Datasets:
willtheorangeguy
/
2017-Go-Time-Transcripts

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
1.49k
[1615.20 --> 1615.72] What's CT?
[1616.86 --> 1617.42] Yeah.
[1617.52 --> 1619.38] Sorry, that was too much of a tangent.
[1619.78 --> 1626.72] CT is a way to get CAs, the ones that sign TLS certificates, more accountable by forcing
[1626.72 --> 1632.24] them to disclose all the certificates they sign to a log where, long story short, they can't
[1632.24 --> 1634.18] be hidden or removed.
[1634.60 --> 1640.62] So you could use the same ideas to make a build server publish all the things it builds to
[1640.62 --> 1642.70] a log where they can't be removed.
[1642.70 --> 1648.38] So if they build something with a backdoor in it, it will show they can't hide it and
[1648.38 --> 1650.12] only provide it to some people.
[1650.12 --> 1656.82] And then you can prove it has a backdoor because it doesn't reproduce because you can't build
[1656.82 --> 1657.72] it from another machine.
[1658.28 --> 1659.64] Now, that's one step further.
[1659.88 --> 1661.26] It's called binary transparency.
[1661.54 --> 1666.44] It doesn't have that much to do with Go, but Go is a very good language to start this
[1666.44 --> 1671.68] because getting reproducible builds is incredibly hard with anything else.
[1671.68 --> 1678.54] Like the Debian project has been trying very hard to get the whole like dev repositories
[1678.54 --> 1682.96] reproducible and they're jumping through hoops that you can imagine.
[1683.54 --> 1687.14] With Go instead, you just set the same Go path and you're done.
[1687.76 --> 1687.86] Yeah.
[1687.90 --> 1691.84] I mean, you would just have to make sure you have the same Go tool chain, right?
[1692.24 --> 1696.82] The right Go version, because theoretically it wouldn't be, you wouldn't produce the same
[1696.82 --> 1698.84] binary if you upgraded your Go version.
[1698.84 --> 1699.52] Yep.
[1699.76 --> 1704.16] And I guess you would have to track the binary for each platform too.
[1704.58 --> 1704.80] Yeah.
[1704.92 --> 1709.52] Because the resulting binary would change for the Windows build versus the Linux build versus
[1709.52 --> 1711.32] the ARM builder.
[1711.80 --> 1712.06] Yeah.
[1712.16 --> 1717.84] You have the whole matrix architectures and operating systems, but those are like, I don't
[1717.84 --> 1719.04] know, 20 or something.
[1719.56 --> 1720.66] It gets bigger every release.
[1720.74 --> 1721.38] That's the best part.
[1722.04 --> 1722.64] Yep.
[1723.16 --> 1726.34] We're supporting 32-bit Spark on a Raspberry Pi now.
[1726.34 --> 1727.34] 32-bit Spark.
[1727.34 --> 1728.34] 32-bit Spark.
[1728.34 --> 1728.80] 32-bit Spark.
[1728.80 --> 1730.00] Is there really 32-bit Spark?
[1730.38 --> 1731.70] Not on a Raspberry Pi, no.
[1732.56 --> 1736.70] No, I made that up, but it was apparently nowhere near as funny as it should have been.
[1737.34 --> 1738.88] I'm like, you're confusing me here.
[1739.68 --> 1741.46] I can run Go on my Deck Alpha.
[1741.58 --> 1742.00] How's that?
[1744.56 --> 1745.78] Break out the Commodore.
[1746.42 --> 1749.22] I remember back in the days when a Deck Alpha was a big deal.
[1749.22 --> 1751.32] I mean, not even joking.
[1751.44 --> 1753.12] You can run Go on mainframes now.
[1753.82 --> 1753.92] Yeah.
[1754.10 --> 1754.32] Yeah.
[1754.72 --> 1754.98] Yeah.
[1755.50 --> 1758.22] It's just insane watching Go take off.
[1758.44 --> 1762.04] It blows my mind in just a couple of years everywhere that we see Go.
