Kubernetes SIG Node, 13 Dec 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Node 20221213

Description

SIG Node weekly meeting. Agenda and notes: https://docs.google.com/document/d/1Ne57gvidMEWXR70OxxnRkYquAoMpt56o75oZtg-OeBg/edit#heading=h.adoto8roitwq

GMT20221213-180519_Recording_1920x1120

A

Hello, uh it's a signal. Sigmatic. Limiting uh today is Tuesday December 13th 2022. welcome everybody I think it's the last meeting of uh this year. uh We can discuss uh next meeting later, but uh probably will cancel it so first and the most important topic for this meeting was 126 retro and 127 planning I congratulate everybody with release of uh 126.

A

um It's time to start doing. 127. oh before I did that um I want to say that uh so for all right, PR.

B

A

So I think deems is doing a great job of cleaning up PR's um Ronald I know approving stuff I closing PRS that are not very needed to have some long-standing issues, um so yeah we're down to 196b below 200, which is a great news. uh Let's keep working on that and I think we can clean up a lot of things before New Year.

A

So with that, let's go into our 126 letter and I need to share my screen.

A

You should be able to see my screen now.

C

A

Hey Let's uh walk through the list of uh caps that are even prepared for us uh that were immersed in 126 and then once we go through that we can discuss what went uh well, what went uh badly so we can help ourselves with the next cities.

A

So first Google credential provider, we keep making progress. This is not the very important cap because we want to minimize the three and three dependencies, so we want to like this is needed for everybody. Community will definitely benefit from that, but it's hard to find people for this kind of work and I'm glad to be making progress here.

A

Metrics uh we did some progress on metrics as well, which is a great news. There are any comments on this cap.

A

Okay, no comments here, I invented flag, so we keep making progress in uh uh minimizing kublet I think event will help us a lot in terms of Google diver hat uh we entered Alpha, which is, uh if you want to, try it and give feedback on especially on reliability. uh Of this events, it will be great Dynamic resource allocation, um yeah. We uh it progress here as well: um uh solid, uh solid uh improvements around uh devices.

A

um Port failures uh it gets into beta, um which is very good news. uh Many HPC uh workloads will benefit from that by Saving resources. When something got started for the night and suddenly I discovered some issue, we will not waste all the customer resources any longer as they can configure it properly.

A

um Okay, multi-numa alignment, topology manager, I have very little to say about it. Beyond that, topology manager needs to be uh G8 soon.

D

um Actually, this is slightly different. This was um an addition to topology manager and uh topology manager. Policy options were introduced as part of this cap, and this specific prefer close. This pneumonode was the first policy option that was introduced.

A

Yeah, um what I meant is uh topology manager itself would be great to GE and I knew, which is always great yeah.

A

um Thank you, Swati uh CPU manager, um uh G8. So we started the GN managers uh we already discussing um uh resource manager, plugins and the first step is to graduate whatever we have entry. So we have a solid foundation to build on so CPU manager, device manager will graduated yay um yeah. That's all uh caps, I. Think historical statistics is uh we're doing.

A

Okay, uh we're doing I would say average average release uh in in terms of number of caps um and I'm, not sure whether we have data on how many we missed uh yeah I think we missed a few. uh At least I missed some uh myself yeah.

A

So if um anybody has a comment on what what went well and what uh didn't went well, this series, please speak up.

E

I'm trying to look at the top here, um I feel like we've kind of settled, into a sustainable pace like between maybe eight and twelve things in a given release.

E

um I think that's good. It shows like some consistent velocity I'm curious if people are upset by that velocity.

F

I think that that's what I think that's a capacity direct and.

G

I think, depending on the.

F

Size of things we just yeah go up or down that number.

A

I can say that one thing that didn't go well is uh we needed to ask Mark to apply Milestones to all caps uh during the flying stage? Mark um I, I can't say more things to you, but I think. Thank you for your time.

