Kubernetes SIG Instrumentation, 27 May 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Instrumentation 20210527

Description

SIG Instrumentation Bi-Weekly Meeting
May 27th, 2021

A

uh Welcome everyone. uh This is the sig instrumentation bi-weekly meeting today is may 27th. uh We have. It looks like a few agenda items uh merrick it looks and and damien you guys have uh the first agenda item. You guys want to go over.

B

It I mean you wrote it, but I can talk.

B

uh Okay, uh so yeah on some recent discussion on one of the prs on metric server. We got to pretty quick agreement that we could do better on the api server or custom metrics api server library um to make it easier for uh also other uh make.

A

B

Use it so the problem that we found out or discussed was that prometheus adapter uses both custom metrics uh observer library to serve apis custom, metrics and external metrics, but also imports matrix server and uses part of this code to to don't re copy.

B

The api machinery part- and this basically creates a problem that magic server is bound to or can cause breaking changes, even though it was not promised, and this definitely us more component start for implementing um core metrics as part of their adapter, and that would be great for to simplify for those use cases and and and allow just an outer event adapter just to use one library and and decide what metrics that they want to provide and not don't think about like uh finding out that there is also like totally separate word.

B

uh So I think I originally proposed the name, but this is totally uh a name that is one working process of creating a metrics api uh could be a metrics api machinery or maybe han knows more because api people have like. There is a lot of project with the similar naming and there is definitely naming his heart a project to move. I think, basically, we would move the uh the metrics part of api machinery into custom, metrics library and rename it so it's.

B

It could be more obvious for people how to uh where to look for. Is it version right now, uh which one.

A

B

Yeah, I think there was a proposal. I think that must know to use the kubernetes uh versioning. This is, I think, related to that most kubernetes libraries are exposed from kubernetes staging us, so they're attacked as kubernetes, and this library, like is external, so it's not, but I think there is a proposal to do that.

C

Yeah a there is api server to kind of had some version in there, but they've still nothing planned. Yet they still need to be tackled.

A

Okay, um yeah, I mean I I'm okay with it like. What's the ask here, you just want a repo.

B

Everyone wants the ripple. uh No, it's just a discussion. It's just like make sure that those changes are visible. Either public people can have uh maybe uh yeah at their thoughts. uh Maybe naming could be outsourced if someone has a idea- and it may be also, someone wants to help with that.

A

I mean wait, this is already in a separate reverb, though right. Yes, I think.

D

That's just the question was to to move the abstracted resource, metrics api api machinery out of metric server into the. What is now the I think, custom metrics, api server. Repo um right is that what you meant yeah.

B

Yeah that, like this is the code change. We are just discussing like a part of like the low-level code, changers of also thinking about breaking changes or renaming to make more it more useful and more discoverable breaking changes. How we can handle those uh and all like if there is any problem that we don't see or we could maybe not don't, do a breaking change or like yeah yeah. We want it would be good too.

D

So this is, I think it.

B

D

Always the plan like, I think it's just something like sully- did the refactoring like three years ago, and I think we said that the next step would be to move it, and I think we just never did it so like I'm totally up for this.

A

I mean this is basically a component base right, more yeah, yeah.

D

It's a bit more general than that or basically more specific than component based, but it's yeah.

B

But there now there is a component api server like there is component base and there is a wrapper for component base for api server, like or controllers like it's. We are building the layers, yeah.

A

uh Yeah I mean it sounds reasonable. um Is there anything that we need to do in order to get this.

A

D

I suppose we need to request the repo name change.

B

Any concerns about like breaking change is it for people that impart.

A

I mean maybe we should just create a new video migrate, this stuff and deprecate the old one. That way it doesn't break for.

C

New one and bring both code base into one place and even like custom matrix education hasn't been detached uh in a while between the whole refractory. In my opinion, or at least some refresh having.

D

I think I would leave that up to whoever wants to do the work like how you want to do. It is up to whoever does the work like. If you want to do it then, and you want to do it that way by all means, you have my approval.

A

Yeah, but uh one of us has to actually request the name, change or get the repo, so it would be helpful to know which one of those we should do.

B

I I don't think there is like you just need to approve. I think you don't need to do that like requesting yours like yourself. If, if someone wants to take it like we can yeah just create an issue, maybe someone can pick up and it anyone can propose. I think we we migrated that original request for group, state, metrics and metric server.

A

Okay, I mean I, I think it sounds reasonable.

B

Okay, I think that, oh, if someone wants to propose a name, please, I will leave you someplace.

A

Okay, so we have two more agenda items but elena's not here, so we should probably move it to the next session uh when when she will be here, uh there's one thing, though: I wanted to talk about uh david, specifically uh the metrics grabber thing I don't know, what's going on there.

E

Oh yeah, so actually it looks like there was a change to make a bunch of the um control plane, pods, insecure ports, disabled and they added in order to accomplish that they added a uh proxy to the test and the proxy is super duper flaky. I actually, I'm pretty sure you were on a reviewer for one of.

A

E

A

Yeah yeah, I was just wondering what was going on.

E

So they they broke it and uh it sounds like they're trying to fix it. They. So it's not on us uh yeah. So we we have an issue open right that we all got pinged on and I cc the person who made the change and they linked to a different issue that, where they're tracking progress so there's 101 894, which is the r issue and then in it uh they link to 102 o 50.

E

whoops that didn't work.

A

Well, I thanks for looking into it. um Elena was uh mentioning that this was kind of a problem, so I thought we should probably talk about it, uh but it sounds like it sounds like it's not on us. um It sounds like it's. Whoever owns the proxy.

E

You're assigned to the remove metrics proxy pr right now, han, which is supposedly the fix that they're referencing for it.

A

Oh didn't I did I review this already. It needs a rebase or something so I haven't looked at this.

A

This is quite bold, uh no and also they're, saying that it's not a flake.

A

Like yeah, this is, is this even resolved? I mean like? I don't even know if.

E

Or they may, it looks like there's a separate attempt to deflake.

E

The metrics grabber tests that merged six hours ago, so let's actually go check and see if that.

A

Helped are you looking at the uh the link that elena generated yep.

E

A

E

And we get to find out whose internet is faster.

A

E

A

That's that uh for everyone else following along this is the link. It looks like it's still happening.

E

A

But again, this isn't really us right like this is like some random proxy.

E

Well, that people.

A

E

We own the tests and someone added a proxy, so we can either ask them to revert the change that added the proxy, but I it's been almost two weeks or were responsible for fixing it probably but they're looking at it.

A

E

Just going to comment on the issue saying that it's still failing.

A

Yeah, it will it's still failing, and also um this remove metrics proxy, which.

A

Is I mean I I don't understand what I'm supposed to do here. This needs a rebase and the the comments make it sound like it's not actually ready.

E

Yeah so in the issue the flaky test issue, they say they're working on it in that remove metrics proxy pr. But there's some problem, that's blocking it. They call it a debugging handler problem and they ask if they can merge a separate pr to mitigate it, which has already merged, but it doesn't appear to have fixed okay or maybe it may also be. Do people have to rebase on top of their fix before.

A

Yeah yeah yeah, I think they have to rebase. I think the problem is this: this pr has not been rebased after after that fix.

A

So we can just, I guess we can just ask.

A

A

Okay, all right, uh I guess that was it. Thank you for updating me. uh I was.

D

A

Not up to speed on this one, uh okay does anyone else have any agenda items that they want to talk about? No okay, then I guess everyone gets 13 minutes of their life back thanks everyone. It was nice, seeing you all right, bye,.

D