Kubernetes SIG Cluster Lifecycle, 26 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Cluster Lifecycle - Cluster Addons 2021-10-26

Description

https://docs.google.com/document/d/10_tl_SXcFGb-2109QpcFVrdrfnVEuQ05MBrXtasB0vk/edit#

A

How is the background.

B

All right welcome everyone to the cluster add-ons meeting today is the 26th of october and yeah. I think welcome everyone, and I think lee had a quick update from flux and oci things.

A

Hi friends yeah super excited to share this. uh We mentioned it in our previous meeting that the carvel team would be making a contribution to flux, to support, oci images being pulled as a source and that proof of concept slash draft pr has turned into a productive and fruitful discussion.

A

uh It's sorry, I think it's pr 450 on the source, controller, repo and the flex, cd org, I'm not mistaken, and um yeah ben hale from the carvel team, with vmware made that upbeat or made that upstream uh kind of contribution of that internal work.

A

uh It includes the oci image api and the flux team has left some feedback. We change the shape a little bit and it looks like it's gonna happen. So that's. That's really great there'll be core support for oci images. Assuming that everything continues to go smoothly there inside of flux.

A

um For sure I started opening up some conversations with the argo team when we were at get upstairs slash, kubecon about doing this in argo as well.

A

At some point, we probably want to try and reach out to the fleet folks from the souza rancher ecosystem, because we want to see support for pulling these oci sources all around for all sorts of add-on. Me type things and continuous delivery bits, and it's good because we were talking about using this as a delivery format inside of the add-on operators as well, and so we'll have this kind of packaging capability all around the ecosystem.

A

And if we can see some adoption of these oci packages become more popular. For instance, it's it is in tons of community edition red hat's, also using it in openshift.

A

um Microsoft is doing all sorts of stuff with oci bundling right now, then, that's kind of like the little big ecosystem picture of that- and this is a small piece which is um flux, is going to support oci images and uh go check out the code yeah. That's uh that's! The update there from cluster add-ons kubecon oci image, stuff.

A

Related there was a very interesting talk at kucon that was using oci registries. They were streaming an irc type protocol in a way. So people were chatting off of the image layers in an oci registry. You can go check it out.

A

Weird weird stuff: there.

A

Yeah, I I don't know if we had any uh other updates uh from a.

C

Cool builder, declarative.

A

Pattern stuff or cops stuff, but that's all for me.

C

On the flux side, uh I think uh it looks like it is ages before the kubecon was 10 14. I believe- and there was one flux- road map to general availability by hideous and michael bridgen from v works.

C

So I had moderated that and they had said they already have achieved what you call the general availability- and uh I don't trick like right now- the details, but definitely they had done a one hour, 40 minutes presentation was it yeah, 30, minutes presentation with some q and a and uh sorry for the not being able to reclaim because too many conferences you know one after other, I went through kubecon then went to the 5g world forum, then the next one quantum week, so I literally lost track of it.

C

But flux was really in good shape and I'm happy for them. I hope we can help them to add to cluster api and few other crazy things. I'm thinking about. Can cluster api address the quantum computing can cluster api address, acceleration which is being promoted in the sense of gpus and how? What do we need to do to enable such kinds of things is running?

C

In my head, but at this time I'm more focused on the standard development organization of how to make they call it as nisk niskis, uh noisy uh intermediate scale, quantum era niskara.

C

So I am defining something called rask, which is reliable, advanced scalable, quantum era. So how do we move from niskara to raskara and how we can plug in the cluster api, which is required to enable these things? This will be in my head for as we go forward. So that's what I will bring it to this forum. Thank you.

D

Is that is that from kubecon? That's something I really miss from kubecon is like new ideas and and different ideas.

D

But it sounds cool prakash.

A

Yeah there was certainly a lot of talk about gpu acceleration and.

A

The you know: it's you, you have this like new compute resource and to get these workloads scheduled on them. You want to use other people's workloads and sort of an add-on format. You know the drivers for the gpus themselves are um kind of add-on-e, and uh so how do we configure our clusters to provide the platform for that compute uh and then also to schedule those workloads? And then you have like quantum accelerators or that sort of thing, as well different cpu architectures for different kinds of code?

A

A

I'm like thinking now like having a flex bootstrap repo that just shows you how to set up that stuff. It's probably something somebody could sell so.

A

Thank you, prakash.