H

No problem, yeah I, think that the there was a little bit of issues this round with a bunch of sigs not like uh just with the chairs and TLS, not having the right access, but we're getting that sorted out on the release team.

C

A

Was there any troubles with API reviews this time around I feel that we only had like one cab that needed some API review, but uh it was CRI API, not the API API.

F

So dra was big right and I'm sure it had some API components, but.

C

F

Don't think we had problems but yeah I.

A

Would put it as a good thing? We didn't have any problems as API reviews.

H

I think prr reviews too.

H

I think there was a less like delays for prr reviews. This release.

A

Yeah, it's fair um boss went very very smoothly, even though we had some last PR reviewers this time around. uh It still went very well.

A

How about uh I haven't been around during the uh quad freeze stage? Was there any troubles getting things in last minute? I remember. It was constant the problem previously uh and we even introduced the soft fees at some point to make sure that people, not latent caps, be around to the last minute.

F

Yeah I think, like some things, did learn during the last weeks and that's always going to be an issue given folks have multiple things they are juggling. We.

B

F

Can always aspire to try and merge things earlier in the cycle.

A

Did we use soft freeze this release? uh Sorry.

E

I, don't I don't think we did I mean some of the items we tried to push are along so like um I think until we get the In-Place resource resizing thing over the hump, we should always identify that as like I could have gone better. I know that but I think. Actually we did a great effort to to get that uh ready to be good to go.

A

F

With getting the the CI stuff sorted out, so I think the pr is in a better state. So hopefully we can get it in sooner. In 127.,.

A

Yeah understand we had some CI issues uh this time around right. We had more test feathers.

A

Especially with C group with blue transition David, do you want to comment you unmute yourself.

I

um I think the secret V2 stuff is actually a little bit unrelated. That was just a test in for change. I, don't think, that's related to the release necessarily um I'll, just kind of just we were trying to update the OS images um just happened to coincide with the release, but.

I

I can add another one here, I think in general, I think um looking overall I feel like the Sig in general, has kind of uh capacity for kind of one really big cap at any release. So, like I, think this release, we had two right. We had the dra and the in place resize and so, like I, think we had to pick. You know. Dra was in kind of better condition and and uh more tested and more ready at that point, uh but I think that's something that, like we'll have maybe moving forward.

I

Sometimes we have to have. We have to kind of prioritize between two two large caps.

I

It's hard to find bandwidth for two like that.

A

Wait for next release, we plan to do sidecars all.

F

Right so I think one item on the event well is I, think good collaboration across folks in containerdy and cryo to get evented like like uh over the hump.

E

Yeah I was trying to think about um the the budget comment um I feel like we struggle with a budget on external facing um changes, but we've seem to do better with internal facing stuff, so, like I, think the Plex stuff, even when it's spam components outside of the direct Sig here seem to work smoothly and I. Don't know if that's just a symptom of it's harder to change the API of kubernetes than to um improve the internal implementation.

F

I think makes sense to me I think direct. Also, given the conversations we've had and had around like sidecar like agreeing on the API and the UI is always harder.

E

How many CRI changes actually happened um and if they were actually yeah.

A

You don't think about it. I have a agenda topic for this um and it has a list.

A

So I wrote some CRI policy outline is a problem, and uh so in 1 26 we had um uh window spots and books uh host Network, um uh just documentation, change and event at black. So we had four big I mean three big and one small change.

E

Yeah so I'm sure your cup is going to say why that's too big or too bad or too too good or too awesome. But um at least uh the transition to the CRI doesn't appear to be slowing our ability to evolve too much.

E

Yeah I mean part of me wondered when we moved CRI to V1. If it was just going to be done, but it seems that we um are open to continuing to to iterate and add capability.

C

A

A

um Okay, is there I think we at time I wanted to allocate like 15 minutes for this uh discussion. If there are any more feedback on how care process went uh or you want to share it in private, somehow we can start a slot conversation um I, think action, items I, don't see any specific action items out of it or Beyond just do better planning and maybe try to do soft freeze again uh to um just uh advocate for faster and earlier reviews.

