Kubernetes SIG Windows, 12 Feb 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Windows 20190212

Description

Kubernetes SIG Windows 20190212

A

All right, hi, everybody welcome to super windows. Meetup. Thank you all for attending. We don't have a lot of items on the agenda today. So there's a few general updates I'll go ahead and make them, and then you know, if anybody has any topics they can bring to me. So so let me start from the talks that Craig had posted Craig. You wanna talk a little bit about that very quickly. The line sure.

B

Absolutely so I shared in black and the email list outline for the upstream documentation for the windows containers intended to land in the docks during 1.14 we're gonna go GA. At that point, the conversation I have with the sigdoc team about a week and a half ago was around the you know the fact that adding a new node type impacts a lot of different things within and.

B

Discussing the trade-offs of how to figure out the the best and most effective and efficient way to create the right box- and they asked from the exact theme- was instead of to try to make PRS to a bunch of different pages on the website, right away, to consolidate all of the content into a kind of a coherent document which we can then make a decision about whether or not it makes more sense to have those kind of distributed changes to all of the different pieces or to have you know a more central windows container guide for kubernetes and then have small changes to small changes to the relevant pages where there need to be specific, call-outs and so I was asked to put together the kind of outline.

B

So what I did is I created the outline by consolidating input from prior outlines and content, full request for prior release? Attempts of this and the information from the cap and I've consolidated that into the outline and I, would ask for anyone who has the time and energy to take a review of that outline so that in the next week we can kind of finalize the outline and start working on the content itself.

B

What one idea is to first thing is to figure out whether it makes sense to fill the content out in the Google, Doc or I. Don't know much about what it takes to reformat in Google Docs into markdown, but I need to I'm going to talk to the sigdoc spokes later about it today, but the the near-term ask is to add your input to the outline level content. There are any specific things you think need to be called out anything else. You want to add.

A

To that no I think that you know there. The outline is really good. It encompasses a wide variety of areas from user guides to installation to what's working. What's not working has a contributor experience, section I, think that you know we're gonna look into some members of the sig to come in and fill in some details around especially CNI, so I just wanna see a lien and ban to talk about and David as well. Obviously so Ben and David can talk about flannel Eileen about the ouvea.

A

No, yes, so we can fill in some of the sections around networking. But overall, this is a good outline that would keep enough information to end users are coming in about what's working. What's not what kind of types of apps they can bring into Windows and and also a section around experimental features or features, are coming down.

A

The line like JMS a container d, calico and and other things in our near and long-term roadmap, so I thanks Craig for for basically putting everything together from all these various sources, including your own input, and once we know how we should start filling in the content where that's markdown or Google Docs.

A

Let us know Craig, and at least me personally want to pick up a few sections and- and the one thing I proposed on that was, we can have people are willing to contribute to pick two or three sections each or subsections and give them a deadline right.

A

So they fit the fillable scene, who can go through a round of review, make sure that we're all riding within the same type of content and detail level and then adjust if necessary and then to the next or three sections and who can complete the docs that way from a from a timeline perspective.

A

You know, even though the dog's deadline is far away. It really is not. When you look at the kubernetes 1.14 release timeline, we will have code freeze on March 7th, which is exactly three weeks from now. Pretty much and then the dogs seem to be completed 11 days later on March 18th, so my goal is: if we can get the dogs in a good enough state by the end of February, it'll, give us enough time to do reviews and iterations and and and all all that stuff.

A

So definitely, even though we have a month really time flies. It.

C

Does and I just wanted to note that I've.

B

Lined up some Doc's resources to help us with editorial and formatting work as well.

A

And I put the link to the docs that the Craig created on the in the meeting notes all right next by the way, Patrick is not joining us, since you guys know Seattle is in Snowden the only thing you can see the spacing, though, so you guys haven't seen that picture that she marries by the way that has that, so he won't be able to join us.

A

Alright, did you wanna give us a quick update on GSA I know that your PR to create the new report is not complete yet but you're getting there. Yes,.

D

Request is in I'll, just basically ping Erin again today to get it through and then other than that the entry PR is still waiting for review. So please take a look, no chance. I paint you too, as well I, think she was out last week. So she just got back and acknowledged. Saying it's in our it's in her head and she'll. Take a look yeah.

A

Unfortunately, I was also out last week, traveling, not out of work so I'm, basically in a pile of I'll, get it as well. My apologies no.

D

Worries no worries so yeah, that's kinda in a holding pattern and John is making progress on the books that we described in the gap. So basically the web books are being developed in a private repo and as soon as that, Cube repo is up. Leadership is.

A

Excellent, thank you, I think Tyrone on your repo last night as well. So just give you my hours and she doesn't reply all to him again. Awesome.

D

And Jeremy has been helping us quite a bit with test setups and debugging various DNS issues, mommy.

