Kubernetes SIG Node, 1 Feb 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Node CI 20230201

Description

SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk

GMT20230201-180516_Recording_1388x1120

A

Hello, hello, it's the first day of February 2023. uh It's a signal. Cia weekly meeting, welcome everybody.

A

As I said at uh Brian joined hi Brian, um so we have a few agenda items today. Do you see my screen.

B

Yes, I can see it yeah.

A

It looks good perfect, no black skin today. um Okay, so um I have two agenda items. First, I added it here is um I I've been looking like I've been doing. Some cherry picks for uh for the bugs that we had in HTTP connections uh and I was wondering if all the Test passing for previous releases and I discovered that there is no test for people to say leaders.

A

um So it used to be that uh if you go to test grid here, if you go to Chic release, you will get in the past. There were like a bunch of tests in um informing category to go to informing, and you would see some uh signal related tests today. There is none, uh meaning that, like we but don't run any tests, uh it doesn't seems like we're running any tests and if you go to uh signaled.

A

Then you have similar things so that it is blocking, but it's all on, uh like 123 is the oldest and like I, think it's just left over from something and I believe that those ones were used to be created by a release team. They used to create those for second forming and they also kept it in our tab.

A

um Now it's not created, like I, think uh somebody dropped the ball inside I'm like oh, maybe it's just for I read too many times and like release team decided to take it oh away so yeah. We have this uh small problem that I I really want to address like I think we need to fix it and I wanted to ask you for anybody here on a call aware of like when something happened, because it probably happened. Well I've been on Parental leave, so maybe somebody has some context on people.

A

C

A

um I would suggest you started small, um maybe and some not conformance.

A

Security is blocking.

B

I think we can have a grid for each um for each of the supported versions with at least as you said, conformance tests.

B

So we can have at least a very small signal of what's happening. There.

A

Yeah, this would be nice.

C

um Mike, did you say you you're gonna make some or did you say you found some? What.

B

So I'd say it would be a good thing to have uh we don't I, don't.

C

B

We have any at the moment right. It's probably just configuring a task return, conformant tests with certain pixel versions,.

B

Okay, if the window have an issue, I can I can open an issue for this. Thank.

C

A

And we also discussed with them like before that uh we like, when we discussed operating systems, which runs uh which we run against. uh We wanted to check. Like recently, we uh we have many tests on C group V2 and for C group V2. We needed uh more up-to-date operating system, so we like switched to um cost 101 I believe on special version of Ubuntu uh I, don't have on which version, um and um we now know that that's there, but then I like as I, was looking at previous release.

A

I I was uh realizing like do we update like do. We need to update course, version to um previously versions of releases, so in other words, if there is a like test that you create for 124, we will run on latest course like we will run on course. That was at a time uh more than four uh 124., so I think we need to lock this uh releases on specific uh family of open operating systems. uh This way we can guarantee that there is no regression, because new.

D

A

System will be being used for old version of kubernetes.

C

Yes, that sounds right.

C

A

And this uh begs the question: what do we need to do for um uh current Master Test like what will 127 be running against a long term so, like we have one half year support window in open source for kubernetes, so we need to decide which operating system to use Mike. Can you take the section item and decide.

B

Yeah I can't create an issue for that and decide as well. We may need to set up some policy for future versions as well. um Yeah, let's, let's start with the with the correct one with 127., the upcoming version.

C

Sergey, do you see that comment from Harsha.

A

Yeah I'm about to uh watch it, uh so let me write it down and then, like I'll.

A

Saying that uh he has a problem with the microphone. Sorry, um uh let's talk over chat, so we were thinking of heading one, not E2 conformance to the release. Blocking. Do you mean release blocking that is a sick release, working or seek release Broken Top here we have two tabs like one is. What we think is the release blocking and one is what city release uh observe as a it is broken.

D

Why is there two different? You know the history there.

A

um Mostly because CG release wants very specific signal and they want to control what like they want to know what to look at. They don't want to go through to every single, seek and check what's going on, um so they uh created their own tabs.

D

So I would say uh if we need, like I, think the eventual goal is for proper sick release blocking, but if, as a way to signal that we need to have Sig node release blocking first and then prove the stability there and then upgrade it to proper Sig release. I think that that's that makes sense.

A

Yeah I think that was the ideal and um why is it C group E1 and not V2, uh go through release blocking? Is there a specific reason.