A

Any more comments on that or we move to 1.7.

E

I think the only thing I I'll just say, I openly struggle with, and maybe we could all figure out ways to help is um people will reach out to discuss their cap and whether or not you know all the things that we have in a release it. It is hard to keep all this in our in our individual heads and so.

E

I'm I'm wondering if uh well one we should probably think about as we get to 127 if we can expand the list of folks, maybe within domains that we are comfortable taking like approver on a cup um uh given.

E

If we look at the past two releases, I guess but I just know personally, it can be sometimes hard for me to um keep everything in my head and for those who I give feedback to or I'm like you know. Can we just pick one of these two or three different things right now? It's largely just a consequence of trying to absorb everything that people bring forward, but.

E

Just top of Mind here that we, uh if there's people, have been working on particular domains, um we want to continue to refine it like I'm thinking about the device space. We did the Resource, Management, Group and stuff, maybe um Kevin or Swati or others can start thinking about you know. Do you want to take ownership in that that area around caps going forward? Just what can we do to give a little more specialization so that um we.

J

Can all save some.

E

Brain capacity.

E

I pick on the dra one, because I think that was one I gave feedback there, which says uh there were like three or four different concurrent resource management, related ideas and personally I struggled to absorb all of them.

A

Okay, uh it's a good thing, uh good addition, so we have two of them like because it's uh I think are you mostly talking about earlier review by by experts so like there is less details to think about.

E

To grow people's scope and a Sig right, um okay, uh and so you know uh we're all working on really complicated areas, and so, like you know, particularly as we started the resource management group, maybe we can think about ways to grow enhancement. Scope with that, because.

J

E

Are bottlenecked on on caps generally by I think maybe two or three people? One of them is me.

A

And there is some also some secretarial work as well. uh That needs to be uh helpless. um Okay, is there any more feedback.

A

Perfect, uh let's move on to um 127 planning.

A

Oops not strong one. uh This one.

A

I think what happened is uh Raven took everything that didn't fit into uh 126 and put it here, and people started adding more uh caps in the table.

A

K

A

Bigger for everybody to see now, do you want to go through the list? Typically,.

F

Yeah sure I mean we can oh yeah I think maybe we could have ordered it, but it's okay. Let's try to get through as many of them as we can, because I think we want to give priority to things like in place vpa that didn't make it in 1.6 uh all right. So, first on the list we have cubelet plugin model based of dra uh Marlo. Are you around two basically.

L

We have a park that works, but we haven't based off the array, so we have some more actual work to do to get that through, but we have dedicated resources to work on this and I. Think we've spoken already: we've had our first working group meeting with that included Derek Swati and Kevin and Sergey we're all there and a variety of others, Francesco, um okay, myself and a Thomas, um but we are also happy to help take some load off going forward.

L

um If there's other ways, we can help with other things, just to minimize the impact for reviews, Etc.

F

M

I mean I think this.

F

Is a great meeting where you know we? We are looking for folks to sign up for reviews and approvals as we go through the list so.

L

Yeah, so we're happy to try to make it so that we're we're minimizing load on the team there. That would be helping us Shuffle work around. So, if you have ideas of how we can we're happy to.

F

So for this particular cap, do we have uh reviewers and approvers identified.

L

E

You consider the cap actually ready, or are you viewing this as something you want to get into a good shape in this set of meetings we had this week last week, we.

L

Want to get into good shape because we wanted to make sure that the shape of it was was what the community wanted. So we still had one more meeting on Thursday to go through that and then we can, we can run for it on code. The code is not ready yet.

B

Like meeting minutes or recording coverable somewhere.

L

There is the recording um we need to get that to Sergey to get that there and there's also been multiple posts in the Channel with uh times and requests for people to give times. Yeah there's.

E

A calendly linked I think, but um there's only been one meeting Alex to my knowledge.

