Kubernetes SIG Node, 11 Jul 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Node Sidecar WG 2023-07-11

Description

Meeting notes and agenda: https://docs.google.com/document/d/1E1guvFJ5KBQIGcjCrQqFywU9_cBQHRtHvjuqcVbCXvU/edit#heading=h.m8xoiv5t6qma

GMT20230711-160509_Recording_1920x1020.mp4

A

Oh, it's a signal side, car working group meeting, uh it's uh July, 11 2023, welcome.

B

A

Two of us here um and we have big PR- emerged right.

B

A

Okay, so let's see yeah.

B

I I guess you asked the the approvers to do some review yep. Okay, thank you.

A

Yeah I was trying to work with people to understand, concerns and walk through design and implementation details.

A

Okay, um we had this issue uh Uber issue outlying, all that left. So we.

C

A

Merged, yes, and now we have few more uh probes for sidecar life cycle hooks right. Yes,.

A

And you're also have end-to-end tests.

B

Yes, I already submitted of PR, for both of it.

B

Maybe do I have to separate them, not necessarily okay,.

A

So this one is.

B

Yes, but I think we need need to review the the next one first, because it is the pre-factoring thing: okay,.

B

It only updates the naughty three things so that you can just review it easily.

A

Okay, so you have three right: why is listen to and testing this.

B

Is yes, two two tier.

A

Okay, this is life cycle Hooks and probe. Yes, one always. You just said it's merged right. Yes,.

B

Yes, yeah got it.

A

Okay, so this one is uh main one.

A

And not into e to E I can I can help with immersion that I I would assume it's just one file right, hey thought.

A

C

Prime day doesn't.

A

C

Yes, last medium ran over. Unfortunately,.

A

We discussing what's left so we have two PRS on enabling probes. We also have um this PR uh total lifestyle, like um compilation of total resources,.

C

Yeah, so that one is actually I need to rebase it now.

C

I can do that.

B

C

A

Okay, uh so this radio State calculation, it's automatic right, so we don't need to do anything.

B

I think it would be already covered.

A

We may need to have a test, was it.

B

A

B

Not sure, yes, maybe you have to have the two tests per day. Yes,.

A

Okay adjustment- you just need to to.

C

Be done is that so that's uh I'm score based off of the like the new Total pod requests.

A

It's um so make uh sidecar almost core, not more than container reports in a like a regular report: The Irregular containers, oh okay, so, right now our situation is whenever you have a sidecar which typically much smaller than regular container, uh it will be killed really fast because proportionally it's very small comparing to the node so like what happened like all the side, cars being killed uh very early and then slowly we can reference to others.

A

Changes right! Oh no! It's about process! It's not about sidecars, not about continents.

C

Yeah, so it's about yeah processes within a container yeah.

B

Is it? Is it about the eviction in Cuba left side.

C

um Sorry, my change or this change.

C

A

Maybe a month ago told send the pr and I think it's more storage already and we changed from C group. We do how we terminate processes. So if you have a own problems and all processes within the signal will be terminated, so all processes within the container will be terminate all together and you will not get up in situation when inside the container. You have one project still running and another processor is already terminated.

B

C

So the thought was that it was like a problem that handle well by anyone to have individual processes in your container kills, as opposed to the entire container being killed.

B

A

C

Then this one is about.

A

Kind of a cross-container, so we want to minimize chance of regular, continuous running without sidecars and that's why we want to adjust this old score.

C

Is that caught out in the kit as well? Okay I want to read that so I'm going to try things out what that was.

A

And uh we also have topology and CPU manager, update I, think it's easy to discover what needs to be updated. There is like a few methods that iterates through containers, the only problem- maybe any dependencies like um anyway I don't know like- should be a reasonably easy, but.

A

Like we have Crystal covered this one, we have a PR and then we have some tests. A little bit of adjustment, yeah I think we're in a good shape. uh You may may be able to finish all of them, I think um score and topology manager. If we wouldn't finish in this release, it's not a big deal um yeah. We really want all that, because without this uh it wouldn't be a fully functioning for istio, for instance,.

A

Okay, yeah I, don't have anything else for this question today.

A

A

Yeah um yeah I, remember like Matthias, was working on documentation, so I think maybe.

B

A

C

A

B

I thought it was in.

C

Drafts still looked at it.

A

On the website.

C

All right there, it is.

C

I'm pasting the link to the side door, doc update yeah, it's still pre-draft.

A

Let me also need.

C

I've been trying to answer a few questions about it and kubernetes user. On slack, there's been a few people asking about it, so I think it'll. There's some excitement for the ability to use it.

A

Okay, um let's try to see like what progress we will do without us. I need to do your tasks and uh then next week we may want to ask us if anything will be done.

A

So, thank you very much. Thank you. Everybody thank.

B

B