Kubernetes SIG Cluster Lifecycle, 5 Nov 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 20191105 sig cluster lifecycle

Description

Kubernetes SIG cluster lifecycle meeting for 11/5/2019

A

Oh today is November 5th 2019. This.

B

A

Standard said, closed, July 6. ah um Let's take a look to see what we have in the agenda. We only really have any standard status updates. Are there other things that folks wanted to discuss at all.

A

The only topic I can think of that I wanted to bring up was that for me, too, I had done some good work in sort of taking some of the things we had laid out here in the--the for the face-to-face that we're going to be having at Keuka. My plan is to go through this in more detail.

A

This coming just not this week, but next week, so I might ask the sub-project owners to step up and to maybe have a separate call for next week for us to go through the slides or to they can verify asynchronously.

A

That would be helpful. So as a reminder, we're gonna try and get the slides done by next week. So that way we don't have to scramble for a good con I.

C

Was just gonna say, I volunteered to help with that, but I II I have not coordinated any efforts. No do I know of any I. Don't any efforts that have yet been coordinate. Otherwise, I don't is there something I can help with or should I coordinate should I create some slides, I.

A

Think slide updates are the best way to do this. So if you go over here and you're here,.

A

Hopefully, folks can.

C

See that I was not at work. Yes, we can. I was not aware of this. This is the this is one that you've created as a start-up great. It's.

A

The start and I believe I shared it with this group in a private chat. Whoever's, quite just be speaking. So let me put this link inside of the meeting notes.

A

And what I think I'll do is if it's, okay with some project owners who are listed in this section here is I, might set up a separate call for next week, sometime probably mid next week for us to just make sure that the slides are in order, maybe 30 minutes to make sure that to serve as a forcing function check. You know, question oh.

D

Yes, sorry, a little confused, there's this these, these slides that you're making this is for the project roadmap. Is that different than the face to face that you briefly mentioned at the very beginning? That is.

A

The face to face pretty much.

D

C

Don't actually technically.

A

Have a face to face is anymore, so we have to make an excuse by saying, like it's a meeting of some kind, so I think project roadmap was a good way for us to get new contributors involved, as well as talk about long-standing issues and questions that we've had got it. Okay, thanks yeah.

A

Are there any questions there.

A

Good once twice three times all right, we can go into the standard sub project updates to get more feedback from the community. Look where you want to start us off, yeah.

B

I read some notes for qadian, only three quick items, so we try to resolve some end to end test failures right now, and these are not caused by Q medium. If you have been following this they're caused by XI GHC de kado also changes in kind, so we are walking mean for Brits are working on that and also we are coordinating with the release team to fix this.

B

Another PSA here is that we start the discussion of how to consume the cube, a DM, API config more easily by tools, like course, API or in general, by high-level tools. So this is a working group, the component starboard standard issue. We already discussed the problem of how to consume the the cooperate type properly from a high level to like a beta. So please comment on the issue. If you have any ideas, another one is that we in verse 17 we added support for docker 1903.

B

This is only on the validated limo, so you can still use even newer versions of docker. We are also updating the sed server, 23.4 and potentially also updating coordinates, but this is still to be decided.

A

That's all what see I know this is a short cycle. What's the current date for code, freeze, I,.

B

Need to check I think it's the level for November. Let me check. Oh.

A

It's kind of soon is that, like the dates they they keep on changing the names for things, so this is actual nothing that bug fixes occur after victories right. Yes,.

E

It's cold, freeze, I, just checked it's on the 14th of November. It's.

A

Really soon to do a core DNS update in that short period of time is a little sketch.

B

Yes, given it's not decided yet we might drop this I'm more concerned about our work in progress, features like windows, getting windows to beat them. Also we have a machine, readable output, I, don't know that I need to talk with whoever is assigned to this item.

E

A

Have an updated test job specifically for comedian in Windows, no.

B

I spoke with someone from people, though, who is working with that, but it's I believe there are some problems and I'm skeptical that we are going to get these to be the oldest icon.

A

If you can reach out in the SiC coaster life cycle channel to to figure out who the point of contact is from the window side I'll be happy to help wrinkle there. Okay.

A

Are there any other questions for the Lumiere.

A

Then once twice three times Daniel, it's a DAT em.

F

Yeah, we have a dentist running a problem, so that's that's gonna make everyone's life easier. Finally, it! Finally, it happened. It took a little while figuring out things in and prowl, but that's there and we're actually running a TDM in docker with system D. So moving to multiple number of members will happen soon. So yeah, that's pretty cool.

A

Is there it now it sounds like you're getting towards the container is space? Is there any plans to do in, like maybe 118 or 119, to start to do initial POC of integration with Cuvier middle.

F

Yeah I think yeah. There's there is an issue open for supporting a TD in static pods. That is that, let me double-check, so we also added a roadmap and I call it that particular.

F

If, if it's not then I think that would be there's definitely interest in that, so that yeah, that should be it's not on the near-term I'll yeah it is. It is on the roadmap, so yeah, potentially the next, the next release any.

