Kubernetes Storage SIG COSI, 21 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Storage - Object Bucket Standup Meeting 20200921

Description

Meeting of Kubernetes Storage Special-Interest-Group (SIG) Object Bucket Standup - 21 September 2020

Meeting Notes/Agenda: -

Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage

A

All right start it: okay,.

B

C

B

Nope, it says: screen sharing disabled.

A

Oh sorry, I need to make you a co-host.

A

Okay, china.

D

B

So, as of last week, we discussed a deletion workflow.

B

We were talking about deletion in terms of it being a long running operation where sometimes, if the number of objects in a bucket is very, very large, and if you're trying to delete the bucket the amount of time it would take to finish, the deletion call will definitely exceed the deadline on a single grpc connection.

B

So what we discussed was whenever this reconcile loop for delete, starts, we let the grpc connection time out. If it does and then, if it does timeout, then we put the reconcile. You know unreconciled object back into the queue that needs to be processed, but the way we put it back into the queue is such that it it does a exponential back off and uh over the last week uh after thursday's meeting, we actually implemented uh our core.

B

I would say, like a shared controller, lib library, that does this uh right now we're testing that library, um so that uh all the different components in the architecture can actually use that before it can use that now, um the next step, so I just drew a quick chart of what needs to be done and like what are the next milestones we can look forward to so.

B

The first thing I want to say before starting this is um uh the cap is in a position where uh it can be reviewed. um We have uh kept it up to date, uh based on any new feedback. We've been getting so if the people here could spend just uh 10 minutes out of your day to look at whatever parts of the camp is relevant to you and leave your review on the on the cap. That would be very much appreciated, so the this the review is the step that is.

B

This is the last step before we can merge the cab and start pushing this code into the kubernetes six repo, that is in the kubernetes six organization.

B

So, in the meantime, we're still using our uh private organization container object, storage interface. uh We want to stop using it as as soon as it is possible that as soon as the cap is merged, okay, let's get into milestones so the first step um with the way we're proceeding. um So the first thing that that will be able to show is actually a quick demo of the most basic workflows uh I'll get into that. Let's just quickly cover what else um is there?

B

So after the demo, in terms of development, we will be uh developing to uh we'll be trying to get to the milestone of minimum viable product, which is quite not alpha yet and I'll explain why, shortly and in parallel, we'll be doing an api review. The api review and mvp will go hand in hand. That is if the api review comes back and if there's some feedback that needs to be incorporated that will be incorporated into the mvp when that gets solidified.

B

We'll we'll uh also have a demo at that point as well, and uh after that we will uh push for alpha in alpha there'll, be testing set up, there'll be a you know, some at least small amount of examples, and maybe uh very simple docs.

B

um So alpha is not just uh you know: basic working product it'll be a little more than that. Now, um let's actually get into what what first demo I'm thinking about. um So the first thing I I want to aim towards is is actually a very simple demo of provisioning, a green field bucket, so uh a green field bucket is where you know it's a brand new bucket we're not sharing this bucket or anything in the first demo. All we're doing is only one uh bucket.

B

We create using the bucket access request and bucket class, and then we provision access for that bucket by creating a bucket access request and uh once the uh and bucket access class, and once it is provisioned rocket access provision, we then mount it into a pod and access the bucket from the part. This is the the first demo that I want to aim towards.

B

um The reason for this demo is because it it touches upon all the three components in the architecture, so um the three components are central controller manager, cycle controller and the vendor provisioner for the cycle controller, and also uh the node adapter, which is the csi driver that will put the bucket into the part.

B

um So we've started working towards it. Like you were saying earlier, um we have defined the go apis for all the six bucket objects, uh bucket request, bucket access request, bucket class worker tax class buckets and bucket access, and we've also developed a shared controller lib so that anyone who is or any component that's using this bucket star api bucket apis can actually simply just import this and start start using it without having to rent the wheel within the project.

B

um The thing that that's left to be done in terms of like shared code is having a test framework that all the different three different projects can use is not the highest rarity right now, because we can start focusing on developing the actual uh components.

B

So in terms of the actual components that is the central control manager, sidecar node adapter. The the bullet points in green show what has been done so far, and the bullet points in orange are left to be done.

B

So we've started working on this after uh we, we got approval on the design and the cap. We when I say that I mean not the merge itself, but where we all agree that this is the right design um in terms of uh central control manager. uh After the shared lib was shared, controller library was, was developed.

B

We quickly went from there to having at least a dummy listener for all the different objects, and I want srini to give me an update on this, but as far as I know, srini, we we have that we've we've written an admission controller that will ensure that there's only one bucket class in the cluster correct.

E

That will ensure that bucket class is specified in the incoming bucket request. um I don't think we, so the way is when the br comes in. If there is a default bucket class, we will use that um to populate the br. If the br already has a bucket class, we will let it go through. So there is uh no check that multiple classes have default uh annotation at this point, if there are more than one class has a default annotation.

