Kubernetes SIG Node, 28 Mar 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Node 20230328

Description

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

GMT20230328-170451_Recording_1920x1214.mp4

A

A

Hi, everyone welcome.

B

I haven't started it yet I still switch it for the button.

A

I I press record: okay, all right hi. Everyone welcome to March 28 2023 signal weekly meeting. uh We have a couple of items on the agenda today, so let's tackle the second item. First, so Todd raised this topic. No, not third uh I, don't know the first name Pedrosa. So he raised a topic in signode slack about the um kill behavior on c groups.

A

We do so in c groups, V2, there's a new knob that allows us to kill all the processes inside a container when, when there's an um scenario, so the proposal is whether that should be the default Behavior.

A

And there are some comments there, I think uh in general, people are in favor of killing the container as a whole and I think that makes sense because it's hard for something outside the container to know if a particular process is killed and it's just cleaner to restart the application.

A

uh Any thoughts comments but folks on art.

B

Yeah, in fact, for uh sidecar containers you adjust on the score for sidecar, so it will be equal to the main container and I think uh the same exact reason. Everybody prefer everything to be killed at once, not uh like individual, not even individual containers but the whole Port, but I, don't know whether you want to go with whole body.

B

I will probably want.

A

Yeah it it may make sense to do it at the power level or the container level.

B

And continue to definitely makes sense. Port level may make sense for um sidecars. That's why we do it with sidecars.

A

Yeah, maybe I think uh Best Way Forward is probably to open an issue and then debate the container versus pod level knob. Maybe we can start with container level and then figure out if odd level makes sense as well.

B

I think the question is: do we want cap or we want just an issue.

A

ah I'm, not yeah I'm, not sure I mean we can start with an issue and if it seems like, uh if folks feel like okay, this is a significant enough change. Then we can decide if we need a cap and then think about restoring the older Behavior through some novel I. Don't see a reason why? But we can start with an issue. I, don't think we need a kid.

B

Okay, um uh Jose Luis. Are you on a call, I think Jose.

B

No um okay I will reply on the slacks that we need to have an issue and uh Bitcoin.

A

Okay, yeah taught us a comment about wanting to retain the behaviors I think it's best to be start with an issue and then continue the conversation there.

A

All right uh we can move on to the next topic, uh which is the annual report uh so we're putting together the 2022 signaled annual report and Sergey has kindly created a starter DOC for us. uh So our ask is that the community take a look at the dock and add comments there, so we can collaboratively put together the report. So sorry, do you know when it's due? How much time do we have maybe.

B

Yesterday, I don't know um nobody did it yet, so it wouldn't be the first uh late for this report and we're mostly late because it was code freeze and test freeze which took a lot of energy. Do you want to just go through the questions uh all together, and maybe there are some opinions on the call.

A

All right, Sergey I, have the issue where my mic is getting paused when I share.

B

Okay, let me do it.

A

A

All right, yeah, uh thanks again, we can see your screen. uh So the first question is: what work did the Sikh do this year? That should be highlighted.

B

A

C

B

A

So maybe we can I mean sorry, maybe we can cover what's been happening on the CI sub project side, like the work that went into making the jobs better.

B

Yeah I think by the media. We achieved very good results and we even sent this CI report, but then it's degraded so like right now we're trying to uh get it back to on track. So we can highlight, but I don't think it's like true highlight: um okay,.

A

All right, uh we can talk about the progress we made on the permabatas like graduating or closing out things at the black ink.

B

B

Yeah we did a good job of like the topology manager position, for instance, uh long-standing memory manager, yeah.

D

B

um Okay, I also wanted to say um about. uh Maybe what we did is we said that continue like we removed CRI view one alpha, two API and it cut 1.6 or, like continuity, 1.5, I, think out of uh support or even a 1.6 I think.

E

Local film was also last year, if I remember correctly,.

A

The deprecation.

E

A

Yeah removal, yeah I, think so too. We can double check but yeah. That's that's good to highlight yeah.

B

And with that, we we kind of arises a bar of dependency versions, like even the series and kernel version discussion uh when we said that uh Cisco's uh settings will only be supported. Certain uh kernel version.

A

Maybe I mean we did a bunch of new work, so we can say that yeah we started new projects and then, when we go to the kept SEC section, we get the details of the new work that started. uh We can say c groups, V2 went GA because I think that's a big event coming up. B1 is gonna, get replicated in various dependent projects.

A

B

B

We started this um uh work group for what is it?

B

Let's go down um security pro? No, not security profiles. Is it security profiles. We started this new audible today, but I. Don't remember what some Hardware something yeah. My memory uh is not very good around it. I wasn't involved very much.