A

Other questions for a Daniel at all.

A

Good, what's twice.

A

Thomas mini cube sure.

D

We did a a 1:5 release the other week. We actually had to follow it up with some point releases to address some issues and shortcomings in her for newly introduced bugs and so we're planning to do. 1/6 for Q, Khan and hoping I guess. One six will mostly be I, guess performance related.

D

We are planning on demoing at cubic on during our deep dive talk a couple of new features: VM free deployments, multi node and a prototype Mac OS menubar UI we're also testing a PR bot to report performance regressions. One of the issues that we've had is that it's hard to understand the performance implications of incoming PRS.

D

Sometimes people claim that there's a massive speed up and there's actually not it's actually in the other direction, so we're right now it's on a very coarse, it's very coarse grain, and that it's just reporting that, like start takes longer, but where we would like to get is implementing a PR bot that actually reports that these functions take longer than in a previous than in master. So the code for this is in the Dominican repo. If anybody is interested.

A

Just real curiosity, these just standard sort of behavioral driven tests like sort of intense doubt, tests, yeah.

D

So it's right now we're basically we're mostly concerned with cluster startup.

D

So right now it's just measuring that, but we're so it's measuring just basically the average across three runs for master versus the PR and we're looking for a signal that there's a noticeable difference or not so where we plan to add like a list of other actions to perform, but so far we're just focused on start. If.

A

You found any performance implications in qadian that you want to optimize feel free to submit PRS there too, as well. Yes, we we.

D

Expect, maybe not on the we don't know yet on the start, latency, but one of the places that we expect to be launching PR so that the first step is measuring. Latency the once we get latency down we're going to start measuring overhead of some of the kubernetes services and try to dial down the overhead for local kubernetes type solutions, and those PRS will definitely go upstream. So but.

G

If we do see something.

D

For cube ATM, we will definitely be putting it upstream. We don't, we don't want it. We don't want to be a special snowflake here. Can.

A

You elaborate a little more and overhead besides lecture latency. Are you talking about overall size and space and efficiency as well, so.

D

I'm mostly focused on CPU overhead I would really like to get a running, kubernetes cluster down, to say five to ten percent CPU overhead on a MacBook Pro. We are a ways away from there right now. We have not started doing measurements in this space, but that's part of what the PR bot is designed to design to function. For so there's.

A

Been a lot of ask the reason I ask is, in the history of time we've seen a lot of people wanting to run kubernetes across sort of edge edge devices, all right, small, smaller and smaller footprints I call it the grenades, toaster yeah.

B

A

A joke, but it is a valid use case right. Yeah. A lot of these folks are running on really small armed environments and optimizing efficient resource efficiency and utilization is super super useful. Yes,.

D

Yeah we've we've done some discussions about whether or not we need to start looking at memory overhead for people who are doing unique configurations, but so far it has not been a focus for us.

A

Okay, if you have a an issue, an umbrella issue or a tracker, they're I, think they'll get. Those who are optimizers might be very interested in that yeah I will I will find the link to it. You know I could medium publicize a little bit more on the mailing list, because this is a state space where we almost get pinged on a monthly basis for people who have like something. Some weird thing has been broken inside.

A

You know the resource efficiency is not that great or some weird thing is broken inside API server, initialization on some really resource constrained environment.

D

Sounds good I will I'll make sure to do that.

A

Any other questions for Thomas.

A

And what's nice few tips all right next up.

H

Jason the street yeah yeah, so I have a list of the most recent releases that we've had for various cluster API projects feel free to take a look and they're all V 1, alpha 2 oriented releases and for the most part, they're addressing bug, fixes I. Think cap G, you may be kept. Thee may have some features in there, though I'm not familiar enough with them to to stay other than that. We've been making a lot of progress on the proposals for v1, alpha 3 I.

H

Think the only one that's still seeing relatively heavy comments at this point is the control plane cap I'm waiting for the last comments to come in from somebody who's actively reviewing it right now to do a final pass through, so that we can hopefully get that get that proposal merged, otherwise I other proposals to probably start landing shortly.

H

Outside of that, we've been doing a lot of work to improve the IDI testing for the AWS provider in particular, and thanks to dims, there's been a lot of work going on for testing conformance of kubernetes, build artifacts against both the GCP provider in the uws provider, so we're starting to see actual upstream conformance signal running against cluster api. So hopefully we can use that to continue chipping away at the removal of the slash cluster directory, and the only other thing I had was a topic mostly to discuss with Justin.

H

We've had to tweak some some fixes with the AWS janitor and every time some changes are made. We keep getting asked by the janitor code. Doesn't live under AHS goes even we're deploying it. You know as part of Bosco's, and I have to go through the history of how the project originally was a standalone project that was running his periodic job.

H

We kind of jumped in and co-opted it to also run as part of Bosco's, but I wanted to check with you to see if there were any objections to keeping the dual usage but moving where the code lives to the vascular's directory versus the maintenance directory and.

C

Just confer it's the same repo, it's just a different directory yeah exactly it sounds fine if it makes your life easier. It's fine with me cool.

