Cloud Foundry CF on K8s Forum, 12 Jul 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Cloud Foundry On Kubernetes Working Group Forum for 12th July 2023

Description

2023-07-12

[DB/GC] Memory limits (and limits in general) for build pods
[GC] Branch protection exception on korifi-ci
Need a PR to https://github.com/cloudfoundry/community/blob/main/org/branchprotection.yml to override the auto-generated default configuration
[GS] Team reorg
[RI] (Pardon the multi-part question) What’s the roadmap like for the next releases? What are the major press-worthy features in the pipeline? When can we expect support for a cf marketplace command? What about support for deploying Docker containers?
Supporting existing Docker containers is an easier incremental step to introduce Korifi to new audiences

A

Yeah so far, I haven't gotten responses from other folks that I've tagged.

B

If I wonder, if it makes a lot of sense, then uh to to go to the agenda, I was hoping to speak a little bit about the future shape of Affairs.

B

Given that we have some team rework going on but yeah. If people won't show up I, don't know about this and also I'd like to live a bit earlier today. So I have more a bit like 10 minutes or something like that. More.

A

Yeah we might have to punt on some of these things. I'm gonna see just added some of them, but if he's not here.

B

A

It kind of texts, then.

A

I mean uh Karen I can certainly really any updates um to the other. Folks, too,.

A

Did you want to mention that team ryorg thing or like I always touch on that.

B

uh Yeah I mean the. The thing is that we we heard the news from Kieran Giuseppe they're, going to be relocated to some other projects and that American colleagues are remaining, but uh it will be just one parent capacity, so I, just I was just curious. uh If, if there is some clarity on things like are all Engineers remaining on Griffey, but just one pair is going to be working in the open source or are just the people remaining, and maybe uh if, if we.

C

B

Ideas, if who is going to take Giuseppe's Place, basically to make questions that that are in my head.

A

Oh yeah sure I I think we're we're working out some of that on the VMware side, um but, like my understanding, is that we'd, um like we'd, still see all the American folks contributing um and probably just on you know some sort of rotating basis and uh yeah I. Think we um we're working out who, on the American side, um would be stepping into Giuseppe's place on those leadership roles. So I I think that's close to getting subtle, but I don't want to. You know, commit anyone before that's completely confirmed.

B

Okay, so that's I, guess that's good news because uh we'll be working with all people, although we're a different time zones. So that's another thing like how do you envision? Because when we're working with Giuseppe I guess it's easy, we had to ipms when we need to and I guess this will have to change a little bit, maybe a bit more async.

A

Yeah, that's a good point.

B

So yeah that that's from the main things that I wanted to discuss with folks but I guess we can still talk about this next time.

B

um Yeah, as far as far as I hear it's not particularly clear yet so maybe next time we'll have more.

A

I think we'll have a little more clarity, um yeah next time for sure.

B

Yeah I can see. The drum has a lot of questions, so maybe we can try and answer here at least yeah, so I think we haven't released for a long time right, March April, something like that. So I think we have a lot of you know things.

B

We just finished the async jobs uh so that, basically, when you delete some resource, you can query if it's gone yet I can remember that we did a lot of big things but I, if, if other folks can help me with uh uh because I was in a Long Live recently, so my memory is a bit. You know we.

D

Did some k-pak optimizations one was reusing the cash so that builds after the first one uh go a lot quicker. Previously we were creating a new kpac image for each build. So it's a nominee. Well, it didn't have good cash opportunities anyway, so that's good, also um rolling deployments.

D

um So previously it was CF push. It would wipe out your deployed app and replace it with a new one. Now it's a lot more kubernetes like where it'll just replace the running pods one by one. If you've got more than one of them.

D

Those are a couple of nice user-facing things that happened.

B

um Yeah I think so. Basically, you can now pick a build back from the ones that you have available and skip the detection phase.

B

What else I have a lot of stuff because we haven't released for a long time? uh Maybe uh we can send you a list of like more a more like a specific list offline, because I'll have to dig into the like stories or commits to give more details at this point.

E

Zero eight milestone in our GitHub think GitHub project, but I can only see four issues. There.

B

Are definitely more than that I guess it's just not zero maintained Maybe uh but yeah there there's a lot of cool stuff uh user, user-facing stuff! That's going to come out and I! Guess! Maybe now that we've done the job say pick, maybe we should release so that we get this stuff out.