E

um I don't really catch that right now.

B

Okay, so that needs to be done right. uh Do we do that for service classes? How do we do that? We don't um no storage.

F

B

F

Talked to john about storage classes, and he said there can be as far as he remembered there could be multiple defaults and the result is unpredictable.

E

Right, I think the code picks up the first one that it sees. I guess I don't know I mean that's probably I read that automatically.

C

I think it may also return an error if it sees more than one.

E

That can be done.

B

Yeah, so that needs to be done um so we should ensure that there's only one default bucket class. What was the thing you're saying about um so you're saying if a br comes in before there is any default bucket class and if the br does not have a bucket class field, it returns an error correct. No, no.

E

No, if there is a default bucket class, the br gets populated with default bucket class. If there is no default bucket class defined in the cluster, the br will will not have a bucket class defined. Then it will throw an error.

B

Yeah, okay, um I.

E

B

If uh okay yeah, that makes sense, I think that'll- be useful I'll make the codes, including.

E

And it is easy for me to catch the error if there are multiple bucket classes. Okay, we have default, that's not a difficult thing to do, but should we do that or I don't? I have to look at knowledge class.

B

Okay: okay, yeah yeah, please look at storage class and uh let's, let's have the same behavior for bucket classes in terms of uh default, yeah, okay and and what is the progress of the central controller manager in in so.

E

Yeah, I am now I started using the uh the new framework, so I had code that was using the external provisioner before right. So I need to put that back into the the new framework that should not take long. Probably.

E

uh I will be working mostly on that today. So.

B

Okay, okay, all right so yeah, so, first to start using the new uh library. That's that's the thing you're working on right now, right, okay,.

D

um Okay and is rob on the call.

B

Nope, okay, so rob rob, has been working on the shared controller library that you know is explaining here. um We actually last spoke on saturday uh shout out to rob for his dedication.

B

Actually, everyone in the team um and what we last spoke of was the the we had to do some testing on this on this shared controller library, and uh so so he's actually implemented uh the logic and, and just uh the testing part is left before we can. We can uh we can kind of brand. It saying you know it's, it's it's in good shape,.

B

Again on on the sidecar side as well, uh if I uh last uh you know, we had a sync up: uh we, since you're rewriting from scratch, you know we'll have to start over on all these different events, one by one in terms of moving on to the next component in terms of node adapter. We have krish starting out on this chris. Do you want to give us an update on where you are.

G

Yeah sure so um you can hear me: okay, okay, uh yeah, a lot of the initial boilerplate code for the identity, server and node server has been written for the csi driver, and so I'm just working on uh adding functionality for querying the bucket access request and bucket request. uh Okay, get the values for the uh for the grpc.

B

Calls okay, so this is actually literally. The second point here um sounds good okay, so for the demo, uh the things that need to be done are colored in black text color. So so it looks like chris. Chris is working on this part in terms of node adapter.

B

We'll need two more steps, which is to create the volume mounted into the pod in terms of sidecar, we'll need to reconcile bucket, create and bucket access, create and write a dummy, provisioner a sample one and also have the deployment uh spec for it same with deployment spec for the node adapter. I've missed that um in terms of central controller manager, just listen for creation of bucket request and bucket access request and respond to that and have the deployment spec for that as well.

B

E

That is possible yeah that can be done quickly. Okay, um yeah go ahead. Sorry.

B

So uh I expect so you know this. Is uh uh you know this? Might I think I think a reasonable expectation of when we can do the demo is about two weeks from now. Maybe three weeks from now, um it also depends on people's availability, but it is, it is doable within that much time. um So so that's that's what we're looking at as of right now. So so that's that's my update for, um like our development plan. I think uh now that you know I managed to capture in this format.

B

uh We can. We can follow the development closely, all of us and and uh start contributing code pitching in um and and help make do this faster, even if possible.

B

So if there's anyone on the call who wants to contribute to the development you're welcome, um please maybe just reach out to me on slack kubernetes slack and, and we can go from there, um yeah that that's that's all the update from me.

E

So one other thing we did was uh we changed the api groups to storage.kids.io.

B

Yeah yeah, as of uh last week, we uh had this conversation about changing the api group to storage.kdio, and that has been done as part of the api definition. um I think srini you, you have a pr open right, so I know.

E

uh Yeah first pr I yeah there was some kind of check in now after that changed that so that I did a pr yesterday, night yeah, that's open! It's true yeah, you merged one of my pr. So, okay, you review that yeah yeah.

B

Let's yeah, actually you know what I will um I will share. I will share that pr also on this chat. People are welcome to um review this yeah.

E

And couple of things as a couple of pr's, I put out onto the main kubernetes six uh for uh starting the ci okay, uh so that's basically copying the the release directory uh with the update to the to the headers. I mean the the uh you know um see.

