Kubernetes Azure Provider Subproject, 7 Oct 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Azure Cloud Provider 2019-10-7 Kubernetes Cloud Provider Subproject Meeting

Description

Azure Cloud Provider 2019-10-7 Kubernetes Cloud Provider Subproject Meeting

A

All right, good morning afternoon evening, wherever you are, this is the at least specific time. October seventh meeting of the cloud provider as your group and I want to welcome everyone.

A

So this is a community meeting and the Cooper neighs community and that's such we're all obligated to adhere to the community standards, and that means remember that this meeting is being recorded and will be posted to YouTube and that we all should be good to each other and with that I will paste the link to the notes and the chat and anyone who has not yet, please add your name attending.

A

Oh look. Steven Rd did that. Thank you and I see you're, taking notes. Great I would like to welcome Aziz, who I have not seen in this meeting before we're a small enough group. Maybe we should just do a quick round of introductions, so we all know who who each other are I'm Craig Peters I'm, a p.m. at Microsoft working in the open source container infrastructure area Aziz.

A

Maybe you can introduce yourself Lee yeah,.

B

This is Lisa, I am currently working with VMware and we I will give a part of favi networks, which is quite a quite by VMware. Exactly and I have been working on different cloud deployments of IVs and loan balances, so we get deployed in different cloud environments, I sure they do and we have been exploring IKS as well as different kubernetes expansions. Also. So when shift and other environment so I was interested in what's happening, is group and wanted to participate if I, whatever I could but great welcome.

B

C

Hello, Aaron Humphrey, working from also Microsoft Anna, mainly focused on a case and improving the integration of communities down era.

A

Likes Jack I'm, just going in the order that your videos show my screens good.

D

Morning, I'm Jack Francis I'm, an engineer Microsoft primarily working on a KS engine which is a cluster provisioning library for a KS and, in addition, has some tool chain wrapping on top, so folks can build sort of ad hoc self managed to various clusters.

A

E

Everyone I'm Steven, Augustus I, am a senior cloud native architect at VMware. Also, the emeritus chair for Sega sure I work on I work on a bunch of different things: some released us some PM stuff and also I'm one of the primary maintainer for cluster API Azure. Oh.

A

Let's go down the agenda, I'm really glad to see everybody here, it's great so.

A

The sign up from the last meeting that was pending was the bucco migration and I see Stephen. You created a tracking issue.

E

Yep, so I will I wanted to wait until it was like a reasonable hour to send out an email and people would be paying attention to it. So I'll send out an email today regarding regarding the migration I doubt there will be too many people up in arms, but you know it's something to be aware of, essentially for anyone who isn't aware we're planning on migrating the cloud provider, Azure repo from kubernetes from the kubernetes work over to the kubernetes cigs org. That is just for consistency.

E

We're sig, based projects should be under the kubernetes, a gorg as opposed to kubernetes. There may be some concerns around name changes of repos for github.

E

The first name changes is free, essentially, they'll, do a redirect from kubernetes the kubernetes eggs, so that shouldn't be an issue and then also when doing imports for for the cloud for the for the go repo itself, we have to redirect or set up kpo and say that gates that I, oh so say, batkid sattaiah, a slash cloud provider will point to will point to the correctly, though so I'm not too concerned about the the imports but I'm curious. If anyone has objections for any reason before we we go for it. So.

C

For this items, would the Odin go import pass deal works so it way do some such as go cat with the original wood wood. Then a steel.

E

Work so the go get with the original should still work, because what it'll do is you'd be go getting kto, slash cloud provider asher, which would redirect to kubernetes slash cloud provider badger which would redirect to kubernetes cigs once it's done so as long as we only do one redirect, we don't like, we don't change the name twice. It should be fine.

C

Okay, so wait: wait, don't need to change the the go past it right.

E

We should change the go path, but.

C

E

But if you have an, if you still have the old go path, it should be fine. For now we need to change the go, we'll need to change, to go path to cig dot, k, dot. Io SIG's got case that I, oh okay, yeah, yeah it'll, be.

C

Ok, so yeah in that way, they are actually a lot of things. We need to change, as I said, the testing from the pipelines to publish the image and so on, and also in our course they. They are also some go past here, yeah, so.

E

I think I I think the initial stuff should be like. The initial concern should be fine in terms of translating the the path correctly. We should we should double back and fix it. All I'm saying is that through the essentially through the redirects, it won't break or it shouldn't break, but this is something that I would like to try. Maybe you can you can help me enumerate on that issue, things that will have to change as well.

C

E

I want to I want to try to move on that as quickly as possible, because I think this is. This is a time that will have the lowest amount of people using the repo rate yeah.

