Kubernetes SIG Node, 7 Feb 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Node Sidecar WG 2023-02-07

Description

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

GMT20230207-170414_Recording_1344x1120

A

Hello, hello, uh it's um February, 7th 2023, it's a sidecar working group. A meeting welcome everybody yeah today, I think will be a short meeting. um We about to enter the like enhancement, freeze and uh we are on a Finish Line I wanted to just check. What's going on with uh issues and um oh, what active work is happening so, first of all on issue side um on bureaucracy, part of it. Oh yeah, there is a change in templates that needs to be picked up like between I, created the template and like um implemented.

A

There is a change that needs to be picked up and I will do that uh yeah, some Fields um I think biggest problem right now, like is. There are two big things today.

A

um So first is uh we don't have an approver uh because Derek I was told that Derek is off this week, so I was being in that a lot of time and he wasn't around so I will um ask Bruno now and maybe I don't want to help his approving the cap, because that was generally okay but- and he didn't uh uh didn't say anything uh negative so far, but he also didn't um give you an approve.

A

um I, don't know what uh how to react on that, but I mean I, think I'm, Ronaldo approval and uh don't uh uh just uh put in a standpoint should be enough.

A

Okay um and then second one is uh in this cap.

A

I still asked to write down specific tests that we will write like packages that will be writing test for.

A

A

Yeah there are a few uh comments too on unaddressed on this.

A

I need to. We need to address it um if somebody has energy to help with that, it will be great.

A

Maybe if you have time to reply in these comments and say, like maybe change the text, um if you have time, um I will be looking at comments and like once I address a specific comment. I will resolve it. So uh this way we can synchronize and coordinate.

A

Okay, well I hope. I will have time today to clean up completely and yeah.

B

What's the deadline, it's Friday.

A

You want to do it before, because uh this uh things, you think, is direct. That Derek is not around, so I need to make sure that we will have somebody to approve it. Okay,.

A

um And let's talk about implementation a little bit, it's also full of uh roadblocks uh and uh findings. So first I thought: if you can update, what's going on with this uh application.

C

So it's still just waiting for a reviewer um like it has been uh triage is accepted, but it's waiting.

B

C

And I guess I: can it.

B

C

Signal stuff and then uh some sexy, casual and stuff, so I'm not sure if that makes it a little more complicated or.

A

C

A

Does but let's try to find somebody so on what about this? Second part when we expose it as API I, think we agreed on that part. Do you think you can start working on it as.

C

Well, yeah yeah I believe so um yeah I think there were some last uh from you know. Some uh requests for a change in the name but yeah kind of yeah I can actually start the hot status as well.

A

Yeah I'm not sure about the, like I put the latest uh names that you already done in a gap. um My only question is: do we need to put uh name calculated uh in there like suffice, calculated that I will indicated? uh This is not a new information, it's actually the same information you can obtain yourself, but I'm, not sure how valuable it will be, maybe to limit ourselves. So maybe just don't do it.

C

Yeah yeah, it's always easy to change later on as well.

C

A

Thank you. um So this issue, um yeah I, found a new issue about started field. So um we have, we want to rely on a field started um to continue. Status, has an indication whether startup broke completed successfully uh and those indications in field started and I. Think it's a very useful field for us to understand. With the sidecar completed its installation and next in container can be started.

A

The problem with this field is, it is not initialized. We need containers today, so we free to Define semantic for it, and I wanted to uh I mean it's easier to divide semantic when the field is like I mean today, is a pointer and I really wanted to make it a non-pointer. uh This way uh it's either false or true, so you don't have this Middle Ground like today, you need containers, has a started filters now uh indicating that um we don't like we don't care about started like I.

A

Don't know like how to express it, but uh um I think it's it's not ideal, so I think uh started. It needs to be either true or false.

A

It shouldn't be this like undefined state and um then when we will start implementing sidecars I, wonder if we should just switch all the unique containers to start that field or just uh um and then make a semantic of started for a neat containers be like container completed, or we need to just follow whatever semantic we have today and uh for sidecar, specifically only look at started field, foreign.

B

I would be in favor of doing like for the normal containers and if, if you look at the the answer from Team.

B

Yeah so he's against changing it, because it will break the client go again. But if we take the option two, it's uh almost as if we we set it to a non-pointer.

A

Okay, so you suggest we just keep it falls and for when do we change it? We need containers, we don't have a proper manager any longer, so we need to have a different state at which we will change it for any containers. If you want to change it to true for regular unit containers.

B

Okay, so I don't know.

A

Exactly yeah I also don't know. um There is a point in time where we will set started at uh field uh started at indicate that the process has started, um but maybe the same time we can set the start that we need container specifically because init containers doesn't have any post start hook or startup props. So it's pretty same time, um but yeah I, don't think.

B

Today, started at is is on the board. Level is not on the container level.

A

There is a container status, dot status, dot started.

B

A

Yeah, because we want to make sure that customers can terribly shoot individual containers right, yeah, I, think that might be unpotted as well, but I'm pretty sure there is one in container.

A

um Okay, so I don't think we will need to change semantic of started for you need containers to be fully completed like so. We can have unified Logic for unit containers like one started, stools and press it to the next one. So I think it's uh it's too big of a change in unnecessary change. I think, okay, yeah. If there is an agreement, so I I don't have any strong opinion like because we need to change it or not I. Just uh it's.

A

It feels a little bit strange that uh some fields are easy to change and, like we break into client, go quite often, but then some fields are like yeah I, don't change it like it's uh uh I, don't know okay. Lastly, we.

B

A

B

A

As well like um I, think more and more we talking to people like uh one more. We realized that this change will break a lot of use.

A

um Yeah I think Joe found some like. So there is a whole ecosystem of libraries that have depends on client goal, and these libraries are not like. One level depends: it's like multiple levels of dependencies, so library takes. It depends on Library like open Telemetry, for instance, that takes dependency on client go. So there is like three levels of dependencies. An updating client go in this uh situation becomes um becomes a quite a trick, especially when there is no backward compatibility.

A

So I think suggestion was that either we initiate a project of creating an article relative library for that will guarantee backup compatibility or we don't touch client go for now and again, uh it is what it is. um We just need to deal with that.

A

Okay, so I think that's uh all we currently doing for any containers is implementation tasks.

A

Is there any other comments, suggestions.

A

Okay, I think once we pass the stage, we really need to start thinking about uh test like um there is an In-Place Port update, um PR. That is uh big and every time something changed it like it's 30 base after the base, like uh people spend so much time on that. So I think we need to try to decompose our cap in small tasks as small as possible, even though it's like maybe some like preparation causes like then you just insert another parameter and you go down like you, sell limitation in your containers as well.

A

This kind of decomposition would be very useful, um so I think, maybe next time we can get into once we have cap approved and uh this out of mind, we can start decomposing uh into specific tasks.

A

um Okay, is there any other topics for today?

A

If not, I would really appreciate if you have helped me with the pr review and like reply in the comments and I will be also updating text as well. Thank you. Bye.

B

Goodbye I will have a look tomorrow morning. Thank you.