Kubernetes SIG Testing, 6 Mar 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Testing 2018-03-06

Description

https://docs.google.com/document/d/1z8MQpr_jTwhmjLMUaqQyBk1EYG_Y_3D4y4YdMJ7V1Kk/edit

A

Okay, welcome everybody. This is the suggesting meeting on Tuesday March 6th, it's being recorded and will be uploaded to YouTube as soon as we get around to it. Today we have a pretty short agenda, so please.

A

Cola's possible because you might come through there, so the only two things we have an agent today is a question from Tim about the current status of the released blocking jobs and then a question for me about where we're at with the type-check job type check job and what did that cut? Enabled to mungus your foot.

B

Yeah Ben is here Ben: do you have like an updated status of where things are with regards to sequester lifecycle could be DM jobs, yeah.

C

So it looks like in the past, they've actually been using. Three distinct builds for every run, I'm digging through that and trying to sort it out. So hopefully we can hit that and get it down to at least two I'm hoping to have things green again, but I don't really know and I'm, not sure exactly what was failing before these issues.

C

B

Was only a small number of of actual issues and I thought they were flakes from what I saw. I didn't see anything major that caused me consternation, at least at this point we are running separate tests right now and our role and set up to verify it will have data back to compare once what's the test suite even runs so great.

C

I've been manually running it with different parameters and looking through where it gets everything from I'm, not sure how long it's going to take to sort things out. There's a lot of weird things going on with how it competes the GCSE locations well,.

B

The I'm mr. no one really owns it, I think that's kind of the problem. It was created by a couple of folks who don't really own it just kind of had to pack something together, but now it's become pretty important for the release. So until we have a replacement for it, we have to at least get by yeah.

C

I'm trying to at least get it to run cleanly, but there's logic in communities anywhere and cue, medium itself, in Cube tests and in the e2e script, all determining where to get votes from so getting those. So why not? It's fun! Okay!.

B

Well feel free to poke me on the other channel. If you want me to wrangle some more cats together, okay I can help.

A

Okay, cool, it looks like Ryan's, not here, but maybe better Cole. Did we make the Churchill do tech check? It's.

C

Running it reports, it's not blocking it. um I was going to PR that I wanted to check and make sure that just definitely sure that we continue to run across for the older branches. Then we won't have any problems and that everyone's okay with the fact that we won't be running it in batches anymore, um not type check but cross. We don't actually have support for conditionally running things and bachelors other than branches, I believe so right now, we're actually just saying always run cross, which is not great.

C

So if we can use type check, we're gonna turn that off and lean on type check pretty heavily. Instead, I will probably PR enabling that today, but we also need to like wait for lazy, consents, the mail with mailing list, etcetera.

D

Have we run into any cases where the cross build has caught something that type check didn't? Is that a I mean.

C

Not it's not common, it's technically possible because type check doesn't validate the actual, build logic so like there's, the actual cross build involves more than just building to.

D

Cocoa, should we switch the cross, build to a post in it or I mean? Can we still run on just batches or NCI and we will.

C

Run it on presubmit not batch when you change the build strips okay.

A

Okay, that sounds reasonable and some of the API differences I believe, was the reason that writing that ended up vending some of the NGO libraries right to do so. Yeah.

C

There have been some upstream fixes, they already exist, but they're not in a released, go version. Yeah.

A

Cool and I guess the last thing, but I kinda wanted to bring up I. Just everyone saw it. There was an issue open against esta for yesterday, I believe asking if it would be possible to start blacklisting specific statuses on pull requests in order for tide to ignore them kind of a thorny long-simmering discussion. So I don't know if it's I.

D

So the main idea be against this. Is that a tide kind of wants all the statuses to be actually important, I guess and if they're they should be green, otherwise they're, not an important status. That being said, I think that's a little too optimistic and not all statuses are going to follow that across all repos.

D

So tide is just gonna need something to handle that it should I pretty I didn't I'm, probably also.

A

Have a question like to point contributes at this and maybe asked a question like what was the messaging look like when we're like Oh in this repository? You do need everything to be green, but here sometimes I think.

C

It's pretty well understood: it's. We restricted the specific cases like reviewable that the Rico is just ignoring that and that they're using LG TM and approve instead um I wish they'd turn off the status, but I I, don't think we get enforced, that we should discourage it being used for actual tests and we're pretty.

A

Certain that we're not the required statuses that we used to have like, where are we with that right now, I'm proud do we have? Is this a little under that we.

D

Kind of need them to be required to avoid a very unlikely race condition if they're not required and tide is just checking to make sure that everything that exists is green. Then, if the PR just immediately is picked up by tide before anything can report I, don't know about any of those things we'll think everything is green.

D

So if we, if they're required, though github will say that it spent its it's.

A

Waiting for it right, so if, in those cases, could we ask that review will be non required, would that be an appropriate way to handle? Also, do we want logic in tight well,.

D

Making it not required won't tide, we'll still consider it required so title.

A

So I guess the combination of making it not required and then making tide only care about required statuses. Yes, okay,.

A

Well, that's what we have on the agenda. Is there anybody else that wants to bring something else? Oh.

B

Just a PSA that we have testing comments, be tomorrow for folks who are doing conformance work or it's interested in the testing framework stuff.

E

I, just posted on this, like channel I, need just for how I'm running cube test with the local up cluster to run conformance stuff. So it's turned out very well so far. So thanks a lot for everybody here who has helped with the work.

C

Thanks yeah cool I.

A

Said well, everyone gets 20 minutes back. Every Tuesday.