Kubernetes SIG Windows, 16 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Windows 20190416

Description

Kubernetes SIG Windows 20190416

A

I got that started so today is the 16th of April, and this is the sig windows meeting, and so just a quick reminder, since this is recorded and will be posted on YouTube as well as all of our other me just like all have all our other meetings are so, fortunately, Michael had a conflict today and he can't make it so. The first thing that was on the agenda and notes I had to push out to next week, and that was the survey results around the use cases.

A

So we'll have to pick that back up later.

A

The main things that I had had on the list were making sure that we are getting back on track with the enhancements that need to be approved by the end of this month and then getting the project board updated.

A

It's so, in addition to those- and anybody have any other stuff that they wanted to talk about.

A

Okay, so let me share this out real quick.

A

Hold on here, they change something zoom where I can't just share an individual app anymore. I have to draw this little rectangle around it. Okay, can you see that okay, yep, okay cool all right.

A

It's not the one going to show you: okay, here's the right tab.

A

So, just like we did last time, we've got a backlog, column set up for a version 1.15 and then the things that are live in progress or review.

A

You know we've got tracked over here, so there are a number of things that were carried over from version 14 that still need to be reviewed and merged. So if you have anything in that list, please make sure to take a look at it and make and make sure there's nothing blocked on it and then I'll be asking Craig and others to start pinging people on those they're missing reviews this week.

A

I think that one's already done actually just get that one out of the way, and so, if there's things that you are planning on working on, make sure that they're assigned to you and that they're in the backlog or, if you're, already working on them, make sure that they're here in the in progress column and it's basically the same same thing we were doing before and.

A

So I'll get it's a good mine, updated in there later in terms of project planning. The release team has got a approved schedule published. We see if I can just find. It, though, can I have the link in here.

A

A

Okay, that's ahsoka force, enhancements phrases, xxx.

A

Okay, so that's the release schedule, but the next thing that we need to make sure is: you know, two weeks from a day, all the caps have to be marked as implementable.

A

So last week could they talked a bit about the container D stuff I've got the document still linked down there below and I got an initial review from from you, jus and deep thanks for thanks for your feedback on that and I made updates last night and I'm. Taking that to sig note next hour we haven't been out, we haven't talked about the cube ADM stuff. Yet is there anything that that you're ready to talk about?

A

Is there anything that so, if Talia and Lumiere um did you have anything you wanted to cover on that yeah.

B

I think I can just briefly give an idea of what we're planning to do so right now like there are a few. Are there several differences between how a nose runs like a windows node and a linux? No drugs and kubernetes so like one example, is cube. Proxy is not run as a daemon set in Windows. It currently needs to be run as a service and another current limitation.

B

There is that right now queue proxy needs to be run after flannel, otherwise it will fail like it'll crash, so there needs to be work there, where I'm going to basically introduce some like waiting mechanisms, so that it, the service, can still run, and it won't crash and then separately, when you deploy when old, then it'll actually start to work as intended.

A

B

Then the goal with this project is we're going to basically have to wrap, cube ATM in something else that something else is still to be determined. Hopefully we get that, like all of those details figured out this week and then we can start writing next week, so we'll make that deadline, but so pulling the binaries is one thing: that's going to be handled in the wrapper for it and cube. Atm itself should be responsible, I think for starting and registering cubelet and cube proxy as services mm-hmm.

A

B

So one issue that I see here is for Windows different values for the config files. I, don't know if that's necessarily true for Hewlett, but definitely for Q proxy. You pass in different values. Basically network type, we've Amir seemed a little hesitant to expose something that would, and you can bring him directly. That would allow the user to specify the network type that they want, but I think with this wrapper approach, then we can have that we can have that parameter exposed at that level. So that should be okay.

B

Bootstrapping is going to be handled by cube, ATM, basically we're going to handle, join and reset, and then afterwards, so the wrapper will be responsible for pulling the binaries in the beginning and then afterwards the wrapper will also be responsible after the node is joined and after the services are started, it's going to be responsible for starting up flannel.

A

Okay and the wrapper is something that you had just run once this isn't like another service that would be services right, yeah,.

B

No, the wrapper should only run once you may need to I mean right now. The CNI is like flannel and Linux is deployed way outside of this, so the like, the last part, may be neat like that will need to run every time you bring the note up. Yeah.

A

B

We're going to try to separate the so it's not like how it is currently where you have this full set of scripts, that you're deploying every time that kind of like skip a lot of things. The beginning part will only need to be run once and by that I mean like pulling the binaries and everything yeah.

