Kubernetes Cloud Provider Special Interest Group, 2 Mar 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Cloud Provider 2022-03-02

Description

Leader Migration GA Graduation, clearer docs on deprecations, annual report and more.

A

All right, hey folks um today, is march 2nd, uh 22., okay and uh yeah. This is the bi-weekly uh cloud provider meeting um so yeah. We changed the format a little bit uh in the last few meetings to do triage first, so I'm gonna go ahead and um just start doing triage.

A

I feel like some of these reach triages.

A

B

A

um Let's start with this one, this is a failing test for the uh cube control sku test on gc.

A

I would think this is like a sixth cli.

A

A

Yeah this is 601, so I will remove us.

A

A

Okay, the next one is azure load. Balancer is never updated when no status changes.

C

uh So bridget bridget did add.

A

C

About that in the agenda that uh she was gonna, look into it and she find herself cool.

A

Cool, uh I will.

A

I thought this was actually fixed. Already: okay, well yeah I'll, let bridget take it, but I'm just gonna triage accept it for now.

A

Thanks for that, uh okay and then the last one cubelet service cannot bind on port, um both ipv4 mpv6, dual stack.

A

It's probably a node issue, but maybe there's a cloud provider component here.

A

While the kubernetes do both, uh let's see.

A

I see so they they were able to get cubelet with dual stack working without the cloud provider, but not the with the cloud provider. It's not working um from the club and the outside has cubit just to correct the details.

D

A

Okay, so this thing is just like a documentation thing where, if you have a cloud provider, you can't override that that you can't force dual stack. You have to let the club provider do it, which I think it's expected.

A

C

On a side note here I feel like, is there any way that we could put a pr in to update some of the documentation, because I'm noticing that last message there about you know still valid in 1.23 on the cloud provider and someone showed up in the in the chat channel next to slack channel. I think a week or two ago was also asking they had some question about this.

C

I've seen it come up multiple times now and I think people are getting confused because they think that it's been deprecated, because the official like site docs, say it's been deprecated in 1.23.

C

We probably just need to update that or something.

A

E

I agree but yeah, I think the sorry just just quickly. You mean the cloud provider flag, umico yeah, that's fair! I I will say I believe it has actually been deprecated.

E

It's just that I wonder if maybe they're not they're a little confused on what deprecated.

C

Means I think I think so that was that that was my impression. Is they they believed that you know the signaling for deprecation meant that they couldn't use it anymore, and I was trying to like yeah.

B

It's not clear that the external option, for that is the opposite of the deprecated options. Right, like the um it kind of sounds like the entire flag, is deprecated right.

C

Right, I just think yeah. I think it's just not as clear as it could be. Let's put it that way.

E

Yeah, sadly, I I think I'll make a point. A lot of people may assume, because we've deprecated. That means you you can you you, you may be able to not have to use it, which is not where we are.

C

Right yeah, maybe even just adding a link to the deprecation policy in that place or something.

C

But yeah I just I happen to see that comment there and it it made me remember this conversation I had and then you know I've seen it come up multiple times.

A

Yeah makes sense: does someone wanna own that fork to clarify like the docs or the messaging around it and the various components.

C

I mean yeah, you can assign it to me since I brought it up. Okay, nice.

C

I'll look into it, put a patch up and pick a few people here or something.

A

That's good and I can assign you to this issue too.

A

I think this is yeah similar similar need for better docs.

A

Okay, I think that was the last last one. Let me see here.

A

Oh, it's just not oops.

A

Daddy the same thing here.

C

A

All right, okay, great! Let's uh move on to the agenda items then, uh okay, so I see an update from bridget about the issue that we just triaged. So let's get that uh I don't personally have updates on gcp uh walter or joe did you?

A

Oh sorry, I missed aws um nick did you have any updates on aws here.

B

uh I don't think I have any um sure, germany, I don't either.

A

Okay, cool yeah, uh yeah gcp- I I don't have, I think, like the yeah.

F

I can't think of anything either andrew um I'm just going through my head, and I think there are some things going on internally to make sure release is going out and stuff like that, but there's no nothing really newsworthy.

E

uh I actually have one thing that uh maybe I think we have inferred a few times on gcp but uh from jared, uh which is that, at this point I believe jared has been doing cops. Testing use uh for gcp, which we haven't actually made a d any sort of significant deal of. But it does suggest that there is now a sort of somewhat viable option for people who want to look at using uh kubernetes on gcp to use cops.

F

Yeah he did a couple things there that were interesting, um so it's good for h, a testing in particular. If you have an h, a testing gap, it is a solution.

