text stringlengths 0 1.82k |
|---|
When I give my little workshops, I say "If you're five engineers or fewer, there's no reason that you should be running a microservice architecture." You're just not having any of the problems that microservices solve. This architecture you describe is a great alternative, and one that you can kind of adapt to microser... |
**Carlisia Thompson:** Yeah, I saw Peter's talk at Golang UK and he was very discouraging of using microservices. \[laughter\] And I really liked it because he went through all the pain points, some obvious, some not so obvious - obviously, he has a ton of experience - and he went through many pain points that you can ... |
But at the same time, Scott's remark is to the point I think, because as I have been learning more about packaging in Go in the recent months, Go makes it so easy to really contain your code through the use of packages. I wonder if for Go the heuristics would be different as far as a feasible size for a codebase. |
**Peter Bourgon:** Do you think it would be bigger or smaller, Carlisia? |
**Carlisia Thompson:** I'm wondering if it could not be bigger and still be very functional, if you use the features of the packaging; now there's that internal feature which allows you to hide even more things. And there are companies using monolithical apps, like Digital Ocean, and there's another one that I forgot n... |
**Peter Bourgon:** Yeah, I think it's completely viable. I guess I don't have the stamina to give my workshop again here, but there's so many problems that come when you're splitting your business domain along process boundaries. If you can avoid doing that, if you can split it on package boundaries and then wire thing... |
Ben Johnson has an opinion about this, but I've tried to use it and it often fails for me. Maybe I'm holding it wrong. There's lots of opinions about this sort of thing; I'd love to hold a panel about that as well. |
**Carlisia Thompson:** I would, too. I'm super interested in that. By the way, the other company I was thinking about was Google. Google has a single repo. But yeah, Ben Johnson was talking about that, Bill Kennedy promised that he's going to start writing about that as well... I'm working on a talk and a blog post abo... |
**Erik St. Martin:** \[24:19\] It depends on how coupled your repos are. If you have multiple repos that are highly coupled to each other, then the testing story and deployment story gets more complicated with that, too. |
I think it's interesting, because a monorepo can still be broken up, right? If you look at the Kubernetes codebase, or the Docker codebase, there's a command directory which kind of has the main packages in it, and then there's like a package directory that has code implementations of those things. So there's still are... |
It's really hard though, because anybody who's worked for companies large and small, you start "Well, it's good for these scenarios but bad for these others" and you can join the other team depending on kind of what the exact use case is. I love monorepos for many reasons, and then I love splitting it up for others. I ... |
I haven't seen Ben Johnson's post on what he's recommending there... At least I don't remember it. It might be my seven-day window thing. I need to look at that. Is that on his Medium blog? |
**Carlisia Thompson:** Yeah, he has a blog post and he has a repo with an example. And actually, if you look through the pull request, there's quite an instructional conversation between Peter and Ben going over the tradeoffs, it's very interesting. One thing's for sure - a monorepo would make it easy for dependencies,... |
**Erik St. Martin:** Yeah, so I wanna get into dependencies and I wanna get into Go Kit a bit. But before we do that, it is time for our first sponsor break. |
**Break:** \[\\00:26:46.05\] |
**Erik St. Martin:** Okay, moving on. Carlisia, you wanted to talk about... |
**Carlisia Thompson:** Dependency management. |
**Erik St. Martin:** Dependency management! |
**Carlisia Thompson:** That little thing... \[laughter\] |
**Erik St. Martin:** That tiny little problem nobody has, right? |
**Carlisia Thompson:** Yeah. |
**Erik St. Martin:** Peter, I know you have a lot of views on this, and some of the tooling and how things have evolved, and you've kind of followed this all the way through its course, which has been "As long as it's solved..." |
**Peter Bourgon:** Quite on the contrary... I actually have almost no strong opinions about it. What I do have a strong opinion about is that it gets solved, yeah. Precisely. |
So how did this all begin? Back in the early days, in the pre-1.0 days, we all decided we should go get our dependencies, and because goget identifies projects in sort of a spacial dimension, they identify them by name, but provide no way to identify them in sort of a time dimension, there's no way to identify version.... |
This hope-driven development worked surprisingly well for a number of years, in the sense that Go gained popularity and people were still shipping productive production software with it. But in the open source ecosystem it has sufficient differences to the Google monorepo and to the Google way of doing things, so as we... |