A

Well, I'm probably going to need to drop because I'm in a conflicting meeting time block right now, but I didn't want to miss this. Let's see all your faces also talk about the productive things happening in the ecosystem.

D

A

And thanks for sharing.

D

That uh that carville link as well the that's uh or that the poll for for pull 450. Thank that's. I'm gonna. Take a look at that.

A

Yeah yeah, that's been a really uh really cool and uh productive conversation seeing lots more across company interactivity because I think the d2 iq folks also have been contributing a lot as well as microsoft in the fox project.

A

Cool all right. Thank you, friends, lovely to see all of you, I'm going to hop out.

B

Take care lee thanks.

B

Yeah justin did you have anything you wanted to talk about? I think a couple of pr's landed recently in the cluster.

D

B

D

Going yeah, I uh I think we're this is all trying to get the final steps for integrating into into chaops. uh We are very close, we're not there yet I apologize, but uh we are we're getting there and um hopefully very soon, I'm trying what the next step is.

D

Yeah, I think, actually um maybe I'll, put something on for next time we need to. I think we need to discuss the our back model in general for operators, um so I think uh actually I don't know if we need to discuss it, but anyway, uh that's one of the things that's been um top of mind is, I think we've talked a lot about the rbc model.

D

In the past uh we've had this motion notion of a very restricted rbac model, where the operator um assumes that the rbac roles are pre-created for it, and basically I've been doing some prs to generate those are back roles um and some tooling to help uh for that case.

D

um Basically, building on some toshi's work from google summer of code last year, I guess now um to um polish it up and deal with that uh that restricted mode, um and so there should be a couple more pr's that are going in just to make it super easy to create.

D

um Give it a manifest to create the operators that work in this in this style um and then I think, I think we're good, like I think yeah I need to get it actually all wired in, but that was the that's the current uh area, where I'm working.

B

And do you think we're going to need like more people to to discuss this.

D

I think I think the next step is to get the first one in there uh in some way, and um I think uh you know there's a good community of people that are actually like interacting with the core dns operator, which is going to be our first one. So I think that's we can do that largely on on github and then uh we've had other people, particularly from cluster api, that are interested in how uh operators that have rack permissions in particular in particular, like they really look at thinking about uh cloud controller manager.

D

How those should work. So I think if we could get one of those going as well, that would really see the discussion and we could actually have a ton of uh a ton of really productive discussion and hopefully traction at the same time.

B

uh What was the cloud.

D

Thing called the cloud controller manager is the extraction of uh you know the cloud functionality from coop controller manager, and there is now a aws specific one and a gce specific one, and uh as your specific one, uh you know insert your favorite cloud here for every cloud. There will be a cloud specific cloud controller manager instead of having multiple clouds linked into coop controller manager, but it does mean that now there's another component that has our back permissions or sorry, not our permissions cloud. Iam permissions is sort of the problem.

D

It has to have permissions to do things on the um on aws, for example, or on gce or on azure.

D

Okay, I think I said our back before, which is completely wrong.

B

Yeah, I was just wondering like because I noticed like in the in the past couple of weeks we had less attendees here, but you know more pr's coming in, which is is good, so it sounds like a lot of the long discussions we've already had. But when you said like we should have a conversation I was just wondering like do. We need to pull people from somewhere or you know, re re-energize, this group, or you know it's just the thought I had.

D

Yeah I mean I'm happy with it that there's uh like that, we're having the conversation on github as it were. I like that. I think um there are different uh things that it's appropriate to have on the different forums right like the.

D

The when it's more about like execution, I think github, is a fine venue when it's more about like how should we do certain things. I think this sort of thing is is a good venue. I think we've already had a lot of the discussions uh here, uh and so I think we can. We are now mostly on uh on uh execution type things, but we'll see uh what happens.

D

um I think the next step is. I think we've talked about um the sort of restricted our back roles in the abstract.

D

uh I don't think people have seen them in the flesh as it were in the code, uh and so it would be good to have one that actually demonstrates this and where people can say. Oh, yes, I I like this right. I don't like it. Obviously they're gonna like it, but you know where people can say. I like it.

B

For sure it makes a lot of sense.

B

D

But that's all I've got.

B

Yeah, I've, nothing to add.

C

All right in that.

B

Case, let's wrap it up. Thank you for for meeting everyone take care thanks. Daniel bye.