text stringlengths 0 1.82k |
|---|
**Aaron Schlesinger:** I would go with the latter. I know that might be heresy, but I would absolutely say learn Go, learn idioms, and once you start getting more complex, you have a 5,000-line codebase or 10,000-line codebase, that might be a good time to start looking at design patterns and finding ways to reduce you... |
**Carlisia Thompson:** Can I just +1 that? \[laughter\] |
**Aaron Schlesinger:** Maybe it wasn't heresy... |
**Carlisia Thompson:** Yeah, I agree with that, and I'm very much a newcomer to Go. So by just using the straightforward stuff and learning the idioms, you might not have that organization at the end, but you're going to end up with a much bigger tool belt that you can use. Because design patterns, they don't really ch... |
**Brian Ketelsen:** Alright, it's time for us to take a quick break to talk about our sponsor, Linode. |
**Break:** \[\\00:24:22.17\] |
**Erik St. Martin:** Excellent! So you mentioned that you worked at Deis. I'd love to talk to you a bit about the work that you guys are doing there; there's some really interesting contributions you guys are making to Kubernetes, Helm and things like that. So I'd love to hear about work that you're doing there, and ma... |
**Aaron Schlesinger:** When I first got to Deis, I worked on the PaaS. It's now called Deis Workflow, but at that point it was just Deis. It was basically trying to be a Heroku for Kubernetes - open source, you can go and install it on Kubernetes. Then right before I got there, Helm had kind of become a thing, but it l... |
So it was kind of like this plug-and-play thing where if you wanna use your own logging you can, but we'll also ship you components that can do most of the logging stuff that you'll need. Then there is the router, the routing mesh - you could use that on your own, and it would fit into the PaaS. |
Helm popped up because we wanted to give that flexibility to people without writing a thousand page document with specifications for each module, and having sample manifest and all that stuff. So Go came into play big time with the PaaS, because we had to do things like watch the Kubernetes event stream, see when a pod... |
When you're watching an even stream that screams concurrency, once it screams concurrency, then I just pick up Go. It's just the easiest thing for me, by far. Before we picked up Go there was some Python... Actually, there was some Go before I came to Deis, and then there was some shell script. The shell script compone... |
**Erik St. Martin:** I was just gonna say, what is this? Just a four-loop curling out, long-polling? |
**Aaron Schlesinger:** \[27:50\] Yeah, it would sleep for two seconds and then poll the Kubernetes API. I made the decision to make the plunge after we started getting bug reports where people were saying, "Oh, it missed the pod!" Because the pod would start up and then die within that two seconds. So that's when we to... |
We still have some old stuff in Python and it's working great. Our API server is actually all Python, and Python is super well suited for that. But the other components - our logging system is all in Go, that deals with consuming and fanning out tons of log data. We have a log storage system, same thing; built on Redis... |
The Go choice there has been super helpful, because just by virtue of its simplicity and its concurrency support. Looking back, I kind of think "Why didn't we move sooner to Go?" because it saved us so much strife. |
**Erik St. Martin:** Out of curiosity... All your watchers and stuff like that for the event stream - are you using the Kubernetes client library, or did you just write an HTTP wrapper in Go? |
**Aaron Schlesinger:** We are almost exclusively using the client library. I actually just saw today that they split up the client library and they're starting to pull out pieces to a new repo. |
**Erik St. Martin:** Yeah, there's the client Go, and I'm actually in the middle of refactoring out some of my own logic to call that... |
**Aaron Schlesinger:** Oh man, I'm so excited for that, because right now our dependencies are like a gig of all the Kubernetes code and all the dependencies that we don't need. Once we have that, it's gonna be a couple kilobytes of code; I'm gonna be so happy when we get that... |
**Erik St. Martin:** Yeah, I was really happy to get rid of the whole vendored Kubernetes repo. |
**Aaron Schlesinger:** Yeah, that's a bit of a mess. |
**Erik St. Martin:** So that's awesome. Are you contributing to Kubernetes, or is this mostly kind of tooling built around Kubernetes? |