H

All right and I'll follow up with that. I just wanted to make sure and get an okay from you before we went ahead and just moved things and I.

B

H

You have a question there about whether it's using human ium master or not Tim I am NOT well yeah, so it's using the CI generated artifacts for kubernetes, so it is using cubed en built for master okay.

A

Yeah I'm just making sure that, like if we're, testing or testing tip of cluster API and tip of kaykai so like that, the C icing will be weird for conformance if you're testing it.

H

Yeah, so we're making sure in the tests that we're either testing tip of cluster API or tip of kubernetes, but not both. At the same time,.

A

May in the comment is still that no one besides the cluster, a people would make any sense of this, except for this one denoted way of denoting the version inside the release name. That is actually useful. So.

H

That's actually probably misleading Tim, because that alpha 2 is not actually referencing. V 1, alpha 2, it is. It is an alpha 2 release of version, 0.2, zero of the GCP provider and I. Don't have history for why those versions are numbered. That way.

A

Yeah I'm gonna push on the sub-project to try and address this going forwards if we need to have a flag day than we have a flag day, at least for viewing out for three, is this so confusing? No, what no one will understand this.

A

Are there any other questions.

A

What's twice three.

C

Times, Justin, yes in cops land I'd say we didn't have a terribly productive two weeks through some mistakes, I made and some other things it happened. We had our et tests or our own. You do tests fail, and so we basically spent some time fixing that, but the good news is we are now much. We have better ete testing now that is like back to running the latest and greatest everything I believe and is more consistent across cited us in GCE and skips gingo tests.

C

So if people want to know how to skip that lovely script, they can peek it what cups it up to.

A

Hear the questions for Justin.

A

G

Builder yeah, so I've been working quite extremely on CLI to improve the user experience of using the image, folder and I'll be doing a demo of that in the meeting tomorrow for image folder.

A

Do you have Eckles to push to the YouTube channel bush.

G

I, don't I think jason has been hosting the meetings till now, but yeah.

A

So so long as the meetings get recorded and pushes yeah, we can probably sync offline. If you you know, if Jason goes on a vacation or something that way, you can have permissions to push.

G

And then off to the CLI I want to use that as kind of a base for building the automation and abuse testing the images so I think I'm gonna get quite familiar with a prowl and probably the Google infra dig Lee, so I could definitely use some help on that side. So.

A

The real question I'm going to have now is is when do we form Voltron, but with less dead cats that had the full image builder to kubernetes artifacts to cluster API, all turnkey.

G

What what's your question.

A

There's a question of when but I think the answer to that is as soon as people can get yeah.

G

So I have a lot of the pieces in lots of different places.

G

One of the things that I wanted to start on was building the images and testing them locally using kvn, so I just need to work out how to how to get Google projects the GCP projects and create the correct images in a proud job and and kick those off there's actually very little documentation on that. So start bugging people, probably from next week, I.

C

A

I don't know if Tim is on the call or not I would see. We should sync with Tim's on some of his intent testing work cuz the end state, the ideal end state is to have like you know, the tip tip to tip to tip where he set signal there. So like you'd, be able to the tip of image builder will be building latest images that would be then consumed. You know that actually bundled the artifacts, including cube idiom, as well as all the community's artifacts and then being deployed via cluster API.

G

A

G

Get this just another one yeah.

A

So I think this is a good road map thing to discuss on our on a road map here and talk about it. So you're gonna be at our you could con North, America mush.

G

I'm finalizing that now, but hopefully I'll be there. Yes, okay,.

A

Alright, are there any other status updates or questions comments? Complaints concerns I, saw this comment here, which I think the old invite was languishing. I just I mean update well mo. She was talking earlier. So that's thank you. So much are there any other questions comments. Complaints concerns. Otherwise we get like 34 minutes back I had a question for the image builder is.

D

There documentation on how to get started with experimenting with it, because I'd be curious to look into using image builder to say, build mini, cube images.

G

You saw as part of the CLI I'm trying to eliminate as much of the the steps as as possible, so really be downloading a binary and running it with all the conflict felt yeah, but as I'll produce those docks as video along okay, hopefully the shorter rather than longer, because there's fewer steps required.

D

Another question I hear is: how long does it take to build images right now, with image builder.

G

It's like depends on which version and iteration yeah qmu, based one that can build images in about six or seven minutes. Okay,.

D

That's a asking because right now mini cube images because it's basically building everything from source code. It is it takes about 50 minutes and that's a painful iteration cycle for us, so I'm interested in looking in future directions on for this.

G

Make sure I will.

D

I will pay close attention thanks.

A

Cool any other questions.

A

Going once twice three times: alright, we have a short run this week, I will send out a meeting. Invite to the sub-project leads for next week, some time for us to walk through the slides for kook. um My goal is to have a more cohesive narrative.

A

Like you know, we we have from have this picture inside our slides, which is basically how some projects put together, but I want to make sure that, like the roadmaps and the idea is kind of jive together too, as well, so look forward to an embed coming soon take care of buddy. Thank you all right.