[1762.56 --> 1769.46] So you mentioned something in our show document about latency profiling and Camly Store too.
[1769.98 --> 1770.36] Oh, yeah.
[1770.52 --> 1770.76] Okay.
[1770.94 --> 1772.60] So let's see.
[1773.04 --> 1774.66] Let's go for spoilers first.
[1774.66 --> 1780.32] But latency profiling is what I plan to talk about at Go4Con India.
[1781.08 --> 1787.76] And essentially, the story there is that we are used to all the profiling tools, and they're super nice, super easy to use, right?
[1787.88 --> 1792.40] And you can figure out what functions take the most CPU in your program.
[1792.96 --> 1794.04] And that's nice, right?
[1794.72 --> 1801.20] But then at the end of the day, you realize that that's not really what we are optimizing for most of the time.
[1801.20 --> 1806.52] It's very hard to build an application that takes 100% of CPU.
[1807.10 --> 1809.30] So what is it doing all the rest of the time?
[1809.82 --> 1811.68] Well, it's blocking on something.
[1811.84 --> 1812.74] It's waiting on something.
[1812.94 --> 1815.24] And what metric does that worsen?
[1815.50 --> 1819.54] Of course, latency, which a lot of the times is the one we care the most about.
[1819.54 --> 1827.20] Like, it's rare that APIs are about throughput, about how many things concurrently they can process.
[1827.38 --> 1828.72] It's absolutely possible.
[1829.20 --> 1832.34] But like most of the times, but like this is debatable.
[1832.64 --> 1838.62] But surely there is like a huge interest in reducing latency, making APIs return as quickly as possible.
[1838.62 --> 1846.70] So Go does provide the tools to inspect what functions are slow, are just, you know, taking a long time.
[1847.08 --> 1852.20] But they're not as well surfaced and publicized as their CPU counterparts.
[1852.20 --> 1865.54] So they involve mostly the GoTracer, which can then generate profiles for blocking, for I.O., for network, and for scheduling poses.
[1866.04 --> 1869.08] And these are the things that will actually make you, like, suffer.
[1869.24 --> 1874.98] Imagine having some function somewhere that downloads, I don't know, a tarball from the Internet.
[1874.98 --> 1879.94] That will not show up in the CPU profile, maybe, but it will take a long time.
[1880.12 --> 1883.00] And if your API is blocking on that, it will take forever, right?
[1883.56 --> 1893.02] Yeah, so that's actually really interesting because that debate comes up a lot, especially even talking about garbage collection, the difference between latency and throughput.
[1893.82 --> 1897.04] You know, both of them kind of represent speed, right?
[1897.36 --> 1897.62] Right.
[1898.02 --> 1901.30] And deciding which one you want to optimize for is difficult.
[1901.96 --> 1902.96] That's a very good point.
[1902.96 --> 1918.08] Like, in Go, we almost explicitly stated that we optimize for latency by making the, at least in the garbage collector, by making it faster and faster in terms of poses, but slightly slower in terms of CPU and, you know, throughput.
[1918.60 --> 1923.84] So it's interesting that the profiling tools haven't caught up to the same priorities.
[1924.52 --> 1924.64] Right.
[1924.94 --> 1926.40] Yeah, that's true.
[1926.64 --> 1928.98] So you're talking about this at GopherCon India?
[1929.50 --> 1929.70] Yeah.
[1929.96 --> 1930.66] Yeah, that's the plan.
[1930.66 --> 1933.52] Which is, that's coming up in February.
[1933.92 --> 1935.46] I'm trying to remember the exact date.
[1936.00 --> 1937.20] Does anybody remember a MoFan?
[1937.64 --> 1938.30] I don't.
[1938.38 --> 1940.12] February 21st?
[1940.60 --> 1944.66] I mean, don't book tickets based on my collection, but...
[1944.66 --> 1948.18] Check the website first before buying plane tickets in hotels.
[1949.40 --> 1950.66] You might be early or late.