D

What you mean yeah yeah, getting the.

E

Ci working basically yeah so um xing or anyone uh from the community if they can review that those will be ready so that we can put our code after after we got the clearance so yeah. I just want those uh uh ripples to be uh ready to go so that we don't spend time on or.

D

E

Those things so.

D

Make sense so uh so is assad on this call.

D

No, I don't. Okay.

B

So so uh I'm I I I want to again ask uh the people that are on the call. Please take some time to go over the cap, even if it's just the parts that you care about that you're interested in say, if you only care about market provisioning or access provisioning. If you just review your parts, the co, the combination of evidence, reviews will add up and and will cover the entire cap. So please take some time to review the cap today or during this week.

B

Whenever you get a chance um and then it'll help us move forward.

A

Did you fix the commit?

A

B

And I also fixed the other comment you gave about already exists.

A

B

Yeah so yeah, so that's all from my side. um If there is anything else, anyone wants to bring up now is a good time.

B

Okay, thank you everyone um again. If someone wants to help contribute code and wants to be a you know, developing you know developer helping out with this project. Then then, please reach out on six storage, cosy, channel um or ping me directly, whichever you prefer, my slack uh kubernetes slack id is wlan0.

B

I think let me check yeah. That is me wlan0, so yeah just ping me and we'll go from there.

A

And also, are you using the project board under kubernetes six? I see that you have it or are you using that for tracking.

D

Yeah yeah yeah.

B

Yeah we we started using it uh last week. I think.

A

E

We don't have a whole lot, but whatever we thought and uh krishna also published some templates for the um for the pr, as well as issues, so that pr is also outstanding on there.

F

E

So yeah we need to link the prs here at some point in time. Sit so that people can. If anybody wants to reviews these things so would be accessible. I can do that.

D

C

I have a question and when do you think it's, it can make any sense to um to try- and you know, run some kind of uh example um of of this code like when the demo is ready, maybe right, but but of course after that can do you think it will be usable for anybody who's trying to just create a cozy driver, as as an example, of course, just. Would that be a good timing? Yeah.

B

Yeah, I don't intend the demo to be, like you know, just demo demo quality code, I'm a demo, will be a part of the development uh step. Basically um so yes, when, when when we are ready to show the demo, I expect the code to be.

B

You know such that others can use it as much as is in the demo. So just provision bucket provision access mount back into part. That much will be usable by the time we show the demo.

C

Okay, so that would be something at least I think um would be a good practice and maybe a good feedback point that we can sort of try you know so so I can. I can maybe try creating a cozy driver and then you know be able to give some feedback on like.

B

Yeah definitely that's.

C

How this looks from that.

B

Side yeah, that's that yeah! That's that that was the hope from our side um that will help us ensure that even the way we deploy. All of that is kind of reviewed, and you know some good ideas might come out of it. So.

D

C

And do you know what will be required in order to get to that latest kind of code um building with it I mean how do how do I actually get access to that build, or anything like that? Will the demo be include some guide of like how to build driver, how to deploy anything like that? Yeah.

B

Definitely so uh I I think I've also shown it here.

D

B

Oh here, oh yeah, there uh yeah so there'll be a deployment spec, so obviously there'll be a docker image which will be the build for that demo. um So yeah the deployment spec uh you know. Ideally I want to make it a single customized template where you should just be able to do cube, ctl, create dash k for customize and then url be github.com.

B

Kubernetes sigs contain object, storage interface, something something like that. If you, if you just give that github url, um then then you know it'll deploy the entire thing with the latest. um That's what I do for uh one of the projects in uh mineio where uh to deploy the entire csi driver, which is two different components.

B

It's just cube. Ctl create dash k, github.com menu, direct csi.

C

And the build and the the driver build is it so there are some like the dependencies on the api. I mean the rpc8 and all that, so these will come from later. Yes, somehow right, okay, yeah.

E

Okay, I got it.

C

So it's the details we'll get to it later, just but yeah, but yeah, okay,.

B

Yeah yeah, by default, we we will do it as latest, but uh uh you know the the philosophy of you know like right. You know this development is um always work with latest and uh that's the ease.

B

That's the quickest way for us to find bugs, and uh you know, fix it right away um if we start cutting versions and then start making changes on it, uh we'll we'll end up having problems with, like I mean at this stage, uh we'll have to start back porting already and doing stuff like that, which is better to avoid right now, as a first step until we get to alpha, I think we'll just do always latest.

C

Okay and so I'll, keep that in my timeline, excellent.

B

Sounds good uh if, if you have some cycles to contribute code, uh please reach out and uh we'll we'll get you set up all right. Thank you. Everyone! That's all from me again! If you have any questions, please bring it up now. Otherwise we can. uh We can end the meeting here and uh let's meet again on thursday, all right thanks.

A