Kubernetes SIG Windows, 3 Sep 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Windows 20190903

Description

Kubernetes SIG Windows 20190903

A

I, hello, everybody and welcome to knowledge windows meetup, it's the third of September! Thank you all for attending, as always this is recorded. So please adhere to the code of contact of our community all right, so we're getting close to the release. We are, you know well, thanks for all the hard work that everybody did to kind of get us here and all the been a few Pink's here and there we are holding tint on the dogs for CSI alpha and all that deep is working on that. So here update the PR a very shortly.

A

So thank you tip on that and then I didn't seem to knows that was still pending, but Patrick. Anything on your end.

B

Sorry on CSI or.

A

Anything else like from that this basic dependent for the release right now.

B

Yes, I think all the things that I was looking at our her listed later on in the agenda, because we still have some enhancements repo stuff open for GMS, a cube, a DM and CSI yep, and then I think the rest of this stuff can can move out. We just need to clarify that for the release notes.

B

There is a test so last week, I noticed that the sig windows tests weren't on the release informing dashboard, so I got that fixed and then Adaline and made another update to that. But we've got a one test: flake that I, don't I, don't mean uh in Claudia or currently looking at so the release team is probably going to be picking us with a tracking issue for that and if we're not able to fix it in time, we may have to exclude that test.

B

Do you want to describe the problem Claudia or deal me to cover it.

B

Okay, I guess I'll cover real quick, so basically that the HPA tests trigger a request in the diagnostic that's supposed to basically just spin the CPU up to a certain percentage, and then these the autoscaler tests have a certain threshold set to try to scale enough pods to get that down to a certain range. The problem is that the aps are used to basically measure how much CPU is being used are not not updated frequently enough.

B

So when you generate the load, it overshoots, and so if the target was you know, was 25% we hit like 35%, and so it causes autoscaler to work and create even more pods than were expected, and so the tests fail, because we scaled too far, and so Claudia was looking at ways to see if he could basically improve the accuracy on that a little bit we're putting a workaround that you would sort of adjust the target, like maybe subtract, a few percent or something so that it hits the right to place so he's looking at that.

B

But the release team will probably be pinging us on that. If it's not fixed and then so that's that's the only flake that we're tracking right now on test grid and then today's the deadline for docs, PRS and so I linked the ones that were there for cube ADM. But basically, if there isn't Doc's merged for a kit for each of the features that we're talking about, then we need to make sure that those are in and ready for with you today.

B

So the first one list was CSI. Did you have any updates on that deep.

C

Yes, I'll just post a note, there saying my my main thought there was to update this section around the storage for Windows, who say that CSI support is upcoming and my plan there was to say about you know: the increase support is in and external components will be updated with markers as soon as we have a few of those, but no driver register on the CSI proxy, but that'll be a little bit forward. Looking at the moment for 116.

B

So I think the statement that we want to make is that is that CSI is porta and alpha state on windows, but they don't want to make forward-looking statements.

B

So I think we would need to say something like you need to check with the individual out of tree providers to know when they're supported.

C

A

If there's one or two that we know that work, we should document that and say hey. You know the community has validated X, Y or Z right. So if there's something that you know that works beyond sending them it's very hard. If everybody has to go through every providers report to figure out if it works on Windows, most of them probably might not even say anything about Windows. Some of them didn't even know so it be worthwhile to just say: hey we validated that this CSI works and you know for additional details.

A

Thank you. Okay,.

B

Great so next was cube ad on the placeholder PRS. Are there the first one? The list had had a review comment on it.

B

Let's see, I can see who's on a call because I don't see Kalia or David or Mike I'm.

D

B

You're there, okay, sorry, so, do you think you'll be able to get to dock PR is updated today, yeah.

B

D

The only comment, I think I'm, seeing as dating some of the names to use lower casing like updating the casing, should be very quick and then the other one is mentioning the excellent versus overlay or sorry mentioning the requirements to use overlay.

D

B

B

Okay, yeah just so long as those are on track, then don't we should be okay, thanks for thanks for doing that.

B

And then and then we've got GMS a up next, so I see. Adelina was reviewing one of the last PRS in the testing repo and then I guess it was Jeremy. I was doing some updates to it today. Yeah how's.

E

That tracking I should have it I'm actually running my final test on it right now. If everything goes the way it should then I'm gonna squash all the commits, and then let Adelina take a look at it again. It should be ready to go okay.

B

B

And do we have a dark pier that needs to be merged for that.

A

Those changes, we're gonna dock PR already right. Yes,.

C

The dock PR has already been submitted earlier. This was more like stabilizing the test and yeah cleanup for beta. Those were the two main things: the API changed, the dots for the API teams were already merged as part of the okay.

B

Yes, thanks thanks for that and I see someone added something about DNS search, suffix list.

B

But it was uh yeah.

F

That's me um so this is me yes,.

B

F

We see it sound from Google and kubernetes networking, so I I did that in an issue, because we got someone reporting that us that'll work. So we did some search there. It's not doesn't work. Basically, we try to do is search perfect list so that we can look up a DNS domain name with short instead of the full fqdn and then we custome issued there.

F

So if you can open up that peer, not that issue report, so you can see some details there, so what we say that it's first time if we configure multiple surface search list last party for some article I come in as such lists over the first one work when that's the later ones, and so that's the first issue and we we want someone to take a look if that's our msconfig or if we need turn on some places, because I believe, as we were probably also use that right, because we need to add multiple search term ends in that surface search list.