E

Yeah, so I, don't think you if you missed something, it's a shame but uh yeah, please join Thursday. If you can and I still think, there's more discussion, but for those who weren't there, it was a general desire on what we can do to allow plugable resource managers um for those things that we don't just have entry or maybe ways we can involve the things that we have entry to fit a plugable model and it was a good discussion, so I think yeah great to have you there.

F

So maybe like we need to get the kip in order, and Alpha is the best best case scenario for this one.

L

Yeah pretty much.

D

You can add me as a reviewer for this cap that helps.

E

Yeah um Swati thanks so much for volunteering I also, and then I assume Kevin will be pretty active here as well.

F

Maybe yeah we can but Kevin is he up to what I guess.

G

A

C

A

Move on to the next one, yep.

F

Is not around his Sparkle around I know like they have uh they have a hack, MD or a cap here, I'm.

A

Happy to be reviewed here if anybody wants to.

F

Yeah, you, you can add me as a brewer.

F

This seems simple enough, so I think we should be able to review it.

A

Yeah, that's uh it's very straightforward! Improvement that you've helped by with image, pools.

F

All right next one is extend the Pod resources. Api uh Swati. Can you talk to this one.

D

Yeah I think Francesco volunteered to work on this one, but in general this is like a resource endpoint within cubelet and the API itself has been graduated to GA and we need to update. We need to graduate the feature to GA and I'm sure Francesco is here to talk through it.

J

um Yeah but I think we'll. This is another one. This is about exposing the dra allocation.

J

You're right, though, I was volunteering to review and help.

D

Oh sorry, I I completely misunderstood this one.

J

Yeah, no, it's fine! It's fine.

D

J

Mean it's close enough, so yeah. This is about uh pretty much. What I know about the brief quick conversation with Kevin. Basically, they want to expose the area location through the existing introspection, endpoint and as soon as they get more details from Kevin I will be reporting or he will be reporting. Someone will be reporting and yes, I will be happy to review.

F

All right sounds good.

A

Or Kevin can be its owner.

A

F

So I'm going yeah I.

G

F

Made a pass at this one that seems again simple enough. uh We may need some API review so, but from the note side it's seems straight forward.

A

Your next one is fine grain, supplemental group, yeah.

F

Yeah, so that's the one I was talking about it's okay, so that one is like yeah straightforward enough and I can be the reviewer and approve around. If anyone else wants to review as well. Happy too work with.

A

D

Yeah for this one I remember last time we were having we kind of were double checking and reiterating on the fact if you want to move ahead in this direction, so I wanted to discuss that. I have an agenda item as well about that. But if everyone is happy, I'm I'm happy to volunteer graduate this feature to ga.

E

I think we should graduate to GA and um anything we're discussing on the first topic that Marlo raised up needs to be um inclusive of evolving. This feature right so I, don't think because we're discussing those other things we should stop signaling to the community that it's safe to use this function, given its blood usage.

D

Yeah sounds good and I yeah and like given the discussion we've had. It appears that some of the existing resource managers are going to transition into resource plugins. So even with that into taking into consideration, I think graduating to a graduation of this feature to GA makes sense.

E

Yeah I think all the managers that are injury should graduate to GA and if the evolution of those things or their implementation approach changes it. We discussed yesterday right where that could just be a built-in plug-in.

K

E

um But uh I don't want to put fear in people's hearts that it's not safe to use this function. Given that I know it's widely used.

A

Policies says like we have new policy.

E

uh Maybe it was a part of the resource, plugin stuff. We can think about um how that makes the policies options not need to be in Perpetual non-ga, but I think we kind of recognize that that was never able to graduate when we had the discussion so.

D

Yeah and and the policy options they have a separate feature again and separate process. All together, like the individual options, have different graduation process and the policy option as a feature itself should have its own graduation process.

A

um uh Perhaps us you can add Marlo Marlo. You can recommend somebody to as a reviewer for this cap, so it will be um so be RDU From perspective of resource manager, plugins.

