Kubernetes SIG Windows, 17 Nov 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Windows 20201117

Description

Kubernetes SIG Windows 20201117

A

Hello, everybody and welcome to another sick windows: meetup it's the 17th of november, and it's also cubicle north america week. So as always, it's a recorded meeting. So please ensure you adhere to the cncf code of contact uh poker yeah. I had a question for.

A

Claudio hey, you want to ask claudia your question or no, oh, he has all your problems. I cannot hit anyone.

A

uh Asking him to type his question: okay, let's get! uh Let's, let's kick up our agenda here. Let me open up, see windows.

A

So probably it's gonna be a short meeting today, since I don't see having a lot of folks here since who are uh short-handed with uh keep going and everybody kind of uh trying to attend as many presentations as possible, um so kind of a couple of things on our agenda here.

A

There's the process handle leak, one of the issues that was reported in the the community so mark. I know you've been kind of taking a look at that, and you mentioned that you might have some more time next week after thanksgiving uh to or the week after thanksgiving to take a look at that.

A

um Have we seen that before with uh like, if you folks are using plain containers or is it tied specifically to the cube blood and q proxy, as as evident by the the the you know, some of the restarts that we're seeing in some of these images? Where is it?

A

I like this one that the cube proxies had 3000 restarts? Do you know mark? I.

B

I don't know, but I and I'm guessing, that these clusters were kind of configured without um through some other mechanisms besides either aks engine, which is uh like an unmanned or a managed cluster offering or or an unmanaged cluster offering or aks on azure, and we haven't seen these sorts of issues when we use when we use either of those deployment methods on azure. So I need to keep looking um dig into this after I get a chance.

A

Okay, do you think that this is important for us? We have an opportunity to still fix it for uh 120.

B

When possibly, um this may be an os issue, and I know that this is running on windows server, 2019 and if, if it is an os issue, it's going to probably take until february and march to get a fix. Even if we identify it pretty early.

A

Also, it might not be a cube led or a q prox issue.

B

Yeah, I'm not entirely sure yet, but it could be um yeah. I think I think we just need to keep looking at it.

B

It might just be a config issue as well and we just need to better document how to because it looks like there's an issue. Where was this the issue where something was running out of address spaces.

A

Oh no, I thought it was just no. I don't remember saying anything about address process here.

B

Okay, yeah, that was something else, yeah yeah all right. Oh it does say so. There's an error message um in the in the in the initial comments. It says that there's an error message, because the hns endpoint can't be created because um ipam failed to allocate ipi.

A

B

Ip address is available.

A

This one here, I'm highlighting yeah.

B

Oh no, but that was is this: oh, this is just an easier way to reproduce. Sorry, I'm kind of paging some of this in um no worries yeah, I think we'll know more once we actually get a chance to dig into this.

C

I've been uh looking at the coop proxy for windows code base and I have a pr and decor for improving some of just the general explanations of how the code works, and I was thinking I wanted to add metrics, at least for the user space proxy, because I think we should be counting how many times we're starting go. Go threads up for when we're doing user space, proxy and stuff.

C

But before I did any work on that, I just wanted to ask: is anybody using the windows uh coupe user space proxy because I haven't seen anybody we're using it here at vmware? But I don't know if anyone else is using it, um and so I didn't even know if I was to do this, whether it would get caught in any kind of ci job or or if I could even find anyone to review it.

D

uh Jay, this is the q proxy. That's in kubernetes communities right.

C

Yeah in windows, user space, yeah.

D

We're we're using it in in our product.

C

Oh okay, so do you have the files to do a little pr into user space? Do you have someone that might be able to review it um and slash, I'm just wondering if you use it in your product, do you have nci like if I was to break something? Would there be an easy way to know that I broke it.

D

C

Would happen is.

D

We would we wouldn't catch it immediately, but if q proxy broke, we would eventually catch it. I mean I'm happy to do a bunch.

C

Of tests on my own against it, I just figured if somebody's running nci, I maybe yeah.