A

B

All right, Mike, thank you.

A

Yeah, we can add that, yes, definitely it's! It took so long that it's worth highlighting.

A

Yeah I think improvements to uh Black Performance, maybe I mean evented like we started work on that.

A

B

So you need to be on certain worship.

A

Like I think that that looks like a good list of folks remember anything else. Please.

D

I would mention uh their a dynamic resource allocation because most of the work was done in the last year: okay, yeah.

C

D

Quite a bit of work, yeah.

B

Yeah, it's a good! uh It's a good color, okay! Okay! What initiative are you working on that r b interact as caps Todd added this? um If I can end-to-end tests, which makes sense.

B

Is there anything else, I think we track pretty much everything we're working on.

A

Yeah, most of what we do within okay.

E

Structured Logan that was done through a cap. If you remember correctly,.

A

Yeah but I think it came through uh different uh Sig, so okay, yeah.

B

Infrarily hosting, uh do you mean registry Mike, I'm? Reading the chat.

C

The registry, real jobs, that sort of thing to other providers over time.

C

Mostly Race Free posting them.

C

B

Let's move to the next one, careful in 2022.

B

A

Okay, any event that, like oh needs to be added to the list.

B

Yeah I, don't know how this automation, Works, uh I I, feel the least it's not correct. So this is this was done. um This was started, I, don't know. Maybe we created to Beta in this one.

A

C

A

I, don't think we moved that one to Beta I mean the CRI one. This is the CRA starts.

A

So ensure secret pooled images also didn't.

C

A

Go anywhere so yeah.

C

We can order next it if you want.

B

Dynamic resource allocation.

D

It's Alpha in 126.

A

So the username spaces can be added to the right version here. The stateless Alpha.

A

126, or was it 125 plus probably 125, will need to double check. The second Alpha is going into 127.

B

Yeah this was this, wasn't done. I think this was planned, but was never accomplished.

A

This so no I think the the split didn't happen, the split work, but the Pod networking ready condition was added by uh deep. So there was a renaming that.

C

Happened but I think.

A

We made progress there yeah.

C

A

Okay, I I feel like we are missing a lot of them. Maybe we can just uh get it from the release, notes or other places.

C

B

B

Okay um project house: what areas and sub projects does your group need? The most help is any areas with two of your owners.

B

I think many sub-projects like npg feature discoveries. They all have uh very little like support from maintainers.

B

But MPG clearly suffers from okay. It may go faster.

A

Maybe we can tag for NFD. We can tag. I can tag some folks from Red, Hat and Nvidia that have been working on it. It may be possible that there isn't a lot happening in that area like it's just supporting support for a new PCI IDs for devices right so.

E

Nft is actually maintained on our side, so we have at least two developers almost full time and for a couple of contributors which I actively see in GitHub, so I think we are finding this one again.

B

uh What other projects we have Shira API I feel that it's good so she's, where the actors are complete, not API, again npg, as I said, uh one Tower is supporting it, uh but you may be interested from more topology.

A

I think that's that's really good, like like I know like Francesco, swathi and Kevin are looking after that and say: SPO is also doing well. They have regular releases I. Think Sasha is okay.

B

Okay, do we have any areas inside there um couplet.

D

The site comment it would be good to have uh with somewhere listed the areas of the couplet and reviewers that that would help me, for example, to triage and assign issues and PRS.

D

Okay, not directly related, but I, don't know if it exists. That just would be nice. If, if anybody can point it foreign.

C

Pretty good idea to split it into managers.

A

Google managers.

B

I also feel sometimes we have property viewers and a Proverbs perspective. We have lots of manual progress listed as the reviewers as well, but most reviews are made by reviewers, not the approvers. So maybe you can clean up all units in this way as well.

B

Any ideas here.

A

I think it may make sense to like add, add folks to owners file within sub directories, because they get familiar, like example, would be like uh evented plague right like uh Herschel. Another spend a lot of time uh going and working in that area, so maybe they could be added there. Just one example.

A

B

By the way, as part of this exercise, I started aging more um all your files that signode owns and yeah. If you like, there is a yaml file with all the owner files with your own. If you know more, please contribute, and for now we can start editing those files to review and uh there's more people. There.

B

Okay: let's go to the next one: um what mattress Community Health stats? Does your group care about and measure? So we try to measure active PRS on the top of every meeting when it's active time on CI group. We also looking at anti-strs and bugs mostly bugs, though.

B

I, don't think we measure any other metrics.