L

B

If you want, you can cook me, I can try it.

D

D

And on the approvers I put in direct and on I'm, not sure if there's anyone else would be more suitable for this.

E

I'm fine to handle this.

D

All right, thanks, Derek.

F

Actually, on cubelet power resources to ga uh Francesco.

J

Yes, this is a simple one: it's about uh the the cubelet server itself as a feature gate to actually expose an answer to the uh padres's endpoint, whose uh API was uh already GA in 120.. Actually, it's on me, which I didn't notice this before. Thank you to sergiato, not the internet. Is that so we want to graduate the server endpoint, so the cubelet can actually answer the apis, which should be 90. I expected, is to be documentation and handling the prr review so code. Wise expect this to be very simple.

A

Back in the review.

J

Thank you maybe.

F

Kevin can be the Opera.

J

Yeah about caring I, he raised comments about uh the interaction of this work with dra work. He was proposing and he's doing they are doing. I do think there is pretty much no interaction, because the the actual interaction would have been with the producers API proper, which is already GA, and this is there are conversation in progress. If there are any updates on this area, I will notify you folks.

F

Sounds good. Thank you.

F

All right so in place update, uh we uh I mean. Definitely we need to work on this one get it much so reviewers up to us. Maybe direct you'll cover the final approval for this.

E

Yeah sure I think I did okay, okay,.

F

All right, so we haven't heard any updates on this one uh and we didn't hear anything in 126 either. So this is basically waiting on author. This is split STD out in the studio, lock screen, so yeah.

F

For add app, armor support, I think it just it had missed the deadline, so we'll have to poke them all clear again and see if he has the bandwidth to work on it. 127.

A

F

Okay, uh the next one is uh field. Can you open that I? Don't remember where we ended up on this one.

F

Okay looks like it's waiting on authors.

C

F

Being yep, we need to Ping.

F

It's a sub second, more granular, probes, uh Mike Brown! Are you around.

F

Mike isn't on the call we can follow up with them.

C

um I can sign up for review that if it actually.

F

Okay, thanks Daniel.

A

Thank you, sidecars yeah,.

F

So Sergey you're going to give an update right. Maybe next meeting.

E

No Keystone containers.

F

Especially sauce continuous yeah.

F

All right, Qs class resources, uh Sasha and team- uh is anyone gonna give an update on this one.

B

M

Okay, yes, yeah definitely we'd like to have this in in 127, so yeah I mean they are ready to work on that type. I don't know. Should we have a separate beating like near term on that or I, don't know how to proceed.

F

M

F

Yeah I think like we probably need her to have a discussion on signals to make sure we are all on on the same page. uh So maybe we can queue something up for the next meeting on this topic.

E

Could just be a topic and yeah discussions that Marlo and team were were pushing.

B

It's separated, so it's not related.

E

B

Understand it's a.

E

Distinct use case I was just trying to see if we could add that in I wasn't sure if the meetings that we had yesterday or on Thursday would be exclusive to just Resource Management plugins, because.

B

A question regarding with timing: so if we don't have any more signaled making quiz here and uh the next opportunity will be like already in January, what is the deadline to get this caps uh approved.

B

F

Good question: sorry: did you check I can check with folks on music release.

A

E

M

A

Time, I guess we do.

F

E

Another meeting Thursday though, if you want to join that and I I I, maybe I'm mistaken I, didn't think that that meeting was exclusive to the one subject. I thought it was inclusive of just uh these resource management related topics so.

A

There you want Resource Management in group meetings, I think I.

E

Don't necessarily want that I guess mentally in my head, I thought all these things were coming from our friends at Intel and I didn't quite realize that there were different segments or user populations they're in um maybe I was mistaken. So we.

L

Have different customers.

B

We have multiple teams working on multiple things, not necessarily.

E

Yeah, it's a shared plate that we all feel yeah. I know.

E

So if we don't want to discuss this Thursday, then we could queue it up. As Renault said for the next uh meeting.