F

um Jared also made sure that the leader migration stuff was integrated not into gcp but pretty much across the board.

E

Well, it's also worth noting that I mean, if you are wanting to do kubernetes on gcp, and you want to ha cops is now enough is, is an option which cube up is not honestly.

F

It's the best option, probably um cube up, is I don't know, do you have any statements on that walter? I feel like it's going away slowly, uh if I can get an owner for it, it will.

A

As with all things in open source, um okay cool that makes sense, yeah jared, feel free to add anything else. On the comments.

A

um You spirit, lebron, do you wanna, add anything there.

G

Yeah majorly just back fixes and we're also working on adding a tv pipeline. We don't have some more payoffs at the moment, so it's located into using the just these very much and using the vmst environments for testing.

A

Yeah makes sense.

C

I noticed john's with us from ibm too hey john good, to see you hey.

A

Good to see you do you want to provide any any notable updates on the ibm provider. Oh no! Nothing! At this time, cool.

A

All right uh extraction, migration, sub project. Oh so one thing I wanted to discuss about the extraction migration project. um Is we so you know how we push this meeting back to 9 a.m? Pacific, I'm wondering if for the bi-weekly sub project meeting, we want to push it back to a similar time, but on thursdays.

E

uh I I I think it is a good proposal. I think the right way to move forward on. This is the same thing we did for this meeting, which is to put it up for a vote.

D

uh Yeah you're right.

A

A

Or you know what.

A

Sure, or do a vote during next week.

E

We could also just do a vote by mail. I mean the one thing I will say is that doing a vote during a meeting uh is pr preferentially gives uh power to the people who, like that meeting time, that's.

A

Good okay, andrew to send an email to the sigma list.

C

Given uh given our changing meeting times and kind of the volume of traffic in our slack channel, I wonder too, if we might set up like a bot that could just broadcast like five minutes before a meeting starts, or something like that. Just to help people out I've seen other channels do that kind of stuff.

A

Yeah that could be good. Do you know of a specific sig? That's doing that for their meetings.

C

um I know that I know channels where it's being done so I could, I could reach out and ask like the openshift users channel. They have something set up um and I I don't think cluster api does, but maybe cluster life cycle has something uh sort of. I can do a little investigation, though.

A

A

All right, let's jump into the agenda uh jared, I'm guessing you added that one all right go ahead.

E

I was just wanted to quickly uh I mentioned so uh we do have a couple of uh I think valuable fixes from andrew coming in on the api server network proxy for people who are interested on the extraction migration, uh and I was going to put poor nick under the under the light and ask where we were with the uh web book stuff.

B

Yeah um I made some progress like two weeks ago um and then I had to kind of uh I was pulled into some some stuff, so um that should wrap up this week-ish uh and I think code freeze is we have like the rest of the month, so I I I should be. I should be able to get an update out, um maybe by next uh extraction, migration meeting.

E

Okay, reach out, if you're having any problems, I think several of us are interested in helping yeah, definitely um yeah I'll, make a commitment for for next um extraction meeting awesome. Thank you nick sorry about interrupting you. There uh andrew glad you brought it up, so we can.

A

uh Go back to the agenda.

A

uh Anything else on that before we go to jared's topic.

A

Okay jared: you want to go ahead with the little migration ga.

H

Yeah sure so little migration is planned to graduate to ta this release. However, in order for that to graduate we have, we need a automated task and b we need to. We need to have like all cloud providers that require migration to affiliate testers feature. So currently we have the automated test screen you can.

H

You can check the dashboard here, however, we are, we have only two providers tested this feature which is dcp and aws, so I'm still waiting for input from azure openstack and with that um so this feature I haven't, got any behavior changes in software. I think so like it should. It should work like if you, if you really try it, you shouldn't work at your first attempt, so I'm um I'm here to like call for action, because the great uh code phrase is coming in my boss and we don't.

F

Really want to miss.

H

The other night, oh just.

F

Oh go ahead, joe um jared, given what you've done in cops? Is there a? Is there a low effort path some people could take? Could they leverage some of your working cops to do this more easily.

H

Yeah, like the the test, the test is neutral about cloud providers so because we just need to, we just need to plug in the environment variable like set the cloud provider and instead of zooms, because it's a tuner. It is like a original cluster or like cluster across multiple available availability zones. So we just need to manually specify that otherwise it should be just automatic. It should just work for any cloud provider.

F

Awesome, I I mean, I think, that I think that lowers the bar to anybody that was trying to, or had intended to do this, but hadn't found a path. I think there's now a low effort path. I mean it's still. It's still they've got to make sure that they're that the k ops configuration maps what they care about, but um at least there's there is an option.

