no matter how many bytes were read in, it could figure out and buffer just enough byte... |
**Ben Johnson:** Oh, sure. Yeah. |
**Erik St. Martin:** Because that's a common mistake people get when they're using readers and things like that, to think that they're reading all the content in one read; it could come up short for whatever reason. So that was basically what I was trying to implement a solution for, to kind of demonstrate if you're lo... |
**Ben Johnson:** I look for that all the time. |
**Erik St. Martin:** Yeah, you could get the 'o' and the 'm' at the end of your first read and the rest is in the next, so that you can't just look at the buffer that came out of that. You're gonna have to look at the stream and build a little state machine internally as the bytes go by to figure out whether you found ... |
**Ben Johnson:** I was going to say state machine is probably a good way to go. |
**Erik St. Martin:** I started on an approach, I just didn't have the time to finish it. I think it would be an interesting thing to finish because it demonstrates a lot of the points from both your Bytes Walkthrough and the IO Package. |
**Ben Johnson:** Oh sure, yeah. |
**Brian Ketelsen:** What was nice about that specifically is that it kind of turned into VimGolf where we started with an implementation, and then somebody improved it, and then somebody else improved it, and then somebody really improved it. That's how I learned. That's wonderful. I love seeing the evolution of perfor... |
**Erik St. Martin:** So the interesting thing about that problem was I was like, "Oh, this is going to be easy; I'll build a little state machine." What prevented me from finishing it is there's several little edge cases that come as a part of it, like when you get the partial read. You need to buffer for a second beca... |
I was trying to build the most efficient version I could, and then I kept ending up in these edge cases where based on the number of bytes that ended up being read, it'd be like sometimes it worked, and I'm like, "Why?!" Programming is hard. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.