D

For that I I think it was just chosen for, as you know, one I mean they're they're, pretty similar in stability, but secretion we've been using C or C1 for longer. So I think we just have that.

A

D

Also, that does right now represent the majority of our workloads.

A

Okay, we migration to C group V2 so um and I think many other advertising system has C group. It was a default, so um I'm kind of on the uh edge here like uh at one hand we need to test whatever people widely used today, but uh I really want to make sure that V2 is uh stable for release. So people like people who adopting new new operating systems they wouldn't be broken because of kubernetes.

D

But would you be opposed to adding both to our release blocking tab, First.

A

Technology is working that needs to be uh improved and yeah I think we can add both as long as uh yeah wait, yeah right now, it's quite empty. So let's maybe come up with a few and you will keep it on cryotab as well right, so it will be duplication, yeah, okay, okay,.

D

And then, and then eventually we can I mean, maybe by the time that we actually look at making it properly released. Blocking for sick release, uh we'll have moved to see group three, two or maybe it'll make more sense. That would be the V1. You know we'll have more feedback on that. So for now we can add both of them to power release blocking tab.

A

Okay and uh in terms of.

E

Sorry, that's testing Sorry.

A

So, uh in terms of uh release a blocking for older versions, do you want to add those as well for cryo or like we need to decide how many we want to add for uh previous versions like we can start with not conformance on like uh some simple ones, but then like uh I'm, not sure how how much you want to see can I um uh uh on previous versions.

D

Yeah I I would be I would be happy to test still. The versions We need to make sure the stability of those tests, because we've I mean we have been fighting mostly with, like the user ID of uh Fedora core OS has been like the largest pain point, so we just got to make sure that, whatever job that we do add, we've gotten all the fixes that we have in master.

D

But I I see no reason why, like I, definitely think that we would want um you know basically released things blocking across the board crowd jobs as many crowd jobs is. uh You know the state feels appropriate.

A

Okay, uh yeah, maybe Mike and uh pave the boss um by like making it uh for some uh existing couplet tests, and then we can extend the framework for uh cryo as well.

A

That sounds reasonable.

D

Yep that works with me.

A

Cool so Mike uh you assigned yourself up yeah.

B

With the performance test for now, yeah.

A

Maybe you've started with creation issue, so we can like place to discuss but yeah feel free to send the PRS as well yeah.

B

I'll create issues and update them here in the notes. Okay,.

A

And next item is the U right.

B

uh So yeah I noticed that this test has been you know for a while now I think I tried to fix them a while ago, but it didn't get any actual results and I'm wondering if anybody actually looks at them. These are the Fedora, the swap Fedora jobs, they're, both regular and serial. We have a similar job for um for Ubuntu.

B

Both have been passing, uh I think the cereal has been flaking, but I wouldn't be I'm, not that worried about the flakes, because those are flaking in other jobs, non-swap jobs, uh but yeah I was wondering if anyone looked into the Fedora jobs, if not, maybe it's um it makes sense to remove them because.

D

They have been.

B

Red for a while.

D

uh I, don't think we've worked into them. I think uh it could be good to check on them before removing them. um How about I assign myself to this issue and I'll I'll, either figure out if like this, is something that we can, that might trivially look at or maybe what this test Works skipping or then, if not yeah, we can.

B

B

That might be the case yeah and.

D

Do you uh have you found any uh anything with the Ubuntu or the? Where is this on cost uh either of those have you looked into the failures on either of those.

B

uh I think that the the federal failures are because the the tests failed to start, probably in some of the in, like in the setup stage. Oh yeah.

D

We would probably just this is: actually we might even 127.. huh Okay, yeah I'll I'll, take a look and see what the situation is. Thank.

E

You recently we fixed issue in Federer OS, with the SSS keys, we're not getting transferred correctly, so it might be the same issue here.

B

E

D

It's true yeah. We I think so yeah that we've been failing on provisioning on a whole variety of jobs. For this, so yep uh I'll take a look and see how we can make it more consistent across the jobs.

D

A

Thank you. Well, we on our phone memory, swap it's not very critical to have a wide coverage of tests uh closer to Great NGA. We can. uh It will be critical, so I.

D

Mean we can temporarily.

A

Disable those and then come back if needed,.