B

uh So I, don't know if that answers the question. If you're, looking for something specific Ram, no.

F

I think generally, that um that's helpful. The reason I put that question up is um we're obviously looking at uh some PR cycles and when I say PR I mean press release, so uh some press release around the time of cubecon Chicago, which is November so ideally about a month before that, we want to say that you know these are the uh five new cool tricks that kurifi has been able to master in between Amsterdam and now.

F

So uh that's like the big, um let's see overarching sort of uh question, and so um the idea is, you know: what can we provide as a list at that phase and.

C

F

Time when we had spoken to people, um we were talking about Docker support and something close to like um CF Marketplace or something that can do services on query fee and things like that. So my that's the reason I put in those specifics, but a general answer like what you said is like totally acceptable um I.

F

Think if we find a list of what um we've done between P, 0.7 and now um and also get a list of you know, these are the four or five things that we're looking to do in the next few months. I think that's totally.

B

Adequate mm-hmm, uh it's funny that Service, Marketplace and Docker is coming because that's exactly what uh with sap are interested in like we had specific questions about this as well, meaning that maybe yeah, maybe we have like similar expectations from both sides. The community I think the docker think might happen relatively soon.

B

I think the zip had a draft uh uh yeah Joseph ahead and draft proposal for this one, but uh it's not really something we're actively working on, but it if people are interested, we might prioritize it relatively soon, because it's not very big I guess as well with Marketplace I, guess, that's a bigger topic, managed services and like brokers. So that's.

E

B

Very big thing that that we're also interested in uh at sap so I guess that's.

F

Good the reason, the reason um the docker story is kind of higher priority is because when we go and speak to people and say you know yes, but if we try it not everybody wants to start containerizing for the first time with corifi. A lot of people already have like container-based workflows, and the first thing that we want to go and tell them to do is just move those containers over to kubernetes through Cloud Foundry, so um it provides a nice Ram on-ramp into starting to use Curry fee and experiencing that CF push um thing.

F

um Obviously we want the.

B

The end goal is always.

F

To get them to you know, get that source code in and use Curry feet through that, but having support for existing containers simplifies uh a lot of things for people who we are trying to uh get to try this for the first time. So um there's always that Gap that exists.

F

When we speak to people and say you know you can you can try this, but you know you'd have to point your source and contain these for the first time Etc and they always come back to us with um what, if I have existing containers and things like that. So um it's definitely something that makes it convenient for our job on the road. So to say.

C

B

F

It's also a useful feature in general, if, if that's possible, and um if it's a relatively small effort, um I think it given give us like a big win. um If we can just advertise that in general,.

B

Yeah yeah: it's uh definitely something we can prioritize soon, because we don't have anything like the mini epic, for jobs has done so I, don't know. If it's going to be that small, but we'll have to think about it. We haven't thanks for all it a lot.

F

Yeah, if it can be prioritized in some formal manner, I think that would be much appreciated.

B

Yeah sure it's uh sounds very logical to do so. If uh people from all.

C

A

B

Are asking for it.

A

Yeah ROM I think you know any feedback that you're getting from or that you're generating from like any workshops or um intro sessions. That you're running is super helpful, um I, I guess I'm curious, like what um what the mean appeal would be even um like in that introductory step to using karifi and the CF interface to run a Docker container on kubernetes.

A

You know because then you know it's not handling the build steps, and you know this- maybe not as big of a gap in between like CF, like the CF interface and just kind of rock kubernetes at that point. So.

F

Great, um it's only about uh you.

D

F

C

This was already available with.

F

Like CF for kids, so it's I guess the the only thing is um you're just providing them with options, things for something that they can um see and feel more, it's a little more tangible um when you talk to people about getting their workloads onto kubernetes, um because not every time, uh people seem to want to start containerizing with um you know, paquito and and stuff and um I'm, not sure how we handle like apps that are written in multiple languages.

F

So if they have like, let's say a react front end with some PHP packet and things like that, and they already have a container for that. I think it's easier to just use CF to get the container onto kubernetes versus go through the whole pack. Build exercise um because I I haven't tried, uh tried out any of those I yet, but those.

D

F

Where I feel um yeah, proving that Curry feet just generally Works um and then proving that corifi works for your apps design, specifically um for your environments and things like that, so so I think it's a two-step process that happens there, and so um we probably have like half of that step covered right now um and then we once with the with the docker stuff I, think um it it's just uh a way for us to convince them that you know containers can be managed using uh this and then all the newer apps can follow after that.