N

M

M

Yeah yeah, let's yeah sure, let's let's discuss about this on Thursdays. If there's a meeting, yeah sure I'll give an update there.

F

Sorry thanks Marcus.

K

Yeah I wanted to say something here, so hi I'm, dikshita I work for the node runtime team uh at Google. I might be taking up this cap by the way. So, if you could also include me in the Thursday meetings, uh it would be great I'm just starting to ramp up on this by the way uh brought it up in one of our meetings.

F

All right, you're talking about the Qs class resources right, yes, okay, so I think Derek or Marlo. Maybe if you can put a link to that meeting in the notes, folks can uh join it.

M

Yeah please do or or send to the signal. I, don't think it's just been published there. So if you can't send it to Tamara understand it's easy to pick up.

L

Yeah one one thing to note: is we ran out of time as it was so I suspect, we'll use the rest of it. um If we need other meetings for other things, we should do, though, all.

N

Right, either way like.

E

I said my honest mistake, but I thought uh maybe there was more uh one macro organization within Intel, pursuing a set of topics and I thought that this would have been one of those topics rather than I appreciate the plight of big companies everywhere. So uh uh if, if it disrupts anything, we could just do this on our next signal meeting.

F

Thanks uh so moving on to the next one probe terminating grace period to GA, so Ryan is taking it on uh any reviewers.

A

I can follow the reviews. Oh.

F

I I can be the approval.

G

I

Sorry, just one more thing: if you can go back up for the memory Qs, can you add me and Dixie depletes to the reviewers.

F

So this is not the memory qos divot.

I

Oh, this is not Memory Express, I'm, sorry yeah.

F

Yeah, this is Qs class resources. This.

O

Is intense, I'm, sorry, okay, I'm excited.

F

Yeah yeah, if you need to add memory queue as it should be in business somewhere.

O

Okay, sorry, okay,.

G

A

Which mic and Paul this is, uh if you want to contact me, uh let's discuss but yeah this feature, but I just wg8 any previous approvers.

C

A

Could be a reviewer.

C

I can also take a look.

F

At Peter and Daniel.

F

I I can be the approver s.

F

With Peter and David your topic, next SEA advisorless CRI starts.

G

Yeah, so this one, uh you know where ideally we'd move it to Beta And, Define criteria for testing and performance benchmarks for both of the Cris uh to be collecting the metrics and stats. So it's um would largely be kind of refining the approach that we found when we met Alpha stage. Last yeah.

I

I think we need work on both the runtime side to merge the changes for all the metrics and testing I. Think yeah, that's Peter mission, so.

A

uh Do you plan to do it in other cities would be uh okay.

G

A

C

G

C

uh Sign me up for reviewing.com.

F

Yeah I think this is like con going work on the runtime side and then we see how much see how much progress we can make.

G

Yeah I think on the cubelet side. It'll mostly be you know, we kind of have passively talked about having about a validation tests for all of the T advisor metrics and then also having like perf comparisons. So it'll. uh There won't be as much work on the cubelet side but kind of just administrative stuff to make sure that we're guarding all our eyes and Crossing RTS.

A

I, wonder wonder where we are on uh in case of documentation, uh I think uh documentation around metrics in Kubla is very weak in general. So if we can handle yeah with better implementation during in this capital,.

G

Absolutely that's a great shout out that uh was also on my mind and forgot to mention it.

B

F

Great uh Nord look plugin support for pod, but they're, probably same status.

A

Yeah, it's rotten, we can ping the author, okay.

F

I'm wondering if we should go through all of the next ones, maybe we rotten yeah.

A

And uh what is interesting here.

F

So there was one I I know like David. You are interested in memory QRS. uh What are you able to work on it to take it to Beta I know. Paco also came up last week uh with the question around the threshold.

I

Yeah feel free to add me and uh and Dixie Down, the memory Qs I know it's here or not. If not, we can add it to the talk.

I

A

And if you love this uh details here, I can send a document in a chance.