D

A

D

I'm sorry I I was gonna, say I I, guess it wouldn't really hurt to remove them. If you know, because it is just an alpha and they've been failing on provisioning but uh I'll take a look. um Thank.

E

You I was just wondering if you do, we have a priest equivalent pre-submit job, for it.

A

I, don't think we added any memories. Okay,.

A

This may be a good idea by the way uh Mike. If you plan to work on memory swap, uh it may be one of the items to take a look into.

A

Okay, um this is discussed.

A

um Any more agenda items for today.

E

Sorry I was just wondering I'm, not sure if I, if I got it correctly, so we are adding few node conformance job to seek no release tab.

A

For previous versions, yes,.

D

And also the two cryo, the we'll do, one for secrecy, one and one for secret. Three, two.

E

Yeah yeah, but we are talking about release right, not signal release, Master blocking I'm kind of got confused between those two.

D

Right so we're going for Sig node release blocking to start, and then we can prove the stability of it and then we can petition to seek release to have it on their Master blocking list. Do you think that makes sense.

A

Yep, so that would be perfect. Thank you.

A

Okay, let's quickly go to triage, not much happening there.

A

I cleaned up some of the unrelated issues and uh I was cleaned up. Work in progress.

A

A

um Are you working on that right.

E

Within the problem, hello, yeah, this is working now, so the I'm asking the author to remove the hold. So if you added a periodic job and a pre-submit job or you're waiting for the P submit job to pass, and now it's passing so um once they fix that unit test, which is failing, I think they should just go too much.

A

You also didn't seek release Master, informing, oh yeah, it's because you copied it from somewhere or.

E

I could say: yeah Incorporated. There will probably store him on that. I'll put a comment. There.

A

E

It's a copy of existing cryo job, which is Sigler, is Master informing.

A

Okay, it's like I, don't want Alpha feature to be mastered in for me, yeah I got it um okay. uh Let me leave comments.

E

We're going with the c group V1 we got for this particular feature, it doesn't matter P1 or V2.

A

A

Foreign, because there is no lgm yet I mean I- don't want to put it in equation on author because of this comment fast.

A

A

Yeah, let's check out.

A

All Ryan said that there is a winter storm.

E

D

I have internet problems.

A

As well, but there is no veterans from here- um okay, so it seems to be pulled, lifecycle related and it's the fading Confucius.

A

Is anybody interested to take a look.

F

uh You're gonna assign it to me.

C

A

Okay, there seems to be a small change.

A

E

A

Box, instead of.

A

A

A

C

D

A

We just discussed uh thank you Peter for taking it.

A

And this is uh this was in way to author for a while, and it's not necessarily A strictly related to a node, but it touches some node code.

A

um If anybody wants to take a look, storage files.

C

A

Better error management.

A

Brian, do you want to take.

C

I thought I'd already looked at it, I've been watching him do that anyway, if you need a name, you can put my name on there. That's fine, okay,.

C

I thought he was getting all the approvals he needed on those I.

A

Think he can approve with uh himself it just um ogcm signing to check.

C

A

And yeah I will move it too.

A

Yeah, we're done with this.

A

So we can also review news reviewer. We didn't review this for a while um yeah this is it just looked uh this we just looked this. We just uh looked yeah so uh event, plugin CI, yeah,.

C

A

uh Brian I think you looked at this right before yeah I see your name yeah.

C

I put something else: TVs Nikita's right.

A

Yeah, so maybe you can uh yeah it's.

C

Which one I was, if I didn't understand, do I need to take an extra step. I thought I was done with that.

A

uh Apply the GTM once.

C

A

Yeah um you're a member of community right, so you can apply okay,.

C

You can be.

A

A sign so yeah you can be a employer.

A

So once you apply HDM like approvers, will likely like more likely to take a look.

C

A

uh Unit tests, windows.

A

A

Looks unfamiliar for me.

C

A

What's happened with this one.

A

It has a vegetarian.

A

A

C

A

Is your PR right I.

C

Was I was ldp.

C

What's the title, I can't see it.

F

C

F

um This is going to be replaced by Clayton's um yeah.

A

A

Do you want to close it yeah.

F

Let me I need to make sure that clean's gonna put the test into his PR, so uh I'll do that and then I think we can close them out.

A

A

Device manager- this is what we discussed uh with Francesco last time. He.

