text stringlengths 0 1.82k |
|---|
**Carlisia Thompson:** That's interesting. What about Go Kit Log? |
**Peter Bourgon:** Go Kit Log is about how you manage logging within your process, within the service itself. What I'm curious about is once the log information leaves the process boundary, like on STDOUT, say, how do you get that into a system that is searchable and usable, operationally simply, and without having to ... |
**Carlisia Thompson:** Okay, so are you talking about what Prometheus is for metrics, you're talking about that for logging? |
**Peter Bourgon:** Yeah, something like that, at a very high level. |
**Erik St. Martin:** Have you looked at things like Logstash, or...? |
**Peter Bourgon:** Yeah, Logstash is like a FluentD thing where it's pushing logs around, but it's not actually doing any storing or querying. |
**Erik St. Martin:** Okay, so you wanna address this from the... There's lots of stuff out there for doing metrics and alerting and things like that; you want something that's designed specifically for logs storage. Like, if we had to rethink distributed log transfer and storage and querying today, what would that look... |
**Peter Bourgon:** Exactly. And maybe there's already software that's purpose-built for exactly what I'm thinking and I can just use it. If so, that's great, I'd love to know about it. But I don't. |
**Carlisia Thompson:** But you're talking about in Go, right? Or just in general. |
**Peter Bourgon:** Well, in general would be fine. The only thing in the space that I'm aware of that serves this need is Elasticsearch and it's too operationally tricky, and some other reasons I won't get into. I'm not a big fan of it for this use case, but yeah... Maybe something in Go, that would be great. |
**Erik St. Martin:** Now, Scott has a different opinion on log storage. Netflix doesn't do a lot of the distributed logs. Do you wanna speak to that a little bit, Scott? |
**Scott Mansfield:** So I would probably need to clarify that... We do have a massive logging system; we generate a ton of logs, but I don't... So as a company - it's the same old joke: we're a logging system with side effect of streaming video, and there's a massive Kafka cluster that's the ingest for this, and I beli... |
**Peter Bourgon:** ...HTFS, I imagine. |
**Scott Mansfield:** Yeah, but my view is... I tend to try not to rely on logs, I just keep metrics around. Everything that I could possibly want to introspect. |
**Peter Bourgon:** Totally. |
**Scott Mansfield:** ...and if there's some specific thing... It's like bucketize all unknown errors, then yeah, we'll start logging something. But at that point there's no performance hit because I'm gonna close the connection anyway. So yeah... That's my little speech. |
**Erik St. Martin:** It's difficult when you get at scale; you can't log into the servers and check the logs anymore. You kind of have, like you said, metrics, and then you have the log messages that you wanna be alerted on, so there's these different reasons you want logs. You want to diagnose a distributed -- trace b... |
I think that solves most of the use cases for at-scale logging, because I think it gets too hard... Can anybody think of any other use cases that a system like this would need to solve? |
**Peter Bourgon:** \[01:20:00.27\] I mean, certainly... I cut the logging domain into two parts: one is your typical debug info warn application logging, where you might need this kind of deep introspection in order to debug certain issues. But you don't need a high quality of service; you can drop some of those log me... |
Both of these things is sort of what I'm considering, although they are drastically different QoS guarantees. |
**Carlisia Thompson:** Well, let's talk about structured versus unstructured logging... Nah, I'm kidding. \[laughter\] That's a whole show right there. |
**Erik St. Martin:** The gloves come off... \[laughter\] So we wanna move on to \#FreeSoftwareFriday? In case we didn't tell you, Peter... |
**Peter Bourgon:** Uh-oh... |
**Erik St. Martin:** To your point early on, that you don't really hear from people when all is going well - we try to (in every episode) kind of just throw out a project and thank them. It does not have to be written in Go. It can be a person, group, or project, and just kind of thank them for providing stuff that mak... |
**Peter Bourgon:** Great, thanks. |
**Erik St. Martin:** Alright, Carlisia, do you wanna go first? |
**Carlisia Thompson:** No, I don't have anything today. |
**Erik St. Martin:** How about you, Scott? |
**Scott Mansfield:** Sure... Last week I went to QCon in San Francisco, which is a really amazing conference, actually... It was really high-quality. And I went to one talk on Twitter's caching system, which naturally I was interested in, because I work on the Netflix caching system. They have written their own C-based... |
There's all kinds of things... They use a static size hash table because they used to see these latency spikes at exponentially increasing intervals, which happened to correspond with hash table extension, and a variety of things like that. The talk was really cool... Unfortunately, QCon talks take a while to come out,... |
**Erik St. Martin:** Oh, cool. So it was recorded? It will be out eventually? |
**Scott Mansfield:** Eventually... I don't know when, unfortunately. I have access to the videos because I attended, but... |
**Erik St. Martin:** Oh, brag about it. \[laughter\] So to give you a little more time, Peter, I'll go next. So I have not used this, I'll preface it with this - I have not used this yet, but I was at KubeCon last week and I was talking to the CoreOS guys, and they have a really cool project called Zetcd. It's basicall... |
Alright, Peter, do you have anything? Feel free to say no. |
**Peter Bourgon:** \[01:24:03.21\] No, I've got one. It's a small thing, because sometimes the small things are the best things... I was using grep for most of my life until I stumbled over this program called ack, which was a bit more usable version of grep. And I was using that for a long time until I stumbled over t... |
You can install it, and it installs this pt, and you can use it like a grep, except you don't have to do weird contortions with -r, and grep like you expect it should work. It's super, super fast. |
**Carlisia Thompson:** I'm glad you've mentioned that because Dave Cheney mentioned it, and I was like "I'm gonna install it immediately", and I never did. But now I'm gonna install it immediately. \[laughs\] |
**Erik St. Martin:** I'm in the same boat. I used the Silver Searcher for many years now, and it still installed, and people keep reminding me of The Platinum Searcher, and I'm like "I'm gonna use that", and then I go along my way and continue to build stuff and I forget all about it. So now it's gonna live in my sea o... |
Alright, so I think with that we are out of time, and I definitely wanna thank everybody on the panel. Thank you, Scott, for stepping in, and thank you, Peter, for coming on the show. Thanks to everybody who's listening now, and who will be listening to us when the recording is released. |
Huge thank you to our sponsors, Minio and Backtrace. Definitely share the show with your friends, family, colleagues... We are @GoTimeFM on Twitter, GoTime.FM - you can go there to subscribe, and if you want to be on the show or have ideas for the show, or just questions of the hosts or guests, hit us up on github.com/... |
**Carlisia Thompson:** This was fun, bye! Thanks, Peter. |
**Peter Bourgon:** Bye guys. I had a hacking good time, thanks everyone! |
**Erik St. Martin:** Hacking good time... \[laughter\] |
• Feature freeze for Go 1.7 announced |
• Binary-only packages now allowed in Go |
• Peter Bourgon updates his "Go Best Practices" talk from 2014 to 2016 |
• Discussion of impostor syndrome and code review by Brad and Andrew |
• Upcoming GopherCon presentation on Gopher data science by Daniel Whitenack |
• GAFKA: a Go tool suite for managing KAFKA clusters |
• ChatOps and microservices with Micro and ChatOps bots |
• UNIK: compiling apps into unikernels |
• Micro framework as an ecosystem for microservices |
• Security benefits of unikernels (reduced attack surface) |
• Benefits and limitations of unikernels |
• Data science definition and process |
• Industry applications and drivers behind data science advancements (commerce, advertising) |
• Use cases for data science in business processes and engineering |
• Open-source tools and frameworks for data science (R, Python, etc.) |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.