A

Yeah I think these seem like the most impactful ones. If folks have ideas, they can suggest much.

B

As you contribute to MD help, new contributors engage with your group yeah. Looking at the contributors uh contribution MD file, my first reaction to say no.

A

So we added that document with direct and Don right on how to contribute and climb up the contributors.

B

Yeah, this is the next one yeah. uh This is mostly about uh how you can get started with a group.

A

So maybe this needs some cleanup. Maybe we can open an issue and if someone wants to take a stab at cleaning it up.

A

Like, for example, how how to run local cluster and how to test cubelet changes, stuff like that, can go in.

B

There yeah I think we have information just need to link to this information. Yeah I also feel that the most new contributors have troubles identifying the code pieces like what folders or what like, uh we don't have clear I mean we have this different, like, for instance, like when I just joined like almost three years back I was like what is staging like. How is this like?

B

Is it uh belong here or doesn't belong here, and uh apparently you can go to staging uh root folder and there is a readme file that explains what it is, uh but it's not obvious that you need to go there and understand it.

B

Maybe code structure as well.

B

Next one, if your group has special training requirements for your approvers, uh something Beyond contributors guide is helping here. Yeah we have this contributor ladder. We published it last year, I think it's a good progress. We made and I know that some other six looking at this contributor's father right now and trying to come up with something similar.

B

The other group has contributors for multiple companies.

B

Yes, we do and I looked at Stats and we have how many 19 companies uh with one two plus contributor contributions. I feel this number is low because, like 2020 report has like 43 I think companies are contributing. So maybe it's something wrong with the statistics but yeah.

C

B

It's a correct number: if I trust this, then it's only 19.

B

Has anybody want to call one so uh deep dive into this dashboard and understand? What's going on.

B

Okay, if you're interested, please jump in.

B

Other ways as end user companies can contribute user. Currently, not if one of those races more full-time support. What will they work on and why so I put a couple items here, but I'm happy to hear more so Docker, shim removal and duplication is still a big topic. There are many customers and many third-party companies, which only supports docker uh some help here, maybe beneficial uh an upcoming C group. E1 duplication is also something that will start.

B

We need to start looking at uh I think it will be less impactful, but uh still some support needed.

A

Yeah I think in general, like if you get feedback on items like these it'll, be useful. Like.

E

I would wish I write, also uh like more reviews from end users like for new features. What has proposed or something similar.

D

A

What about this, like, maybe can we send out surveys, I've, been publish them to the signaled mailing list right like okay, if you are using 127 the? Are you aware that these are the new features that were added? Are you making use of them, or not, just maybe after a few months, maybe three months after a release or something.

B

Three months after release, I think it will not be a lot of customers but the most engaged ones for sure. Yeah.

A

Because I think uh the rate at which the releases go out on various clouds and vendors is different rates, so just after release, maybe too soon. But you can see.

B

B

Yeah um now we need to calculate those as numbers. I I took slot number, but we have more.

B

Yeah, we just need to uh Dodge the boot and calculate all those numbers.

B

B

I, don't think it's new in 2022, but I think we just introduced like yeah I'll put it there with it into continue. We may want to do working group sidecars.

B

And maybe this effort on Research, plugins and CPU, we can make a working group.

A

Yeah I think that's been there on and off forever, like I, don't know like uh Sasha. What do you think or it's more like you band together for a particular feature and then well, if.

E

If we, if we revive something we, we should revive under the same in the old name like Resource Management working group, yeah,.

C

E

But I remember like that: I can only we wanted to say what the goals are achieve. It.

A

Yeah I think, like we had another discussion where we said: okay, it maybe makes sense to have it be a part of the main group, so it doesn't.

B

A

Don't feel an urgent need, I feel we are working well in the current model like we got in dra right, which is a big one. Yeah.

E

I'm also thinking what in in the current Pace, we are okay in Romaine signaled, there.

C

Is no urgent need.

E

To revive it and without hook, meetings if needed, when they just happen.

B

Marlo, do you have any opinion.

E

What are your thoughts.

B

I can barely hear you sorry.

E

E

I'm not sure we need another meeting, because we have the two cup meetings we're already running now, yeah what.

B

Do you think my question is, do you want to formalize it as a working group, so it will be officially.

C

B

I I can tell the pool request. We can discuss some pull requests because we need it. Yeah.

E

B

Sorry I think we reached the end of annual report and I need to drop off or no so yeah.

A

uh Thanks Sergey, uh that's the end of the agenda as well. Do folks have any other topics they want to discuss today.

A

All right, uh thank you for joining then see you all next week, bye now.

C