D

A

To take a look.

A

Okay, um yes, I still need to take a look.

A

That was the last comment two days back. Okay, let's give it some time.

A

I, don't need more, it looks like see, storage, more than signaled I wonder what changed the signaled.

A

A

C

A

Seems to be minimal, I can take a look.

A

Yeah it's another test by swatchy uh and we have Francesca assigned so also I actually have all GTM already.

A

A

um Oh, this is new cap um I mean not a new Gap. It's uh they're going into beta 2 for unkilled reasons.

A

There's a lot of comments: what's the latest.

A

Okay, a bunch of characters failing.

A

A

Yeah I guess we're not original somebody else.

F

You should link this to Clayton's stuff. It's probably related.

A

Yeah question isn't here.

F

um I can do that offline. You can assign it to me.

F

No, it's R Phillips, oh.

A

A

And this one don't have LG TM.

A

A

E

A

I think we can review other columns next time as I typically do randomly.

A

Is there no more comments for tests? I will go to bug, search.

A

We have only seven, so maybe we can uh finish early today.

C

Looks like a lot.

A

Seven, it's not the one is 27. Then it's a problem.

A

Oh yeah, still on me um what's happening with that: oh yeah, okay,.

A

A

Yeah maybe related to clothing changes as well, so Ryan What's Happening Here is um they have a port with config map and then they stop a cupboard, um some delete, some config maps and then restart the qublet, and then a config map is not being cleaned up.

A

The question is uh we'll create and changes addresses because derive comes.

F

Likely because he's gonna be changing volume imager a little bit too um so likely you're gonna assign this one to me too, and clean.

A

A

You know profits unable to update apparent profile attached to a port.

A

A

About to see a farmer um and I would I think.

D

A

To be taken, a look.

A

A

C

C

That cubelet would manage shap armor profiles.

A

Yeah completely can apply, apartment profile need feature.

A

Okay, um I will take a look later to understand whether we need a signal or not. So this is something we discussed yesterday on.

F

Right I talked to opaco about it, and one of his PRS merged he's going to look into a into end to end Suite running Upstream about uh upstream, and so um it's in progress you can, you know, assign it to me too.

A

So let me assign it to you, so it will be.

A

um Okay, starting while CTR image note.

A

Yeah, it's clearly this issue I mean not clearly, but so many symptoms.

A

124 very fresh.

A

A

A

D

Believe there is a.

A

Prn fly that will fix it.

A

A

Yeah, maybe I will take it offline and find the pr.

A

So I'm not doing it online, because everybody's gonna call.

C

You're, making good progress on these.

A

Video past an army that that might pass is not working. I'll share, Protestants.

A

C

Statistics package as far as I know well.

A

C

Machine learning people use it.

C

I, don't see how a shared process namespace could result in a resolving path. Problem: I, guess we should read the data, though.

A

Too many levels of symbolic links: okay,.

A

Somebody fixed it out of Frozen.

A

Okay, so I think the main question is that kubernetes problem or it's uh our problem.

C

Really doesn't sound like a kubernetes problem.

A

A

September 2022 so yeah long time ago, anyway, uh we're done with. Is there any other topics today.

C

I got a quick thing: it might not be relevant. I have a a ticket. It's got LG TMS on it. It's for backboarding the registry change to 123.. It's just sitting there I, don't know why it doesn't advance to The Next Step. Should we discuss that? Can you send that ink yeah I'll send it over.

C

I'll put it in the zoom chat here.

C

It may not be urgent in anybody's mind, but I want it out of my Cube.

C

I was working with Ben on these at the time.

C

You've already LG tm'd it too I just wanted to move ahead.

A

And this is for 123.

A

Okay, uh let me show me more now and.

A

Really quick and we have the same for other versions.

C

I was it just doesn't seem anybody's interested anymore. I was expecting to go ahead and do it on other versions. It's because.

A

Of budget, like we closed the new year yeah, but it got renewed I think.

C

Think that they're not so concerned anymore.

A

People know if that's one anymore, just not as urgent as before, and that.

D

A

um Okay, let's we'll now approve it and then we'll get uh reduce managers. So there is a comment about that. These managers need to be verified like this. Okay,.

C

A

You have an approval, we can get his managers to it all.

C

A

Thank you, bye, bye, everybody.

C

Thank you, bye.