E

So yeah, when you have a chance, just comment on that issue: I know I know. Basically everywhere we reference the repo and we can do a hound search for that, which should be fairly simple. Here is hound. If you haven't had a chance to use it before.

E

E

Alright and hound basically allows you to four people on the call who haven't seen it before it allows you to search arbitrary keywords across all kubernetes repos, so you can use that to to find things that you're trying to fix.

A

A great other ish other comments around moving the repo I guess. I do have one question which is: are the other cloud providers also moving the community six or have they already done so.

E

Most of them started off there a decent chunk.

A

E

Them started off there, so the newest ones would have started off there and I think the the ones that were traditionally in kubernetes might have been like AWS and Azure. So I just want to make sure that we're setting the right example in my grading, I.

A

Think that makes it really was just wondering why so yeah.

A

Okay, great I had put as the next item to go over the board and look at what should be done and 117. That makes it sense. The folks I would.

E

Say let's do the board last and.

A

E

If they can jump through these, are you added some items? Okay, yeah, great okay,.

A

So Stephen did you add the fix, aggressive, VM cost I did.

E

C

Yeah, so we have met a lot of issues for our big cutter, so we need actually a lot of steps to to fix the easier I have created the one Phe replace the ocean breeze because at least so then the number of requests would not depend on the number of nodes, and also we needed to cherry-pick this PR to order releases. Also, I I would also do a lot of other works to to to fix the issues.

C

For example, we would order the retry after header for and the response taste, and also we would other really really try and lame is based on the responses. So there would be a lot of code changes there, so so I had only expected, then changes cherry-pick T or the releases, but for this one I would expect it to her pick to order release.

E

Yeah, this seems I'm just scanning through the code now yeah. This seems fine to cherry-pick back as its as its fixing a bug. I would I would say just just tag me as well in in the ones that may be considered for Jerry pics, yeah I, just because I know, there's been a common pattern of uh of. Is this really a bug fix or you know, is this a bug fix or is it a feature for the provider? So let's just yeah edge it so I can help out with that conversation.

E

This is. This is a valuable one to have in them. So I think we can. We should cherry pick it back where we can yeah.

C

Actually, we don't, we only have four days, I think for the finished.

E

C

E

Yeah I think you're right on that one.

C

So I would try to TT send out the church are picked yes, as long as possible. Mm-Hmm.

E

Oh yeah yeah, the deadline is the 11th yeah. If you probably sent out the emails in here or someone sent out the emails yeah.

E

All right yeah this should this should be fine to cherry-pick back.

A

C

A

Action item, though so just to close that one out so Stephen you have an AI to talk to someone about that, or is that.

E

mmm No I've just got a I've just got to watch the stuff for approvals time value is going to open cherry-picks for the for the older release branches. That should be it.

C

Actually, I can approve the terrific PR because I'm also the patch release manager, I.

D

C

So I just need another one from from ceará to approve the char Pig. Okay, yeah just tag me when you're ready.

A

Great so the next item I perking, my migration out of KK.

D

Yes, I added I, don't have anything specifically to say about it other than I think we should track it, especially in this 117 release cycle, because it looks like all all signal is that that's not going to be delivered with 117. So.

E

um So yeah, let me actually assign myself to that one, because I'm handling that yeah, so all we're trying to do, is essentially move it out of tree I. Think that that's doable within the 117 frame time frame but you're right that it does mean that we won't be publishing those artifacts within the standard release cycle right within rather within the standard release process. So but that doesn't mean that we won't publish the artifacts at all right it'll, just be a separate out of tree or release.

E

What this does allow us to do is move faster on hypercube if we need to right. So if there are things that you need out of hypercube feature requests or interests in maintainer, ship, I, think I. Think part of the reason that we want it shifted out of tree is: is it is kind of like DIMMs is saying, is kind of a magnet for pulling in other dependencies, but if we have that out of, we can kind of manage it as we need to I know the Azure folks express interest in in keeping it around.

E

Some other people have I know Dalton, who maintains typhoon, also uses hypercube. So what I'd like to do is find maintainer for this. Ultimately, cig release will be the steward for it, but I'd like to find maintainer x' as well. So if you know people that are interested, let me know playing fair I. Think you mentioned at some point that you are interested.

D

E

Least, I don't want to.

D

I, don't want your doing enough already exactly.

E

I was like I, don't want to automatically signed you up for me, the inertia but.

D

Craig and I can talk offline. We have some folks, I mean if there's sort of general community interest of keeping this around and I. Think we because we're one of the stakeholders at least up to this point. We should do our part and maintain it so yeah.

E

D

I can figure out how to do that. Let.

E

A

Think that Jack and AI is for Craig and Jack to discuss.

A

Microsoft, invest me.

A

Didn't think hey, okay, before we go to the board, are there any other items anyone wants to bring up?