F

So that's the first issue, then. The second issue is we from their term. If we use the club or is Carl, the result will be different from what we, if we use Windows partial commands. Like a result, the u.s. named or in work web requests so that cause I'm confusing. We first have because we say: oh, it didn't work and then we figure out. Oh if we use Windows command it worked, and so we, what we're wondering like and for this kind of command incompatibility issues, will someone take a look.

D

For the first issue you are mentioning, are you.

G

D

That, whenever you are using partially qualified domain names with a dot inside them that it's not resolving, like you, I see you're like trying to do results, DNS name one web server, two dot.

F

Okay, so basically we with that right, if like, if you do from Lennox rack, you can do just real web server. If that's for going. If you are in that namespace say like in my case, I mean test namespace. So if I do win web server to which is service in testing space network, but if I, don't you look up at servicing different name space, so when web server that be Forge, it's in the search list right, if you Linux it, what go? Try the second I get what try everyone right.

F

So if we from the second one, which is one web, server that before Passover that look after local, they found that work. They will return me this way. I can look up a server service in a different name space with partial of Kadir, but that didn't work here. Yeah.

D

I think that's a difference in the DNS clients between Linux and windows, I think it's documented on kubernetes windows on the Kid style, Docs, there's a section on networking and there's differences of DNS explained their basically. If the DNS query you're sending out contains a dot, it's it's treated as if it's a fully qualified domain name, it Windows clients. So that is probably what you're seeing this issue. Okay,.

F

So you're saying that that's basically be for behavior and then what? If so, if it doesn't have that so example, I didn't add here if she says googled are local, which we were added there and then, if you don't have that it is something that will go down local and it will work. Yeah.

D

Just Google should work yeah or Google yeah.

F

Okay, okay, so.

D

Anything anything that has a dot will be treated as fully qualified. Oh.

F

Okay, so this is a difference: okay, yeah.

D

There's our there's a work item.

F

D

The TNS team to make that configurable so that it matches Linux behavior but I, don't know when that will be done. I would have to talk to DNS team.

F

B

Test cases we also exclude all the N dots test cases for the same reason: okay,.

F

And then, how about the other ones? Soap is like curl or DNS lookup that didn't care by his team, or they are handled by saying.

F

D

I would have to double-check, but I do know that, and this lookup doesn't quite like query the same same parameters: I guess, they're the same rich registry keys that coral and resolve DNS name uses since it doesn't demonstrate like DNS connectivity. You should use resolve DNS name, broader done look up, but I can I can double check. What exactly and this lookup is query yeah.

F

It's just confusing because either you don't support it all your support for, like it's confusing it sometimes work sometime it does this thing. You know yeah.

D

So resolve DNS names should always work, if not that there's the resulting.

F

List name: it works I'm. Just talking about those commanded, we are used to England.

A

So so can we take this off lines since in the interest of time, since our turn together release out this week? So generally, let's discuss with with the team I would David offline. But you know the biggest thing is that Windows and Linux tone of the same set of tools that are expected to behave the same either right so take it offline, please, and if we need to update some dogs for guidance, and we can do that.

F

A

Free to ping me on slack.

F

D

B

So so David the that bug is linked in the meeting notes. Today. Can you go add a few comments to it to just clarify that DNS behavior there.

D

B

Last thing I want to mention on this is that we did merge some tests around custom, DNS servers, and so there is a wit if you do, if you wants to your review or add test cases that are specific to Windows and DNS, there's already a file there. So it's easy to do you know tests are. They are passing today for using like DNS server equals 9 and then specifying custom resolution in there. Ok.

F

It is consumed earlier because after I let alright right before I think it goes through the common code. So it would allow specifics in so if that doesn't work, but it's fun knowledge, and also that part myself is confusion between. It says that it's not supposed to work for other thing I, but for one I try right. It actually worked. Yeah.

H

I think that bug was just not updated. Does it's been added to the overlay and wind bridge CNI plugins for a while now.

F

But that part was done up there. That's why I finally said oh, this works. Ok, yeah.

H

I'll update it with a link to the PRS I should have closed this issue.

F

Thank you very much.

B

Okay, thanks to drive it I, guess back to you Michael! You had something else: I.

A

Don't have anything else there on that table, I was just gonna mention one more thing to to David. The Peter. Also linked is that DNS recommendations that we'll have in our dogs? Could you just basically talk about their? You know any slope if you put one more sentence that you know advise people nslookup is not the right way to do this and give them the right guidance. That'd be great.

D

So prepare for that. Thank.

A

A

Alright, so so from high level, let's kind of recap really quickly, so the CSI work, the work is checked in the dogs. Pr needs to be finalized on the GMs. A work is checked in the dogs, PR and there's some testing work. That Jeremy is working on to to finalize that, as that as well as the docs is already merged and then Cuba diem, the dogs, PR is getting updated, but code is checked in sounds about right to everybody.

A

Yep, alright, so we're in very, very good shape. I'll, create the release, notes and I'll circulate them with everybody Patrick. If you're on vacation you'll need to check it, but if you're around I'll also to use wall, so you can look at it and then I'll create the slides, probably tomorrow for the meeting on Thursday and then deep, your you have about five minutes for a demo.

A

If we end up in with any issues, you don't have to give it them all, but it's good opportunity if you want to showcase and- and if you want I, can do the one-minute intro about why GMs a is there and why we need it or you can do it is up to you. It's totally fine either way all right. That's it for me,.

A

Anything else for anybody any questions and concerns any help they need for the dogs, PR or anything else for today,.

A

All right, well, I, really have a good rest of your day. Patrick have a great vacation and we'll see you guys next week, good luck with all the PRS winning today. Thank you, I think.