A

I see, and so if someone wanted.

C

B

I think for that I would need well I. Don't think you proxy needs to be read. Well, I'm, not familiar with that lube Amir is gonna, be handling that part since he's more familiar with it from because he's from the cube, ATM team or like he worked on that project, so we would have to I think I key and I still have to work out the details there, but I'm gonna defer to him on that.

D

Question was on Windows. How would you register is it through an SSL or some other mechanism? Oh.

B

You mean registering the services. I thought you meant um like registering the node, um yeah I think we're gonna use se instead of Venice's them, but I think an SSN has better logging, but there's also work item in our backlog. To identify better logging options. Are there support for both? So a decision needs to be made so I'm willing to listen to what people's perspectives are on that.

C

I'd prefer a bidding logging I know.

C

B

There's still discussion on that. There was a cart in the board before people. There are a couple different options and people have been discussing there. So I think that's still work in progress. Yeah.

A

And I noticed that logger us has been updated and they're using a container D to write Windows etw events, which are which are nice because you can. You can capture them at to say within the same session as other things like the windows, h, CS and h, NS events, yeah.

B

Yeah that changes small I think and had contributed to that discussion too, meaning one way or the other I forget, which way, though yeah.

A

Well, it might need a port, a similar thing over to K log yeah.

B

B

Think that's pretty much it for the rough summary of that. If anyone has I think the dock is available for people to be so. If anyone wants to comment on that, then we'll still be reviewing it. This week.

A

Okay and okay, I just did a dock from last week. Okay,.

A

So is the goal to get that into a cap like what's in the next next week, yeah.

B

Yes and I think Craig is gonna, lend some help there too and I. Think like you me and Luba Mira should meet okay finalize some of the things, so you can just start like I. Think at this point a lot of it is just making decisions so.

A

A

A

Alright, the other other area that I had on the list was GM si and run as user name updates. Did you have anything you want to add there deep? Yes,.

E

So, basically, for the GMs APR that updates the API adds the window security context. That's currently in Jordan liggett's in progress column in the APA review board, so it moved from assigned to in progress. Last week we are hoping to hear something from him pretty soon.

A

E

A

For a storage standpoint is that something that we needed to carry as a cap, or is that just something that we're doing? What's that yeah.

E

Good point so, basically in last week's thing storage they mentioned, you know for the storage proxy that I've been looking at. Although it's out of tree work, they say: it'll be safe to have a cap, so I will just write one up by end of this week and publish.

E

I'll add six storage here.

A

Okay, well, let's see if we got storage computer, do we need any networking cap system this round.

A

D

The networking work is comprised on the container dating okay. You, like all that the endpoints removal, the multiple say my calls all of the work will be done as part of that which is downloading.

A

Are you expecting there to be any major of changes needed like I know, we still have some workarounds in place where we're calling C and I was at once per container as opposed to once per pod? Yes,.

D

So that's the work. That's that's already done as part of infinity.

A

Okay, okay, that codes in the darker shrim code path today, and so it's just not gonna, be called a few years in containers right.

D

A

D

That is over. That's what we have done. That's people like enhance that in that kept. If you want.

A

Okay, got it: okay,.

A

I think that's everything I had so far. Did anybody have any other any other items to add or or Q&A so there's.

D

One word we are currently looking at this: bringing the invoice support so I will probably but I don't think we will realize that support and 115, so we'll probably in the future, but we've work. There is a separate slack and separate things channel for that, so we will work on like we are trying to bring in the proxy support in in kernel mode mm-hmm.

A

D

At least the TCP first.

D

So it won't be realized in 115.

A

Is that something I would require a new Windows version as well like a 1981 or 1982? Yes,.

F

A

And what slack channel is that being discussed in an LOI there's a separate slack or is that within the kubernetes slack? It's a separate slack? Okay and I.

D

Think there's a pink channel also.

C

This is like just envoy working on Windows in general. This isn't like using it to do anything for like it's, not like you're, using it for kube proxy or something right. Yes, completely. Okay, it's.

A

C

Wouldn't need to be on like the one 15 or even 16 roadmap, that is.

D

C

C

A

Okay, I was just going through the list here on anything else, say: I haven't been checking the chat. Let me check that I.

A

Hope Peter link I was looking for thanks.

A

Okay, well nobody's got other top I guess we can close it up a few minutes early and get a coffee before sig node. Oh.

A

Yeah yeah yeah.