A

No great I will share.

A

Up well sure, brother.

A

A

So what we have here is the the board, that's moderately maintained and we have a few items that have been put in the 117 backlog: column- I'm, not on a processor. How any of these guy here but I think the first thing we had to do is go through and just make sure these make sense for 117.

A

That's not sound right to folks, okay, so.

A

First thing is: if they noted that we need to benchmark the as your request from the clock control my manager and looks like Rita is assigned. This makes sense to me in this timeframe, because we have the API throttling issue. Thankfui. Is this specifically to help us get our arms around the API throttling issue.

C

Yeah so for this wine, so with benchmarks, it would help a lot for for katom t to to have some reference to set up the client side.

C

Becouse really me so this is required actually be called the some custom running a lot of clatters to within one subscription, but the areas throttle it based on subscription so at at the at the dock siders. So there are only half thousand ready requests per hour and the first subscription, so we needed this to kick either cut them to set up their Collider rich, limiting okay,.

A

So your vision will change to talks based on this right.

C

A

And open push called no manager image. Obviously we need to do that.

C

Is this one had already been down so I I think it could be no well.

E

Hold on actually this one is, it looks like we're going to be publishing to MCR, like a community image should be published to GCR if they're, a and part partially for visibility right being able to control those jobs and have those jobs be controlled by anyone in the community. So my ask for one would be too: there is an image, pushing job that works now, I'm, not sure if you saw that work from Catherine, but let me find it.

C

So check this out. Yes, so do I have any guys tt public to the also TV idea. Yesterday, I think we we we, we should publish it to post, maybe because area where we we only use MCA. We don't you gia, but for community I, think GCI. It also makes them yeah.

E

So I just I just pop that in the chat, let me actually put it in the notes. I was.

A

Good I was gonna. Ask okay.

E

A

E

A

E

Yeah I can do and a so basically what happens? What happens with that? One is after after a commit finishes: there's a there like once we have it all set up a post sim, it will run that will essentially just publish it to staging bucket right and then, during the release process, we you can decide to both run the Jenkins job for the repo, as well as doing a permanent image promotion right.

E

So the image promotion to land, the SHA of the of the staging bucket image into the prod bucket, and that should be fairly straightforward, but yeah I can I'm going through the process with a few different repos right now so I can I can help out there to.

A

Appreciate that.

B

Just one question: Steven on benchmark numbers of requests so generally has a provision for upgrading. The numbers are, like you know, increasing and enhancing the API limits right. So.

D

B

Saw that you know, if you do the changes based on 12, K and one point: okay, won't it, let's suppose somebody who has increased their numbers, won't it impact them. I.

E

Do not know is the short answer we should. We should probably find that out, but I imagine that this would be establishing a new baseline.

B

As opposed to overwrite anything only thing, I was asking, you can figure it, but that's what I so.

A

Yes, I think it is the best thing to do is because Rita is actually working on that right now, if you could comment on the issue that would yeah.

C

A

Can know thank you. Okay, I am noticing that we are almost to the end of the half hour and we have only gone through two issues. I am going to propose that we have another working session sometime before the next meeting of this to work on 117 planning, I yeah,.

E

We can, we can totally do that. I would say we can smaller circle if we want to for the planning session and maybe.

C

E

It an hour and I'll.

A

Send an email out to the to the whole group: okay: we can do a little doodle and get a couple of dates. Do.

E

A little doodle yeah so.

A

E

Sounds good okay,.

A

We'll do that, since we only have a couple of minutes left and well, ask also is: are there any other issues we should cover today.

E

um I had a question actually, unless someone else does: okay, all right uh-huh. So it's a question about the I realize I forego a ky realize I actually approved the PR, but it's a question about the secret based a cloud provider configuration so we're at the point in cluster API Azure, where we are there's a way to inject the the cluster API, the the cloud provider config, and it's essentially like we land the the Etsy kubernetes Azure chase on within within a config file right, the bootstrap provider for cube ADM.

E

Is there a saner way of doing that where we can leverage the secrets? Instead? Is there a quick, read me on the config or something.

D

There is a config map based vector that the cube liquor consumer, isn't there I.

E

Believe so I haven't used it before, which is why I was thinking yeah.

C

Yeah we have idea, the secret is 15.

E

C

It could be your duty to set up the secret and also.

D

C

We also support MSI right, so so you don't need the credentials anywhere. You just use the the tokens from a method. Yeah.

E

We're gonna have to we're gonna have to wire cap c2 to support Emma sighs. Well, it should be. It should be like one flag change, but well.

E

A

Okay, well, thank you. Everybody really appreciate it and with that, I will send out a message about planning and I hope. You guys have a great day take.

E