E

To ask jared and it sort of relates um to ease of use, but I thought we actually changed the defaulting behavior between alpha and beta to make it easier to use sorry, yeah, okay,.

F

um Do we have, I seem to remember, we had volunteers in a previous pass? Do we have a list of those this.

H

Those volunteers were I yeah. I got. I got an input from uh like c cloud. I think it's it's a calves maintainer like running in a migration successfully with aws because it is kind of the primary you know cloud fighter within chaos which is like the most tested and most so I got the signal very quickly.

H

F

Elmika, do you remember who, on on openshift, was going to look at this.

C

I mean it's essentially the team that I'm on so I'm trying to organize us to do it now. um But I see that bridget you know like. I would love to collaborate with bridget, because I you know, I don't want to try and like take this off microsoft's plate, um but like so, I will reach out to her and just see if like- and you know, give them the opportunity to kind of run this and explain what we've been doing.

C

um But if not like I'll I'll, probably end up doing this myself and just um you know just set up a cluster inside of azure and see if I can run the migration test that jarrod linked earlier great.

F

Are there any other people we should nudge.

C

And if you want an openstack test, I can try and poke some of the openstack people. um You know who work on the cloud provider internally or red hat, but again like this, you know I don't. I don't.

F

Know if they'll, if they'd, want to.

C

Accept this or not, okay,.

F

Yeah, I can't remember yeah, if you don't mind just pinging them, I mean best effort, okay, yeah. I will awesome. Thank you.

F

Do we still have a gap on vsphere.

G

Yes, uh I can't work on it. uh I might need to ping chocolate, uh definitely primarily for uh tasks that I need to do to verify this.

H

Yeah, that's definitely definitely.

E

H

E

So a quick question and- and if you can't answer- that's fine lebron, but uh how many costs do you know how many customers there are left using kcm only on vmware.

D

Yeah, sorry, I don't have that number okay, but but you think there are some yeah they're still because we see issues coming up.

E

Okay, no, I just I mean I there's no point in doing the work. If there's no one who's going to do that migration, but yeah as long as there's someone to do the migration, then it would be good to make sure this works for you guys.

A

And just um just filling in the context in case it's missing like the leader. Migration is only relevant uh if you're migrating a cluster using the entry b sphere cloud provider to the external v3 cloud provider. So like any clusters that are already using the external one, you don't need to worry about that migration, but yeah if there are customers or users or even like open source users that are on the old visual cloud provider like you'd, have to deal with that thanks.

C

I got this question about how many customers are still running kcm only like what's kind of what's the context here on that one. Oh sure,.

E

So so this specific migrator is uh a a mechanism on it to migrate, an h, a cluster running kcm. Only to being an h, a cluster running, kcm, plus ccm.

E

It is, it is solving a very particular problem, yeah totally.

A

Up uh on this topic,.

A

Okay, uh so this was just the a I had it from uh triage earlier about our docs, and then I see nick added the annual report. Pr.

B

Yeah, so just a rough draft as part of it, I added a contributing.md which I heavily borrowed from another six. So just take a look at it see if it makes sense, um I think it's all pretty. At least the contributing md is all pretty pretty standard stuff and then for the annual report. You know I did um try to get the data as accurate as possible. um I don't think I quite got everything yet so might be another revision of this coming but um yeah. If you have time just take a look at it.

A

Yeah that sounds good uh just to make sure folks look at it. I will assign myself and walter. Can I send this to you as well? Of course, anyone else wanna get assigned right now.

B

You could do the try, the sig cloud provider, pr reviews um user. If you want do, we use that at all. Does that even like map to anything relevant anymore.

A

I don't think so all right or I'm sure it's still mapped to it probably has like old users on it and whatever got it. Okay, yeah. We should probably.

B

Fix that it would be useful, yeah agreed.

A

E

So one thing I noticed andrew um sorry going back to the very beginning of the meeting, but our cloud provider repo actually has a bunch of issues in it.

A

That's a good point.

E

So I went ahead and put a link in the chat. If you want to or you can just do it that way,.

A

Easier for me, okay, yeah.

A

A

I think uh so like this one, we've kind of this one has a long conversation with the pr that's already in flight and being reviewed.

A

But uh sorry, let me backtrack. So you're kind of I think maybe like process, is what we're trying to figure out um yep yeah. So we use these issues at some point as like a roadmap planner and like we redirected actual user issues to kk because, like the staging repos there and that's where we actually commit changes.

