Kubernetes Cloud Provider Special Interest Group, 30 Mar 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Cloud Provider 2022-03-30

Description

Agenda: https://docs.google.com/document/d/1OZE-ub-v6B8y-GuaWejL-vU_f9jsjBbrim4LtTfxssw/

v1.24 status:
Remaining PRs that are approved and needs to merge?
https://github.com/kubernetes/kubernetes/pull/107631
https://github.com/kubernetes/kubernetes/pull/109072

PRs that didn’t make it and might want exceptions?
https://github.com/kubernetes/kubernetes/pull/108953
https://github.com/kubernetes/kubernetes/pull/108838

PRs that landed that we want to call out:
https://github.com/kubernetes/kubernetes/pull/107750
kubelet cred provider to Beta https://github.com/kubernetes/kubernetes/pull/108847

Discuss https://github.com/kubernetes/kubernetes/pull/108914
should the cloud provider interface support allow-listing LB classes?

A

This meeting is being.

B

All right, um hey folks today, is uh march 30th 2022. This is the bi-weekly cloud provider, sig meeting um yeah, let's just get started and we'll go um quickly into the I'm actually gonna, I'm actually thinking of skipping um triage today, just because we just came out of 124 code freeze and I'm sure we should optimize our headspace. I think about that. So I'm going to skip triage um but yeah. Let's go straight into this project update quickly.

B

um I see. Aws has the first one about cutting the 124 release.

C

Yeah um cutting uh cutting that release, uh we also merged a um another controller. It's a tagging controller that helps with uh cost allocation in aws, uh so the it'll add a custom tag to to nodes. Basically,.

C

B

um Okay, uh azure update on 123.8.

D

Yep we we got that uh you know just some bug fixes out um and, of course, anticipating 1.24. uh I imagine we'll get the release over the alpha for that ourselves too soon, and uh I just wanted to tell nick that tagging controller, which uh helps with cost allocation is super cool. So you should be proud of that. It's very cool thanks.

C

B

Agreed: uh okay, I see no updates for ibm.

B

uh I'm gonna skip extraction migration because I think we'll talk a lot about those items on the 124 status in the agenda. um So yeah I'll go straight to the agenda um so yeah.

B

I wanted to just uh gather a quick list of any pr's that are in the merge queue and in the milestone that is just waiting for the bot to merge, as well as prs that we might need to file an exception for and also just calling up prs that we did merge um that we want to just uh call out and celebrate um so we'll start with the first list here.

B

Okay yeah, so I added this one. This is the the service controller bug um where the retried services are getting um cached and not updated correctly. So I actually approve this. um I put it in the milestone, so I think it's good to merge. I don't think you need to put an exception because it was merged yesterday or it was approved yesterday, but I it would be good to get like a second opinion on this.

B

So if someone on industry wants to review this pr and just double check reasonable um that'd be great, um because the logic here is a bit um like the bug is very subtle, so I do worry that or we could be introducing like some other um books, but yeah it'd be good to get a second eye on this one.

B

Was that it for vrs that are already approved? That needs to merge.

A

um We had a couple um for the leader, migration that are they're all approved and gonna go in now, but the umbrella issue got kicked out of the milestone, so I just opened a zero day exception for that just to get it back in I'm not saying I.

B

Mean okay, if the pr is a proof that in the milestone at this point, it should be fine right.

A

uh Well, the umbrella issue got kicked out.

A

So I figured it'd be better just to be transparent about it, but it's a zero day exception with no no future changes coming so, hopefully they'll just approve it. We'll put the umbrella back, it'll be done.

B

Yeah, I agree. I think it should be fine. I had a comment here, but I don't know if it could probably be a follow-up.

A

Yeah, we'll probably do that as a pull-up. I didn't see that okay, thanks for helping.

B

Didn't make it and we want exceptions for.

A

So these two are related right.

C

Yeah, so the the first one, I think, is reasonable for an exception. The second one I just wanted to bring up, but I I don't think we should go for an exception unless anyone else thinks we should, but the first one is the just removing the deprecated insecure serving we talked about this. I think we agreed that we should do it.

C

um So I I think it's relatively uh straightforward and um I think the only of caveat is um you know, removing flags that that people might be using, but the same thing happened in cube, controller manager and uh other all the other components. So I don't think it's unexpected, necessarily um so yeah. I think that one's reasonable to merge. If others agree, um we would just need to. I guess, add it to the milestone and ask for an exception.

E

I I believe the deprecation notice has been out for four years now, so I I think on that front, where we're more than adequately covered- and this is a security issue uh and in today's time time frames, I think it would make sense to just do this.