**Erik St. Martin:** Yes. |
**Peter Bourgon:** Goodness... |
**Erik St. Martin:** And ouch! Very much ouch! I'm building stuff against their APIs too, and there's two problems. One, before they did the client Go library, which is just the client libraries, you had to import the entire Kubernetes package, which was terrible. Then the second problem is nested vendoring. |
Following this course, in the early days I almost agreed with Google. Like yeah, you don't really, but you forget that in the early days there wasn't a lot of libraries, so you ended up writing a lot of your own stuff, so it just really wasn't a problem. |
**Peter Bourgon:** And the libraries that did exist were relatively small, so you didn't have this huge vendor tree problem, you didn't have this "dependencies of dependencies of dependencies" problem. |
**Erik St. Martin:** Exactly. They were relatively small, served a very distinct problem, and you could either copy the one part of code you needed, or... It was very small, but you didn't have this tree of vendor directories, where it's like "Great, now I have Kubernetes vendored in, but how do I get it to recognize t... |
**Peter Bourgon:** Exactly. So a lot of people saw this coming, and a lot of really smart people started developing tools to manage the vendor directory as what we kind of all settled on as the place to store your dependencies. That's fine - you can check it in, you can not check it in... It kind of doesn't matter. And... |
\[32:00\] Unfortunately, that was I think in retrospect quite naive, because what ended up happening was we have 13 standards of ways to manage dependencies, and that means different file types, different file locations, different file formats, different behaviors in all the tools. Broadly, you can say that they do ver... |
In the end, it means that when you publish a package and you have dependencies and you want to - I guess it's a pretty sane thing to want to do - bind your code with specific versions of its dependencies, then you're necessarily kind of opting into one of these tool ecosystems, but there's so many that in order to cons... |
So it's a total mess. And while you can always find a path through the storm and solve something for your specific use case or for your specific project, there's no single coherent, teachable, simple way to solve this problem generally. So that was the state of the world. |
**Erik St. Martin:** Even if we all as a panel said you should use this one, it's not gonna matter because of that whole nested dependency problem, right? I use one, but I import a package that uses another one... |
**Peter Bourgon:** I think it was super naive in retrospect to say "The community is gonna figure it out", because what happened was you have these camps, and they kind of ossified and for good reasons - everybody has reasonably good reasons for choosing one style or another, but now we have all these competing standar... |
This is the committee that I was driving as of a couple months ago, and that's what we're in the process of doing right now, to kind of eliminate this heterogeneity in this space, and hopefully help a lot of people. |
**Carlisia Thompson:** So just to get things clear, you're not the head of the committee anymore? |
**Peter Bourgon:** Right, so I was never on the committee... Basically, what happened was I had a number of weeks at my day job where I was working with people trying to get a handle on the dependency management story, and was being confronted with all these tools and workflows, and they were broken for our use cases i... |
So I just kind of said, "Forget it... I think I have enough political capital in the Go community that I can kind of anoint myself the person who's gonna figure this out, but I don't have enough political or technical know-how to actually do it myself." So I said, "I'm gonna be sort of the communications director, or t... |
**Carlisia Thompson:** \[36:08\] Alright, so you're still heading it, though. |
**Peter Bourgon:** Yeah, I guess. |
**Carlisia Thompson:** You're heading the initiative. |
**Peter Bourgon:** Yeah, you could say that. |
**Carlisia Thompson:** And what is the state of the prototype that's being built? Has it been made open? |
**Peter Bourgon:** Not yet. The workflow was we first in a Google Doc describe what I want the process to be, we took some feedback on that (it was great), then we picked the committee, and it actually ended up being a core committee of Andrew Gerrand, Jessie Frazelle, Ed Miller and Sam Boyer, whom you know from the ex... |
So they're the core committee, and then we have this sort of trusted advisory group of the authors and maintainers of the top four tools at the moment, which are Matt Farina of Glide, Daniel Theophanes (I'm sure I'm butchering that name, I'm sorry Daniel), he's the Govendor chap, Dave Cheney of gb, which is kind of the... |
**Carlisia Thompson:** Gotcha. |
**Erik St. Martin:** And to clarify - this is going to be kind of a new approach based on the knowledge of people who have already developed tools, and not some way of interacting with these specs that already exist. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.