**Aaron Schlesinger:** Most of it is tooling built around... When we find warts in Kubernetes that affect what we're doing, or we find warts that are kind of related to issues that we've seen with Deis, then we usually contribute upstream. But now, Helm is part of the Kubernetes repo. So Helm, technically, since we're ... |
**Erik St. Martin:** Yeah, there's more and more Special Interest Groups now that I've been seeing come out. It's too hard to keep up with all of them. |
**Aaron Schlesinger:** Yeah. What I would love to see eventually is some kind of centralized schedule for SIGs, so that we can all do that. We'll just figure out "What is today's SIG? What are they talking about? What's on their agenda?" |
**Brian Ketelsen:** I had to unsubscribe from all those SIG lists. It was driving me crazy. |
**Aaron Schlesinger:** \[laughs\] Yeah, I feel your pain. |
**Erik St. Martin:** Well, and there's so many good groups, large organizations submitting proposals and stuff for expansions to Kubernetes. It's really hard to keep up with all the proposals that are going on. You kind of have to pick your world and hang out there. |
**Aaron Schlesinger:** Yeah. |
**Brian Ketelsen:** Did we mention that SIG is a Special Interest Group in Kubernetes? We're way out of this typical Go world now, sorry. Too many Kubernetes users here. |
**Erik St. Martin:** It's all written in Go, it applies. |
**Brian Ketelsen:** It applies, we're just getting deep. |
**Aaron Schlesinger:** \[31:57\] Yeah, and you know, to take it back to Go for a second, the Kubernetes codebase is extremely interesting from a, maybe we can say an etymology standpoint; the original codebase was kind of written like Java, and then it open sourced, and then people outside of Google started contributin... |
**Erik St. Martin:** ...generated code... |
**Aaron Schlesinger:** Yeah, generated code, there's some protobufs-generated stuff in there, there is a swagger spec which I think generated some code at one point, and then they didn't generate it again, and started just building on top of the generated code. Looking at this massive codebase, you can jump to definiti... |
**Brian Ketelsen:** Absolutely. |
**Erik St. Martin:** Yeah, I'd have to agree. From file to file, things kind of... There's not a whole lot of consistency. In certain areas there's consistency; you can tell certain groups of people work in different areas, but if you're bouncing around the repository, you can definitely see the style changes. I think ... |
As long as people are continuously refactoring a little bit to make more idiomatic Go out of these areas they touch, it's slowly going to evolve into that. But yeah, you can definitely see Java patterns in there. |
**Aaron Schlesinger:** Yeah, absolutely. |
**Erik St. Martin:** So, does everybody wanna talk about any news and projects that have been going on? |
**Brian Ketelsen:** It's a pretty big week for interesting things coming out. The Go newsletter came out today, and they had like an ngrok clone for SSH. That was generally interesting on its own; I can't remember what it was called. But the thing that powers it was much more interesting to me, which is Teleport. githu... |
**Erik St. Martin:** That's interesting, I actually didn't see that. I haven't seen the newsletter today. So one thing I saw - it was either today or yesterday - Brad Fitzpatrick mentioned that they are officially getting rid of the legacy backend for the Go compiler, which means from this point on it will be all SSA, ... |
**Aaron Schlesinger:** Cool! I didn't see that. |
**Brian Ketelsen:** Yeah, SSA or GTFO. \[laughter\] |
**Erik St. Martin:** I'm just really interested to see that, because it's going to make things much easier to continue to write rules to make more performant machine code out of it. |
**Brian Ketelsen:** But look at the gigantic gains that we got just in 1.7; I can't wait 'till people have some time to actually work on some enhancements to that. I think 1.8, 1.9, 1.10 are gonna be amazingly fast and stable and awesome. |
**Erik St. Martin:** \[36:06\] Yeah, especially when more people start getting in and writing the SSA rules. That's far beyond my ability to start looking at assembly language and coming up with these rules. |
**Brian Ketelsen:** Well, speaking of 1.9, Vim Go 1.9 was released. We've got a lot of Vim Go lovers. That was a big release. |
**Erik St. Martin:** We should ask this every episode to our guest - what is your editor of choice, Aaron? |
**Aaron Schlesinger:** Oh, man... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.