E

Which is reasonable. I had just noticed that, like that january, 19th was relatively recent, as were I mean, even even some of the ones from the end of last year. So while we don't get a lot in there, I I was a little worried that we might have things show up in here and disappear if we're not paying attention.

A

Yeah, I agree: okay, maybe as a starting point. We should enable the the triage label bot to add labels here and we can just go through it as part of our uh the beginning of triage every meeting and I think most meetings we won't even have new ones, but at least it's on our radar.

A

A

uh Enable triage labels for.

A

Do you want to go through some of these? uh This one, like I mentioned, is like, I think, we're all in agreement on the fix, but uh I think we just need to review the fix. Actually, it might be good just to mention, like I think, the more eyes on this the better, uh because it does have a ripple like if we, if there are issues with this patch, the consequences are kind of bad like we could end up causing some controllers to misbehave.

A

So I think, having as many eyes on this change would be good.

E

Do we have um like the sig networking, at least a sig networking lead on this, given that it has to do with networking.

A

We don't so jordan and I have reviewed it, but yeah, no one on sick network, so it would be good to maybe get one person from network to review this.

E

Yeah so maybe like either tim hawkin or bowie.

A

Yeah, that sounds good.

C

A

C

That the red hatters, who are like raising that issue and have put the pr up those are some of our openstack folks they're, like very keen to get this reviewed.

A

C

A

um Like definitely for 124 like we should merge this um I've been stalling on it to be honest, a little bit because I'm like I'm kind of uh steeping on it to make sure it's the right patch uh but yeah, I think more eyes is better and uh I will actually actually I'll try to I'll try to get it approved this week, because I know it's.

E

It's important okay! Well, then, if we want it approved, I think it. The the most important thing is just getting someone from sig networking on it. Asap yep, yep agreed.

A

um Yeah thanks andrew walter. Let me just.

A

I'm also gonna get awful to see antonio, because I know he's kind of familiar with the whole modifi stuff.

A

And I will also see dan with him because I know he's been.

G

D

A

Right, um hopefully, little bouncer target services are not deterministic when starting process.

A

This is a fun one.

A

Yeah this is one of those like it is a problem, but it would just reconcile eventually, so it's not urgent, um but maybe it might be worth investigating. Does someone want to look at this one, I'm going to start triaging these, though that work? Oh, that does work.

A

Okay, uh if someone wants to take that feel free, but I'll leave it open for now, but it seems, seems legit. uh Okay. This is a similar one.

A

Okay, similar issue, I'm going to triage, accept it, but I will leave it unassigned if folks want to look at them or assign themselves, I'm not going to assign anyone today, just because.

A

Doesn't seem like anyone's interested, so I don't wanna wanna go there. um This one seems.

A

Controller manager.gov version is shown as the format string. That seems like a bug.

A

Oh, that's because you need to build it with the version, build tag right.

A

A

Yeah, I'm pretty sure like this is not a bug just that, if you don't compile with the virgin build tag, then you don't get like this is like the version library in clanko or something like that, uh but I will accept it and no I'll I'll follow.

B

Up and close it afterwards, actually, because why would the format string show up on on like the default value.

B

A

It's just based on what this: let's, let's see.

E

No, I I think the point andrew is making is that the the format actually has to be evaluated at builds time and passed through.

A

Yeah, I think this, the the format we see is this is the zero value of that gap.

E

Right I mean that format is trying to. I think if I, if it's what I normally see with this stuff, that format is trying to get the get tag at the time of build.

B

I see yeah okay.

A

uh Should we go through more of these.

E

We also like, when we have this many we we don't necessarily need to go through all of them today we can sort of do five of them at a time until we've we've knocked this down.

A

Yeah, that sounds like a plan, so I will look into the adding the triage label so at least we for next time we can keep track of what we looked through and what we did. So that sounds good all right and that's it for the agenda. uh Anything else folks wanted to discuss.

I

I just popped in here like a minute ago andrew, but uh I went to look at the new time and at the top of the agenda note stock. It still has the old one so.

A

I

And then I don't know if this is being published in the kubernetes community calendar, but I went to look there too for the new time and didn't see it. There.

A

Okay, I will I will play that update.

C

Thanks walter, I knew that pm was going to get someone.

A

Well, sorry, what was that.

C

Nothing there was, it said, like 12 p.m, slash you know, 9 p.m, and you know, of course I was driving my ocd nuts and I was just about to touch it. Then I saw walter change it.

C

Okay, cool thanks.

A

All right we'll end the meeting here then and see y'all in two weeks. Thanks later everybody bye.