I

A

I was going to say about memory swap I plan to work on that uh I was applying to working before, but now. Okay, so is anybody who wants to be uh Ura program.

C

F

Sure yeah I think if you have interest on the Reddit side as well, uh Ryan.

G

Is checking out.

C

G

C

Yeah yeah I can vote here too also happy to help Landing any of the last stuff.

C

A

How can we help you as a program.

F

So yeah I think it'll be good to make progress on the other. One is well the CRI image pulling, but I'm, not sure if we can poke the authors and see because they were Keen to do it in 126, but missed the deadline.

F

Which one sorry the CRI image full progress? Oh.

A

F

Another one I didn't see is a second by default, GA I'll check with Sasha and see what's what's his plan so.

A

um I want to go through uh thermobility again uh just to double check what we want to do there up armor, we discussed European team, our curse result. uh Is there any updates on that.

F

Ryan, were you able to take a look.

E

F

A

F

It's still open, yeah Ryan is gonna yeah. We can cover it next time.

A

Yeah, it wasn't touched for a long time, so maybe you can remove it. um I discussed this with uh sickos. They need help with documentation. Mostly um feature can be released as this so I. If anybody interested in doing this work, please speak up or reach out to me. Otherwise, I may help. I may have time to help us do that. It's.

C

Important I might have time to.

A

Continue. Thank you.

C

I assumed between the two of us we'll get it over the line perfect.

I

F

Yes, okay, I'm. Looking into this one I'll, add an update by the end of the week.

A

Is this we have in this cast um apology manager, uh yeah it's in release this one downward API, Pages yeah,.

F

So Sasha is gonna work on that one.

A

I want to put it in a.

M

A

F

I think we moved up there.

I

A

Yeah, um what they can do is to get at least and like make it cleaner and see uh how we do it in terms of uh capacity of reviewers and uploaders yeah.

A

So we have seven minutes left Let's uh try to get through the items relatively quickly, I think we covered this one. uh We're gonna have more comments here.

O

I made a mistake when I was filling out the stock and the other features I added a feature that I presented at the batch script. Can we go to that back to that Sig node, one I, I added it last week and I guess: I missed uh sorry, the other features there was an option about adding a condition for when pods are stuck in pending. Do I presented this at the batch working group and they suggested if I wanna work on a potential cap for this release.

O

I would want to at least add it to this list, but I, yes, I missed, adding it to the table. So I don't know.

F

Are you talking about the supplementary groups.

O

No, no sorry uh signode kept planning. There was a other features option that I think I added last week and then this doc completely changed uh scroll down. Sorry.

A

O

Yeah so uh yeah this was uh I presented this at the batch working group, where it's kind of related to the retryable cap. uh This is my first proposal for kubernetes, so I was mostly hoping to start a a kept process in 1.27 I, don't I doubt it will be fully done, but I wanted to know if that would be possible.

O

So mostly this is like for batch users. You run the problems where people mess up an image name or you have invalid config map or Secrets or volumes that aren't getting mounted correctly and your pod stays and pending and I've noticed that, at least in my my team, we have a lot of code around regular Expressions around matching against events, which is not very uh stable, from release to release and uh I'm yeah I can I understand. I can add this like we can do this separately, but I not really sure the process yet.

A

So the idea to improve status, reporting for uh painting pots, yeah.

O

A condition that kind of reflects pods that are uh stuck due to configuration errors, and some of these are might involved with like this config map volume missing I noticed this is a an interesting one. Where volumes don't give information about the condition or or Reason in the container.

O

They just have an event that says whether or not it failed to this one says: Mount volume setup failed, and this is kind of a case where we match on the event regular expression, rather than looking at a container reason, and some of these other ones are related to like invalid image. Name uh image like air image. Never pull like those are. Those are errors that happen, and then the pi gets stuck in pending.

O

I can discuss later, if I know we're at over time or at time, but I was hoping to at least start a discussion about a design for this one at 27.