E

Jay, can you actually send it my way uh I'll find the proper people who keep proxy from the windows side uh to get to do the review yeah.

D

E

Should be able to also uh suggest some testing.

D

E

D

Mention jocelyn, I didn't realize joslyn was here, but yeah jocelyn is actually the right person.

E

Okay, yeah, I was hoping david- was around as well, because uh he usually takes these questions, but yeah do send it my way, I'll and I'll find a proper people.

C

Okay, cool also because there's a couple of tiny, logical changes that I made in there and I just wanted to make sure I wasn't totally breaking someone. Okay.

A

C

Thanks a lot awesome.

A

Sure all right very quickly, uh before I jump into tomorrow's uh tomorrow's uh point, so 1.20 releasing release things right, so uh the the deadlines are coming so test freeze is on november 23rd, so you know worst possible timing right there right the week of thanksgiving and then dogs right after you come back from thanksgiving, I'm assuming because our major change this time around is just a container d work plus some networking.

A

Are we okay here? uh How are you doing from a docs and and and is there anything that someone needs to jump in a hobby, they're approved or pushed through.

B

From someone from a lot of the container d-docs, I think that those are going pretty smoothly. There were some asks for a technical review from you and muzz. um That was last week. I was out last friday and busy yesterday, so I didn't get a chance to look at that, but it seems like pretty okay.

A

Must have you taken a look at them or not? Is there anything like if you, if I need to pop into something just tag me, I.

B

Don't know the call.

A

Oh, my god: okay, um okay, yeah I'll I'll, take you again um yeah. If.

B

A

Me with something, if you want to tell me something talk to me I'll I'll work, a couple of this thanksgiving week anyway, so.

B

Okay, yeah, the doc reviewers, I think, didn't, have any concerns. They just wanted to make sure that all the information was accurate. So.

A

If you wrote it.

B

I, if you wrote it, I trust you, but yes, let's take a look at them. Okay, I did have a couple open questions for muzz um and that is for gmsa support for windows, server 2019 with container d. um I'm still don't have clarity. If um so, there was an issue filed in the windows, microsoft, slash windows, containers, repository, saying that this was fixed um and the patches were going to be out in the 11c update, which I think is actually dropping today.

B

I am not sure if that only includes windows, server, 2004 support or if it includes support all the way back to windows, server, 2019.

A

Told me that we wouldn't be able to do it this time around. That's what he said not for 1.20, okay, okay, that's.

B

One month told me I.

A

Don't know I don't know if something has changed.

B

We'll just have to call that out in the docs, then.

A

Yeah, oh yeah, let me know we haven't deprecated the docker shim, so that's, basically still people can still use docker team and then they'll get this next time with the next cycle. I.

B

Think that the error or the warning message in the cubelet on startup did land in 120 um to say that this will be to yes, the clock for docker. That's right! Yeah! That's right!.

A

All right, uh tomatoes: let's talk about the windows device, plugin api- I don't know if you can talk now.

F

Yes, I hope so yeah and you can hear okay.

A

Cool that's great yeah. We can hear you excellent.

F

Perfect yeah, so in fort summary, the end-to-end test is running. There's an open issue regarding what gpu is available on this particular test node, because it seems to be at least it's not reported as an nvidia gpu.

F

But I think this is not on blocker right now, because everything else is working. This device plugin is deployed in the test. A test container is created, it says container is requesting the gpu resource, the diet x-ray source and is getting it, which is verified in the end-to-end test. So.

A

Sorry is that the test that's running on aks.

B

Yeah, that's the test. That's the the pull request. Job that's set up with an aks cluster with uh gpu, capable.

B

A

Very cool so mark is there anything else we need to do.

B

Push us through. um I had a couple of questions which I actually want to take back to some folks on in the windows container platform team, because after talking um through some discussions in there, it seems like- and we need to confirm this.

