Kubernetes SIG Network, 6 Sep 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Network meeting 2018-09-06

Description

SIG Network meeting from Sept 06, 2018

A

Okay, cool we're recording this is the kubernetes cig network meeting for September 6th 2018 and we're gonna stick to release stuff and only thing on our agenda. Right now is selling tests, a few of which are marked as release blocking, so they can be good to make sure we're on top of those at least somebody's looking into them word up, or do you have a list yeah? Let me let me try to share my screen.

A

Okay, you know I'll see that see something see something cool, so I've been looking at. Okay, I can't change to that tabs. Whatever you, okay, this! No, not this guy. This guy release.

B

Make your text a little bit bigger case. Oh.

A

Yeah, oh boy, a new keyboard.

A

There we go is that readable, yep.

B

That's better nice.

A

I'll get that out of the way um yeah so I was looking at, was listed as the cig release master blocking set of tests.

A

We look for city network, there's kind of two sections here. There's this DNS failure.

A

And we've got these guys down here, it's all seven GCD, let it bounce and.

C

Tests, someone updated about that that the the feature neg whatever that is got removed and but they forgot to disable the tests. So I think someone was working on that now. Yes,.

D

So yes, I'm watching it.

A

So was that feature removed correctly, or is that something that.

E

D

It's still disabled, so I can send a PR to temporarily disable the tests.

A

A

And then I guess the other thing is the DNS test, which I think someone said they're on yeah.

D

I'm working on that I found the reason why it's failing the coriolis configuration is incorrect, so I'll change that and push up here for it.

A

Cool and that's this guy right, yeah yeah.

D

A

Guess just follow up on the PR to say that you've you've got a fix for that.

A

All right so I think those are the only release blocking tests. There were a couple, other failures, I spotted and maybe they should be released. Blocking at this point looks like the IPPS tests have several things failing.

A

A

Yeah there's anybody on the IPS stuff present.

A

A

Okay, I will try to chase that up and I guess. There's a question of like do. We have the right things set as release blocking.

A

Which I'm not try to answer.

F

Yeah I mean I, guess ipbs is GA now so.

E

Yeah, that's what I would have thought.

E

A

Is just the ass meeting.

A

B

Yeah they're so broken and that's a different problem than it was before. I had a general question around what is the deal with cube ATM? It seems like they always break the upstream repo for communities anywhere which all those tests depend on is like deprecated and only used for these things. What does anybody know what the expected future of those tests and that test infrastructure? Is it's a great.

F

Question I do not know the answer to that.

B

It just seems like every single time that I've taken a look at those cube ATM tests. It has never been networking related. It's always been test infrastructure, so I don't know who owns that. But you know at this point I'm wondering if you know those specific tests that are failing I mean, should we disable those who do we talk to about that yeah.

A

I think we discussed this like.

B

A

Months ago and I can't remember what the outcome was, but I remember there not being a clear ownership of kubernetes anywhere or those tests, and it probably just kind of fizzled.

F

Am I is there any? Is there an issue on this that I'm not tracking an issue on what specifically I'm specifically like? Why do these tests keep failing and who should own it? No.

B

There's no issue: I did file a specific one for this test flake, but there's also an indicate the status of this repo issue, 556 incriminate anywhere. That kind of tries to clarify or something like that, what it's used for- and it says.

B

Well, anyway, I can I'll just link to that in the notes that people can go off and read that issue.

B

F

I just paste it into the doc I'm gonna take a look at that and see if I can stir something up: okay, I.

B

Mean it says once cube: a DM has a different solution for edy tests. They plan to move to that and get rid of companies anywhere, so maybe about 113. It gets fixed, maybe.

F

I would like to go and understand what the timeline is. Okay,.

A

Okay, so it looks like a couple: more items have popped up on the end of our list so, and you want to talk about this services, controller ownership, yeah.

G

C

Andrew one.

G

Of the co-chairs for I left a message in the mailing lists like two weeks ago, pretty much talked about some of the changes that were trying to make with services and routes controller so based on the current owners files for those controllers, it seems like it's I'm, not sure if it's officially owned by sick network, but it seems like majority of the owners in that file are part of cig Network and previously we've been I, guess I!