E

D

Test, but alpha does not have the automated test in there so.

A

Deep in Jeremy the section on GMS, a under experimental supporting the docs that we talked about earlier well, no rely on you guys to film a team that will include I. Think of these around, like two three things right set up: how to use and what's the net effect to the user right, you know what what are they getting out of it so figure out.

A

You know how you can a section that- and you know that I'm also abide by the same deadlines that we talked about earlier, we'll take a crack at it and we might, from from your perspective, you might have to update a couple of sections you might have to update the window section, but probably us the section where they have some of the security stuff.

A

On the on the for Linux containers like Ronnie's username, for example, you should have a link there as well, even if the link says go to these Windows dogs, because when people are looking about how kind of identity can I put in my container, we need to make sure there are where the other windows are. Some special things.

A

Clàudia your Adelina, who wants to give us a quick update on the testing effort and where we are I, know, there's a furious rally of lots of peers that we've got to review. The last few days fixing little things here and there.

A

Can your sunny a little bit muscle damage, that's just me or everybody.

F

I'm guessing my microphone goes through the.

G

Better now, yes,.

A

G

A

G

Yeah, okay, so basically what happened? Updating a guy's engine they guys named John Bertrand, also updated the cni, the ages and I version. It seems that it wasn't necessarily that stable arsenide version, 1 0 17- that caused some test to fail supposed to fail to spawn. So when a pod fails to spawn the framework, waits 5 minutes the hard deadline, it waits 5 minutes.

G

So if a bunch of these pods fail to spawn, we just add 5 minutes in test time and we basically hit the awkward timeouts that prowl gives us and that's already 7 hours.

G

While we did basically is. We went back to Asia CNI version 116. That seems to be better. That was long, we're using until now also we removed a couple of tests that are flaky, 80, 90 percent of the time I'm. Looking for the link until we figure out what what we can do, those tests there's no point in in running them, because you just take up time.

G

We also increase timeouts we're looking around some other tricks, either Vegas engine I also added a patch to randomize the age of zones, Asia locations where these clusters are spawned seriously everything possible to reduce the probability of it, either being a slow deployment or having a lot of timeouts.

G

We also consider we talked I talked to Patrick on a male thread. We are so considering for the Windows testing windows. Tesla should be added new windows tests that we create a separate stage, job and use that one to say, add new tests until they're, stable and then move them to the main job.

G

The one we see on test, because we put only that one job, any tweak, we do might cause failures, or so we want to keep that one stable as possible and just have a staging one where we best things for first of it, especially since we're going to add some new windows tests.

G

Yeah, that's that's mostly. On the upstream side, there are a couple of issues we are with tracking relating to DNS.

G

But other than that, the stability is our main concern right now,.

A

Guarded so I have a few follow-up questions if you don't mind Ellen and Claudia, so the first one is this display get tested. You said there was like 80 or 90 percent of the time, or this.

G

A

On your windows, are they flaky on Linux as well on.

G

Windows there are windows, just a second, so I can retrieve that. There's an issue opened.

G

Its if you can test me, it's the max container back of tests.

G

Its issue: seven one, nine, four, nine, okay.

A

Yeah, we should take a look at that and make sure that if we basically to understand if they're not flaky on Linux, why are they flaking windows right? So we need to understand what the tests ask if it's a timeout something else I don't want to read through the ticket now, but I'll I'll do that later. So our biggest concerns are the uncovering, an issue that you haven't thought about before, or it's potentially dangerous to us. So that'd be the big good thing to investigate.

A

My second question is.

A

How, where are we on picking the image for Windows, and we talked about the docker hello world example? We talked about a golang example and basically, all the things that we talked with Luba Mir, both on the on the ticket as well as in a couple. Various forums was that the need to have this distinction between when the test runs on Linux. It needs to pick a Linux container.

A

You know the smallest 40 megabyte image that they want and when it's on Windows is there 123 megabyte, : or hello world image? What are we in finalizing that the PR was still open yesterday last night,.

H

Amir and I create with is creating a separate test with an image that we build ourselves right now,.

H

Request basically has a new test which is pulling a busybox image, which has a manifest lists, which.

H

It has the images for 1803 and.

H

A

We're going to build our own image, is it going to be smaller than the docker hello world? It's.

H

Busybox image, which is basically using at least half of the end-to-end tests, so basically that image should already exist on the on the nodes and at the very least, the image is based on the witnesses or image which already exists on the Mount. Okay,.

A

All right, so, let's, let's get that out there and close that thing: Thank, You, Claudia.

G

Another thing regarding the manifested so on Windows testing: there is this PR that cloud you made about what is my first list, so basically, all the images now under these private repo, our manifest lists. We then change the one under the each each. We team official repo- if you want to manifest this, if there is somebody from Google or who's testing on GCE the jobs for the GC, we might need confirmation that these manifest lists works. Work for the jobs as well before we move everything to UE.