B

The behavior and the gpu device that shows up in the container is different, depending on if you're running, the full server sku or one of the the headless server cues and one of the things that was a little that I just wanted to get a better answer for um around is that normally the full service keys are only released with the ltsc windows releases.

B

So if there's a difference in behavior between how the gpu device is showing up in the full server skis, which only come out, you know every four years versus the semi-annual releases. um I just want to understand what that is.

A

Yeah makes sense: uh is there a time frame for for doing this, like I know, or kind of right up against the deadlines here.

B

um I was, I was going to look into that this week, um but if this is working then, and- and we have the test to show- um I think probably- we just need to clean up the pr a little bit and.

F

As I would suggest, this should be not on blocker for zpi itself, also, because um the the pull request itself is not about gpu access, it's really only about as a generic device plug-in api, and this is already verified by the end-to-end test. Of course, it would be nice if the intern test really actually verifies gpu access, but this is more than we have to do. You know.

A

F

A

So does does the pr not pass all tests, let me actually see if I can pull it up here really quickly, um and I took a note of that not too long ago. I just can't find it now.

F

Running now and successfully running.

A

No, I understand that but like if we look at the pr is everything green.

B

Is it ready to merge.

A

F

A

Sorry this is the end-to-end test. This is the well we should. Why is the test? We should merge the pr for the into end to s test as well right. Do we.

B

Ask reception to get that to try and get that merged.

A

Let's, let's navigate that to the release team, but I want to make sure: oh this one was merged.

F

Yeah, but this is an old one which broke um bro.

A

That's that's the one that was reverted okay,.

F

Yes, exactly okay,.

A

uh So where's the new one.

B

A

Nine three: two: eight yeah, okay, all right! So does this one puzzle test.

A

A

Okay, job succeeded.

A

F

It obviously does not include the end-to-end test, which is separate, pull requests, but I think your lovely efforts.

B

But we have those and then tests in a separate request.

A

All right, so it has a milestone.

A

B

A

Need we need the approvals now hey uh mark? Do you mind pinging, that person from signal that uh you're telling me that day? That is a little bit more on top of things and see if we can sure that was.

B

Myrnal um he's you know doing some of the cry changes, so I'm not sure if this would be in his oh.

A

That might not be on his wall house. Okay, let me let me actually see whoever basically approved the original one. We should have them. Take a look at this as well, so the original one was.

A

This one here and.

A

Like it was the one, was they get the one that approved it? Oh, come on.

A

Right there, oh, this person do we know kevin. Is he part of sicknote not sure.

A

Okay, I'll I'll I'll delete it after we're done with this here and I'll I'll tag. A few folks, so mark will be great. If you also can improve as well. Can I should I say, approve, or do you want to take a look at one more time mark.

B

I'll take I'll take a look but I'll look at that. This.

A

After like right after this okay cool, uh so once once, you approve, once you put uh approved from sick windows,.

B

Okay and then adelina, I know you were commenting quite a bit. um Do you want to take one another? Look at this? If you get a chance as well.

C

Yeah sure I was on mute I'll uh I'll, take a look.

A

And and tomorrow's you need to make sure that the end-to-end test also makes it through. So if you can help drive that and make sure, since it tests are passing, it needs to be checked in.

A

Cool all right, um any other questions or concerns for today and anything so we talked about docs. We talked about the release uh anything else. Anybody wants to bring up.

A

All right, everybody gets back a few minutes of the day. Please uh if you attend cubicon, we have our presentations on friday. uh You know come and join and see. Questions are coming out there for from users and customers, and uh if you find an interesting session, you learn something write some notes and and send it to everybody. So everybody can benefit, especially if something that we need to pay attention because something new is coming up and we haven't had eyeballs from sick windows, so those are also important.

A

I always enjoy cubicon and I will cancel next week. I believe a lot of folks in the us will be on vacation uh mark. I think you're on vacation, too yep yeah I'll, be mostly.

D

A

So uh so no meeting next week and then we will uh uh we'll meet again right after thanksgiving sounds good. All right, bye,.

A