F

A

Okay, got it so yeah I mean you know. I I could see like saying, like hey sure you, you know, you've got this pre-built container and there's even a smaller set of yaml and some simpler imperative commands that you can use to run it, and then you can build on top of that to actually start using build packs embedded in the platform to take it. Everything from Source like once, you've gotten kind of familiar with that incremental stuff.

E

I can also Imagine people having couples that are not simply not buildable because they're, for example, missing buildbox, or they have some weird CI process or whatever, and they already have their CI pipelines. That is producing Docker images for them, so they could just just to run. Those Docker images on platform on kubernetes is going to be pretty cool.

A

Cool well um is that a wrap for today then.

E

um Yeah, if Ram doesn't have any more questions we can start from the beginning, though those are some technical discussions. Now we can talk about the branch protection exception. Giuseppe mentioned that with latest changes of Branch protection rules for cloud Foundry, GitHub repos is going to be protected, meaning that we cannot push directly to Main um and he said that uh appear has to be made into the cloud. Foundry configuration and Tim is able to approve it and merge that PR, but because team is not here, foreign.

A

It's probably to this file in the community work.

A

That's where the pr would go to override some of the branch protection configuration.

E

Yeah, maybe either.

A

I, don't I don't think it's there yet, um but that's just for uh deviating from the default configuration that would get generated um and I believe Giuseppe turned that on.

A

Let me find the exact reference in the charter definition, oh yeah. So if you, if you look in the embedded yaml in the charter file, there's this um line in there under config, that is um turning on those branch protection, rule generation.

A

Or all the repositories that are in under the management of the working group, and so then that Branch protection.yaml file would then provide some overrides for specific repositories.

E

See so he also mentioned that uh the Machinery that's processing the TMO, which has been broken, so it can still push directly to Curves say, but so this is not going to be the case. Oh.

A

Interesting, okay, yeah that sounds like maybe something to discuss with the TOC, uh especially Stefan Mercer, has been the main person maintaining the actual automation there.

E

Cool and the first thing is about memory limits, so today we had a failure to deploy one hour, one of our reference applications that we used for acceptance and the problem was that build felt with uh built running out of memory.

E

So with one of the recent commits in Korea, people are able to specify memory and femoral storage limits for the build containers.

E

Now the problem is that when you specify a limit previously, we did not do that, meaning that the builds were unlimited by memory, but now they are so the default value of one gigabyte seems to be insufficient.

E

That's why we tried bumping it to something ridiculously high, because we didn't know how much memory we want. So we put four gigabytes, but then it turns out that uh kubernetes and the way it handles Bots. If you specify limit resource limits on CPU, for example, it defaults the request to the very same value, which means that if you provide quite a big number for the limits, then your pot may not be schedulable on the cluster, because you simply don't have that much memory on the Note. So this is like a chicken and neck problem.

E

You don't you want High limits, but you want your pod to be scheduleable, so we came up with a magic number that kind of works for our acceptance environment, which is 1.5 gigabytes of memory. But then we have the question: do you want to always specify those limits? Maybe we have to make them optional, meaning that if people don't care to limit their memory on build pots, they simply make may choose not to.

E

um Just ever mentioned that, according to his memory built, build memories for cf1 VMS is currently Unlimited.

A

Because I don't think that's the case.

E

Yeah yeah, maybe maybe that's wrong, because people are paying for memory for their apps, but who is paying the bill for the build memory.

A

That might also be controlled by the memory limit on the app itself. I'd have to check.

C

A

But there is some limit that gets applied.

E

So if that's true, we can probably the problem is that this limit is a configuration of the deployment. It's not. um It cannot be configured per up, at least with the current implementation. Maybe we should change this.

A

It's worth double checking exactly what um slot controller is setting up in CF on VMS.

E

Yeah but it's a bit confusing because you force people to specify limits and uh we don't have guideline what those limits should be.

E

Okay, I can write start a threat on slack about this to see what other people think.

A

Yeah I think I think that sounds good I. Just don't remember off the top of my head, how we ended up solving this in CF and VMS, and there we've got the same allocation issue, um because that um you know that that memory allocation is effectively both uh the request and the limit um in terms of kubernetes resource specifications.

A