D

Yeah yeah, if you secure something and that breaks something, then the something was something people shouldn't be doing right.

B

So I I agree with that parents, but I also think that the release team uh like they're, not they, might not consider like a removal of an already dedicated thing for four years like release blocking like I, I don't. I don't object to finding the exception, but I think the reason back so something to keep in mind.

A

I think walter's argument about security is a good thing to throw in there to that.

B

Yep sounds good, uh so this one.

C

And yeah- and this one um I don't think I should file an exception for this one. I think I didn't get. We didn't get far enough with reviews. uh I've too many changes that you know are in flight. um So I don't, I don't think the release team is gonna. Go for. um You know the the impact of the schedule that that you know is a good enough reason to make it release blocking.

C

If anyone disagrees with that, um let me know I mean we can try, but I I doubt that it's uh it's gonna fly.

E

It's fair, so I have an odd thought and I'm not sure we can do it, but the the really the biggest reason why I don't like the fact that we didn't get this in, isn't because I necessarily wanted it in this upcoming release.

E

But just because of the number of releases it's going to take to get this to ga right. If we do alpha beta in the a we're now talking about that being pushed out by four releases, which is at the current really schedule well over a year out right.

E

So the one thing that would be interesting- and I don't know if we can do- is seeing if we could skip the alpha and just take it straight to beta, not in the release. This release we're pushing out, but in the following release, so that we get back onto schedule.

D

Okay, if we plan on wait, if we plan on skipping alpha, then wouldn't asking for an exception for alpha for this time, essentially do the same thing.

A

It might be more.

E

Transparency, it would, but the problem is that he to even be able to ask for for a a release, a a an exception to checking it into this release.

E

We'd have to be code complete and to uh nick's point um we're probably still a couple of days out from being code complete, assuming that that nick even has time to work on it right now,.

D

Gotcha that makes sense.

E

Yeah and the other thing, I'm not saying that we'll we'll be allowed to go to straight to beta next time. I just think that we can ask.

C

B

Okay, yeah or like at the very least, the quality of the feature should be beta standard and worst case scenario. You know we go off of one release and we'll be really confident to go beta immediately after.

C

Yeah yeah, I think either way we get it in early in the next cycle.

B

Cool uh any other prs that we want to call out that didn't make it.

B

B

So this yeah this one merged. I know.

B

um Thanks for all the pr reviews and all the iterations, but I'm pretty happy with this pr where it landed and um it looks good to go.

B

um The cubelet credential provider went beta. This landed and there were some other pr stat b2b tests and whatnot.

B

I think that's it is there anything else.

B

Okay, uh all right, then, let's go to the next agenda item. So I added this. Like a week ago, someone was asking for a cloud provider feature to let the provider implementations specify a load balancer class.

B

um Let me sorry, let me go to gif. That shows so they want service controller to accept the list of classes that they use when reconciling services. So the context is that in like several releases ago, we or I had proposed to kept to add um a field called low bouncer class to services.

B

um So it's it's like a lot of other class constructs. You have, except it's specifically for service at load balancers. So if you set the class, um the default behavior for the cloud provider is to then ignore that load balancer and let the next chrono controller reconcile that load bouncer, which is just replacing a lot of like existing functionality that people already have using annotations so just kind of a more like um api driven way to do that or more like native way.

B

To do that, but there's an argument to make that the actual default provider implementation should also be able to watch against the list of known classes and risk like accept, update events for those classes if they opt in, and so basically I think what it would look like is we update the load, balancer interface, um to have like a little balancer class method that returns a list of allowed classes and by default it would return no classes, which is the current view.

B

um So I wanna since that's an api change in the interface. I wanted to bring it up here and see if folks are.

B

A

About that, I think your reasoning in the comment there is really strong like having it as an actual api interface, as opposed to a command line. Flag is definitely more friendly to people who have to redeploy these things.

A

Yep make sense.

B

The caveat, of course, is every time we add a method to the interface that breaks, that um it's gonna break um compiling for all the providers because they have to implement the new method. But I imagine like this one's pretty it's like just one method. I think it's worth doing, but uh I don't know, but we don't decide right now we can let it bake for a little bit and come back to next week.

B

B

All right, um I think, that's it for the agenda. Did anyone else wanted to add anything last minute.

B

All right otherwise give everyone 40 minutes back um yeah if anything comes up for 124, exception, requests or bugs or anything, um please. Let me know I'm happy to help review and whatnot.

A

Cool thanks andrew sounds.

D

Good. Thank you. Andrew cool, bye, bye, folks,.

D