G

Guess we, some of our work, has been blocked or been pending for a long time, because we have to get you from people file and I'm wondering like based on the fact that services and roc controllers it is network related, but also like it's mostly just interfacing, with with cloud EP eyes and not not really like. It is a bit of both right I'm wondering if it makes sense to share ownership with the cloud provider on those controllers I'm.

G

Given that we're going to be doing some work around those in the next couple of releases and sometimes they're like chat with sic network and see if that makes sense. So this.

F

Is Tim I feel like the routes? Controller is completely cloud specific and we probably should never tried engine Arif I it. That was probably a mistake in general, the services controller, though that's there I'm generally put back I, think I'm just about to make a mistake. Pete, it's the end, points controller that generates end points from services. The services controller is what kicks off a little balancers in store calls ensure load. Balancer yeah is there? Is it doing anything else? No.

G

That's as far as I know, it just adds nodes to the bouncer pools and then just configure two little bouncers and that's it old service controller like we should maybe maybe name it to like load, balancer controller or something else. Yeah.

F

Exactly my thinking, I think that is solidly the domain of cloud provider I'm going to disagree.

E

Yeah sorry I, when I read your email, I took it to being moving ownership, which was like a little. But if it's sharing yeah.

G

Honestly, I think it'd be good to just have like one or two developers and say Club a letter to being the one it's all just so like we can kind of. You know, look good to me whatever PR is that you might want to do with those those two I'm.

F

Yeah I'm not saying.

G

That we need to remove it from sig network ownership, because I'm sure you guys will run into issues as well. But okay.

E

What is the proposed work because most the it's supposedly written in a cloud agnostic way, yeah.

G

So one of the one of the big ones is renaming little balancers, oh okay, yeah and the other one is like we're. Finding a lot of writers are running into like API limits with Ross controller I mean we think it's because it's overly aggressive with the way this updates routes, and so we might want to do some caching or something in there or sort of optimization in the future. I think.

F

All those things sound, reasonable, yeah cool work. Thank you.

A

All right sounds good and then Dan you wanted to talk about some 1:12 stuff, so Network policy in particular. It looks like yeah.

C

So I was checking the spreadsheet to the Future spreadsheet, to make sure everything was okay and discovered that we accidentally never actually proposed IP block for one 12ga. We had all of our conversation on the egress feature, P R and then never updated. The IP block one so I tried poking the release. People I have visited the the URL for this PR and the chat I tried poking the release. People on that PR, saying hey: this is actually already done.

C

Can we get it at it and nobody responded and I'm noticing now in an earlier comment that there's a link to a procedure for getting an exception? So maybe we just have to do that, but I didn't know if Tim with, if maybe you wanted to just before I.

F

Think the process for exception is the right thing to do with the PR numbers. Oh, it's.

C

Kubernetes features 367.

F

Okay, I think the right thing to do is to go through the normal exceptions procedure and just say like this was a snafu on our part of not know. There's a 50/50 chance, maybe 6040.

C

All right, that's all I had be the the egress or yeah. The eager spawn looks like everything is on track for it. So.

A

A

That's the end of what's listed on the agenda, any other release stuff. That would be good to chat about here. Yeah.

B

I had a quick question about like bug, triage and then adding labels on issues I forget when it happened, but a while back like a couple of months ago, some of the label things stopped working, I'm, just wondering if there's a set of things that you know, we still know work for that stuff. Like can we still tag things as area cube proxy? Is that not something do anymore?

B

I think Signet work still works, but I think.

F

The cigs, the cigs, are definitely supported and I forget. If area is important, I think area, there was an area and there was component and they were, one of them is supposed to be eol I think it was area well.

B

I used to be able to have the ability to add and remove labels I assume that got locked down so I can't even see what labels there are anymore, but you know I, don't ya, know I can't can't see that anymore. Oh look at that. Okay I know, maybe I abused it and they took me off as a joke.

F

Strong personalities in this group.

B

Yeah I was just wondering what the current best practices were around. You know like triage for delayed labeling for triage. It.

F

Seems reasonable that the sig chairs have access to those things. I will poke it and see what I can wear. Okay.

A

All right, if that's it, then I think that's it all.

F

Right cap, your release- everybody, hopefully in the next three weeks, goes successfully in the major issues and we'll have really interesting meetings in the next couple of sessions.

A

Indeed, thanks everyone.