Kubernetes SIG Node, 23 May 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Node Sidecar WG 2023-05-23

Description

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

GMT20230523-160221_Recording_640x360.mp4

A

This meeting is being recorded.

B

It's uh Tuesday May 23 on 2023, welcome everybody.

B

Okay, so uh the only thing in agenda is Matthias uh you're asking about dog update right.

A

B

um Do you mean website.

A

uh Yeah mostly website, yes,.

B

I think we will need to have like, for example, how to use it yeah. We didn't change anything else to basically describing how sidecar works.

A

Okay, um but maybe also the fact that we now specified the restart policy for the init containers.

B

um I'm not sure if there is a place for that, but yeah. If there is a place that describe startup polish, you can also the main place where it says that you need containers cannot have probes. So maybe this needs to be adjusted.

A

Okay, you're taking notes or are you oh.

A

So I can, if you want I, can have a look uh and and check uh where it makes sense and I could like open up here.

B

We have peer reviewed right now. I love some comments.

A

Yeah but gonzu already answered. Most of them is yeah.

B

A

I followed correctly, yes, I think there are no other comments that I have to fix.

A

If I'm not wrong,.

B

Okay, I will speak for now again and we have a signal meeting later.

B

Okay, um I think it's finish line for this PR uh we'll have follow-up. It should be fine and uh John from sister said that all the test passed right now on istio, so they, like all the test, suits that they had with minimal changes. He was able to run it. He said that uh yeah one thing he found is uh their uh web hook. Injection um injected restart policy on containers and individual continues before it was ignored by everything but uh yeah. It was interesting, fun fine, so he fixed it. I think yeah.

A

B

Did um yeah so what's happening and another thing that that is happening and I, don't know. I hope that um somebody will show up is yeah. I talked to John from Eastern. He said that they don't need it. So there is a feature request asking to uh decrease the back of restart policy timeout. So whenever um you continue crushed, you may want to specify that please restart it uh fast enough. So you don't apply this back off policy, so it wouldn't grow all the way to five minutes.

A

uh Only for Cycles you mean.

B

Yeah, maybe uh for sidecars like we didn't, want to include into sidecars, because it will make sidecars special and people will may want to make containers inside cars if they want this feature, so they we don't want to include any features that are very special for sidecars. That may be beneficial for other containers, so I was thinking about it as a separate feature for any container um and scenario wise. It's it's! Okay, okay,.

C

There's some discussion in Signal the other day about that yeah.

B

It's from stock from gaming team, they wanted to say they have very short, running games like game sessions like you, somebody playing on a phone and there is a game rank in the cloud and it's around for one minute or so, and then they restarted and want to apply like to use different to connect to different uh phones.

C

The other, a really good example of that is, if you're using Cable in Standalone mode on your like you're, not gonna, be getting I only care about Cuba for these like half dozen pods, so restart them as quickly as possible. Basically, foreign.

C

Yes stand alone, so, if you're, using like a if you're, just using cubelet solo, to keep some pods running like in a control plane, for example,.

A

B

Yeah, so this is this seems to be a very small feature uh if it will be accepted. It's like yeah, just API, parameter and then apply different timeout.

B

But I wanted to get more confirmations that is needed for sidecars and John from if you said that uh they don't really want istio to elvoy to get restarted over and over again. They don't see this as a scenario um and I don't see it either I just uh and he didn't remember any customers who was complaining about it. I think mostly because I mean it's so dead, it's so broken that it keeps restarting then like. Maybe entire report is not healthy.

B

Yeah well, but I I'm I'm, trying to dig into this. uh If you hear any scenarios for sidecars just let me know as well.

C

So for the dogs you're gonna try to come up with an example, I mean because like istio and them aren't really ready to use it. So what's a good example to put in the dogs.

C

I was trying to think of one that would be like good, for example, purposes, but semi-useful and I'll.

B

Stay open to the energy collector as a sidecar might be useful.

A

B

We can get logs from, we need containers and then we can demonstrate that uh if collector died, it will be restarted and rewarded.

B

Yeah I don't have any.

A

B

um Try to think right now.

A

B

A

um Okay, so so once this one is merged, then we can check about like adding the Readiness probe as well.

B

And yeah presence probes uh boom score adjustment and that's it yeah, two PRS, basically cool and uh towards PR about correlation of total size as well. um I I will remind you now, but I wanted to him to consider it on this PR. First yeah.

A

Yeah yeah yeah yeah yeah yeah, that's the most important one. Yes, thank you. Okay, bye thanks have a nice day thanks.