Kubernetes SIG Cluster Lifecycle, 6 Jan 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Cluster Lifecycle - Cluster API Azure office hours 2022-01-06

Description

No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).

A

Hello, everyone today is thursday january 6 2022, uh and this is the first office hours of the year 2022 of cluster api provider azure uh happy new year. Everyone hope you all have a good break. um This is going to be a light meeting. There aren't too many agenda items, but um let's just uh go around and see what people have um all right. So I don't see anyone new, so I'm going to skip over the welcome new members, part um and david. You have the first item. Go ahead.

B

All right, I will uh try to keep it relatively quick, so vmss, uh the equivalent of like uh um was it the auto scale. Groups uh in aws uh is introducing a new orchestration pattern uh to the mss and it's called emss flexible, and that would allow us to basically take any vm, that's built in azure and add it to a scale set.

B

Basically, it breaks the vm model out from the larger orchestration model, uh because before in the uniform vmss, which is the vmss implementation that they're there today, um the the vm, the actual virtual machine, is like embedded into the resource model there, uh and uh it has some implications on how the infrastructure is spun up and where it lives.

B

In uh you know each rack and that kind of thing, so uh vmss flex uh introduces uh behaviors that separate out the vms from the actual orchestration model and should provide us with significant performance increases for disk activities. Attaching the networks delete, create.

B

There are some api changes, so the orchestration model isn't going to be completely the same, but um there's a spike out there. I opened up a whip pr and it's kind of set out for a little bit, so it's grown sale, but you can see the rough differences.

B

um We are I'm going to try to get that pulled in, so not not that with pr but probably restart it and take what I learned from there and put it into a new pr close this one out clean it up, get something out there that that's ready for review um yeah any questions, but that oh this should we could run a uniform. um This is also meant to replace availability sets at some point.

B

So uh the way we run machine deployments control planes.

B

We could look at this as as how we run all of stuff. Instead of just uh some things are availability sets. Some things are scale sets, so it's it's. It could be helpful.

A

I'm sorry I was muted. I have a question for you david, um so you say we could use it for all things, but have is using emss flights to back azure machines, as opposed to azure machine pools. Is that.

C

A consideration is that something that would be possible.

B

Yeah, so that's kind of what I was alluding to is is we could create machines and then add them. So, let's imagine we have machine deployment um and in capi a machine gets created. Infrastructure gets created for that machine, so we have a named machine, but we also know the group. So we could take that name machine and add it to a.

B

Dmss flex, orchestrator and really not have it do any orchestration other than just like uh availability set stuff.

A

Yeah, I guess I was thinking about it more. The like the other way around, where you put like everything, is an azure machine in that case, as your machine kind of becomes maybe less useful, and then you say like what kind of like instance is this: is this this particular machine, a skill set or is it a dm? And then that is like a configurable property of azure machine and they can all be in the same sets like in the same flex, set.

B

That's interesting um yeah, I don't know we should we should probably. We should probably explore that um I hadn't thought about it from that that perspective.

A

I don't know this is just me spitballing from two.

B

Seconds of thinking about it, so I don't know, but it would, it would be really nice to make some more yeah have some more uniformity uh across how we deploy compute.

A

um Yeah, I think one would be a good person to talk to. He doesn't attend the meeting very often, but he I know he's thought about the stuff a bit um um matt. Do you have another question.

D

Yeah, I was just curious, like I think you probably told us in the past, but I wasn't paying attention. What is the state of flex, I mean, is it still pre-release or is it rolled out anywhere uh so.

B

It's ruled out, um we can use it now. I think it it might be like public preview. I don't know if it's gone ga, um but it's not feature flagged or anything like that, but eventually.

D

It'll be available everywhere, like not eventually, but within maybe a year or so yeah.

B

There are a handful of a handful, not a handful, like probably two or three apis, that we would like to have that we don't have right now and those are kind of batch apis. So, like um hey, I want to batch.

B

You know re-image, a bunch of instances right now. You'd have to call that individually and that that isn't the you know that isn't the the best way to do it like you, want to call azure with batch of instance, ids and say: hey, go and update these um so they're working on that. uh It's just uh they're.

B

You know untangling uh this big ball of yarn to uh make it easier to work with across the vm provider, as well as the skillset provider.

D

And uh is there a link in the pr somewhere for lazy people like me to look at the api, I'm curious how.

C

You know what I didn't. I think.

D

I can do that. I was just asking.

C

The ducks in the in this deck- um I think the api should be in here somewhere. Oh.

D

It's probably right in here yeah cool.

A

Awesome um another question: um have you.

C

Talked to the club provided folks by any chance, do you know what they're thinking.

B

In fact, I just uh started a thread with penfe from cloud provider as well as the aks team. um They have not started testing with this, yet so what my intention is to get the infrastructure spun up um point them at a branch and have you know, have them start using it to spin up uh test environments, and you know figure out the bits that work and don't work, because there's there's going to be the apis have changed significantly like when you ask for an instance id from vmss.

B

Previously you would just give it. The uh you know. Integer uh id, for that instance and vmss would return that uh the mss actually returns the id of the vm. Now so you don't get an instance id back. You actually get a id to that vm uh in azure, so like the the uri to that vm, so that that's where the model has now broken apart and dmss doesn't embed the vms, but in fact serves you back uh uris to jump to that vm.

B

So there's going to be significant changes in apis in cloud provider likely csi likely. You know all of the things.

A

B

Anyone have any other questions.

E

uh Yeah, so uh I've opened a pr for uh azure cluster refactoring, um so basically, what this pr does is like uh decouple all the common fields, um so before that a little bit of context on why we need this pr, so so to integrate cappy's cluster class. We needed to like uh separate out the common fields from azure cluster, because the azure cluster template uh struct would only have the common phase that I shared across clusters.

E

So this pr basically like removes all the things all the fields that are that can be shared across uh multiple azure clusters and and creates it creates a separate struct and then it embeds into the main azure cluster restart. So the idea is like when we create an azure cluster template. We could just like um embed the common struct instead of like duplicating all the fields, um so cecile has already reviewed it and we had like a discussion on the naming uh uh conventions.

E

um So I like I'd like to have like more eyes on that, uh because it's like affecting uh some important, uh perfect piece of code.

A

Yeah thanks for bringing that up and just in general, I think there are quite a few pr's in the queue right now. They're just waiting for you so try to get through them, but we've been getting more contributions than you know. We have reviewers so trying to catch up a bit.

A

um Does anyone have any questions about that? Pr refactor.

A

And I think I don't know if you mentioned that yeah I think you did, but this is to support cluster class. So probably something we'll want to try to get in in the next minor release.

A

um Any other topics or anything else.

A

Okay, if not I'll, just finish with what I uh bring up at.

F

The end, um what one second, just before we quit uh is there anything left to do for the uh kept the pr for ignition support and um flat car support.

A

Yeah, so that pr depends on the 1.1 cappy release, which is not out yet, and so we don't want to merge anything that it doesn't exist.

A

So as soon as we merge cappy 1.1, we can't merge that it's just.

F

All right, yeah, that makes sense thanks.

A

Yeah nice um yeah, so I just wanted to ask if this meeting time is still working for everyone. I know priorities, change and times change from one year to the other. um So this is probably not the best place to ask like matt pointed out, because it's a vice group since we're here um but I'll, ask in slack and maybe create a doodle to see what the best time is for everyone um and try to see. If you know we can keep being as inclusive as possible.

A

Cool all right, well um good, seeing you all and happy new year and see you in slack.

B

Thank you, bye-bye. Everybody.