N

Hey one quick note on that: we are tracking I, think one Capper on like what conditions around whether a bot is getting successfully started and when that's happening uh so I think that was like the last row in the table. uh You can maybe take a look at that, and maybe we can also discuss if there's something that's common that may be solved uh based on what you're, observing and your use use cases is.

O

That the Pod has Network yeah.

N

Exactly right, I mean the name has been discussed quite a bit and we came up with a new name. But uh yeah I was wondering if you already took a look and if.

O

I I looked at it this morning. I will I will look at it again. Cool thanks.

N

O

I will see uh yeah that might be. uh I have yeah, okay,.

A

But I don't see your cap on the list. If you can edit.

N

It yes, sorry I added in the last rule, but maybe I think the a 1.26 cut happened above that so I'll move it up. I guess I.

A

N

Do it? Okay, thanks yeah.

C

N

A

Yeah and uh um sorry I forgot, who was speaking.

O

I'm, sorry, it's Kevin Hannon.

A

Yeah, if you can take a look at uh this cap and see if those can be uh combined together or somehow improved to do be great, we will save on some paperwork if not come to next meeting and we'll have more time.

O

Okay, that sounds good now, look at it.

A

Okay, um let's go through the topics really quickly. So what you do you want to discuss it now and move to next? One.

D

Yeah, so the first one is already addressed, like we've decided to move ahead with the gear graduation. The second item that I have here is lack of multi-normal systems, which could be a key blocker for this work. uh The feature is beta, but actually the end-to-end tests are not executed on the CI because uh they don't have multi-numa and uh and Sergey. We discussed this in the signal CI meeting last week, I I ended up creating an issue in the test infrastructure, repo that we have If people could take a look at that.

D

That would be really helpful, and essentially, the goal is that we have some machines specifically for features like topology manager, memory manager, CPU manager, that can exercise some of these multi-numa capabilities, and if people have thoughts on this, like I would really like to hear that.

L

Have you spoken with equinix metal yet because they have some clusters designated for cncf projects? So if we can get on those, that might be a good idea.

D

Yeah I think as part of a test infra itself um like we will be having discussions on possible machines. So we we I, looked into um AWS uh Sergey, helped out and provided some machines that from gcp side that could be suitable for this, but again like it depends on what is the cheapest option.

D

If Equinox provides us the cheapest machine, I think we can go with that I'm just looking for input in terms of if a signaled um do we want to kind of move this forward, and you know I, don't want to say, push put pressure, but maybe influence the test. Infra group to make such nodes available in the test infrastructure.

A

Yeah I can help. With that um we can I mean we can always play with a frequency of test execution and run some weekly. For instance, uh it's not ideal, but uh if it's, uh if price concern is too big, we it can help.

D

So yeah, if if people have comments on that, please leave your input on the issue itself and that would be really helpful.

A

Next topic I wanted to discuss CRI API policies we can move to next. One next meeting we make has an update on in place of vertical scaling. uh We may, if you want to call I, hope you don't mind. Moving into the next meeting, um that I can don't, is not here, I updated the teams uh for based on owner files. If you can approve this, it will be great. uh It's part of year, end cleanup of all the files and yeah uh Marlo.

A

If you want to say more about the meeting on Thursday or what you've been doing, um you have maybe one minute.

L

uh Yeah, basically, we had the initial discussion. um There are some concerns um that we need to address, but the model is somewhat sounds we'll have to address them. So basically we need some way to bootstrap at the very beginning um before the resource managers come online.

L

uh If we're using something external I, don't know what else to handle is somewhere, um but please do meet with us if you, if you have time and concerns because we're trying to give submission as possible.

A

Okay and please pause the meeting time again on signaled slack channel. So everybody know when.

A

Great okay, with that I uh we out of time. Thank you, everybody for staying two minutes late um and uh have a great uh holidays. uh I. Think uh it's a last meeting this year um meet you all in January, bye-bye.

C

I