G

So for our job on Asia, they work just fine. It will pick the the proper image for the OS version, but I don't know very much about.

G

What the other windows, testing effort for GC is using what version about our version of Windows.

G

So if you can confirm it works, we can move to to the new to the official repo with everything our test, to use the new images and see if it works we'll get back through. Please think I also think I think I think you on on that issue under poll requests, so you can comment there, don't be great yeah I'll do that Thanks. Thank you. Thank you.

A

So you just also asking on the flaky test so some something similar to what I asked earlier right. So we need to understand what those tests are just removing them. It's not an option unless you know for sure they're not applicable to Windows right, so my need to at least open a ticket to track that, so we don't fall, get those tests were removed. We.

G

Have it so that the issue is in reference to those tests.

A

The PR is closed, oh so that one's up here is closed. Nobody will remember to come back to it. Oh.

G

G

A

Yeah we need something to try to remove this test and we neither need to bring them back in modify them or something else. So if you can create that that'd be great. Oh.

G

My figure I see the riot was closed, I think because it was referenced in an issue, so shouldn't be closed out, reopen it. Okay,.

A

G

Actually, I see the Singh windows, automation, automation, the synchronous project, moved it to Don and I. Guess, that's I, don't know but yeah I. Should we open it? Okay,.

A

G

Do have we do have tracking and this and we're looking into it. So it's it's not Rader. All.

A

Right, excellent, so and I also finishing up that work with the picking up our image would allow us to to make a lot of progress. So you know this is more of a qualitative question work on today. But how far are we from getting a hundred percent green conformance? Like you know? What are we today well.

G

If you look at first grade so with the exception of these problems that we had with the spawning clusters, the only tests that are failing consistently are mmm a test that we have a PR to remove.

G

So it's cubelet when scheduling a busybox pod with most areas should remove, should write interest you EGC hosts that cannot ever run on Linux. We have a PR to add it to Linux. Only out this one can ever run on on Windows, because we can add mount files, but in the previous PR that that tag for Linux on it was omitted. So we have the nippy opportunity to success.

G

The DNS should provide units for the cluster and the demon set test are the only ones who are what which are consistently failing. At this point, the daemon set one is flaky.

G

For the previous runs, I would say 50% of the time it fails so, but that is a timeout issue, I think so, looking through through the times it took for a test run, it just doesn't match you to finish in time. I think I do need to investigate a little more but other than that test seems to be the green. um The only the most important thing we have to do this bit is stabilize this runs because for the last for the past week, clusters had problem spawning and so on so forth, and then.

A

I notice that Claudio has his own repo were severed. 2019 versus the Google repo that has 1803. When are we going to be able to merge and use 1809 in the Google through the Google repos I'm.

E

Working on that presently and by the end of this week, I expect we'll be able to switch to 2019 I. Think.

I

That that issue about adding image manifests list the.

E

I

Testing will solve this to where it should be one thing that could be 1803 or 2019 exactly.

G

It can be used for both age as confirmation that works excellent.

A

Thank you because.

G

The last time we switched images without which the repos, without asking anyone we broke Google, so Google's nesting.

A

Alright, the last item: for me, we had our first engineering stand up last week. We have another one this week and we're going to continue those until the end of March or until 1.14 releases.

A

The biggest part of that is that who had a backlog that was huge, 20 items or so now it's down to 7 and most of them are in progress. So thank you all for not only folks on this team meeting, but a lot of folks that don't attend this meeting. There are have picked up items and they're starting to contribute so a lot of items in progress so including a lot in review.

A

So there's 20 items both in progress and reviewed together, split 50-50 and you know we'll continue that progress and I think we'll be in good shape for for 1.14 Ben. You want to talk about the PRU added on the docks. Yes,.

I

This is adding test results for we're at pivotal. We have this product CFC, our that's a way of deploying kubernetes and we're running the conformance tests and yeah we're trying to add it to the test grid, and so we can be another point of agreement or hopefully on running all the conformance us in another environment, rardy.

B

I

We're already seeing some like difference: yeah I think we have more failures, some of it stuff, I related to known issues that we're trying to work around and some I don't know we'll have to follow up and see why we're getting different behavior.

G

Do you have a chance doc word something some results you can share? Maybe there are known issues we know or know how to work around yeah.

I

I can show you like one of the latest logs of a run. Yeah awesome.

A

All right, I think that's all the items who had in our you know in our agenda, for today anybody has a little concerns or anything else I want to bring up. We have a couple of minutes left.

A

All right have a great day: everybody stay warm if you're in an area that super cold, otherwise we'll see you guys on Thursday, some of you and the rest of you next Tuesday I see.

I