Kubernetes Storage SIG, 29 Jun 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Storage Meeting 2023-06-29

Description

Kubernetes Storage Special-Interest-Group (SIG) Meeting - 15 June 2023

Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.4sh8h79xdb38

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

Moderator: Xing Yang

A

Okay, all right go for it.

B

All right thanks.

B

Hello, everyone today is June 29th 2023. This is the kubernetes storage meeting today we're going to go over 1.28 planning's brush sheet and the next deadline is July 18th, that's the code freeze. So if you are working on a feature, please remember this is the date that oh, the code related to a feature targeting for 1.28 needs to be merged.

B

uh Okay, so uh so can you go to spreadsheet.

B

Thanks so let's go over this, the first one is the cover from resize failure. Do we have Haman here.

B

Does not look like one is there today so uh Michelle do you know if there was any update on this.

C

um I think Hamad said he was going to look into it, so so I think he's actively working on it.

B

B

And the next one is issues related to assuming volumes are monk points, so this is a bug fix this Jing here says: okay, so she's not here, it's probably no update.

B

uh And then next one is the provision volumes from Cross namespace snapshot, PVC.

B

So do you have any update on this? One is anyone here today.

B

Michelle do you know? Even though is.

C

There anything yeah this one is um blocked on um making the reference Grant API pulled into kubernetes core. There is a design proposal out for it, but it did not um get approved in time for this release. So I think this is something we could continue to drive for. The next release and I. Think okay, I think Taco fui was planning to to help push the design through.

B

C

Yeah we can work it as well. I guess it's currently Alpha, but it's uh kind of blocked on beta.

B

B

B

uh The next one is CSI World in health, additional metrics ET test yeah. So this one we uh there's a new contributor who is looking at an E3 test.

B

A

One uh quick question about this: so um yeah, uh actually the end-to-end tests uh so I believe that we had said that I think a few meetings back. We had said that uh we may be able to help out on this I'm, not sure what happened. Finally, so uh it's okay! If somebody's already started on this, that's not an issue, but uh but I was just curious if uh they were still looking for help or not.

B

uh Yeah so so I think for the e2e test. I will check with the this person who is working on this right, I think she's, ramping up. That's why it's a little slow, a lot of things she started. You look at ETV test, but I think um in terms of how to move this feature forward, um because I think we are kind of blocked right now. Right, if you test is one thing, the other things we are looking at, how we can uh really move this feature forward.

B

Do we want to um like add some other um house status?

B

Maybe maybe we need to move them into the PV or PBC status things like that, so those type of things- maybe we actually should talk about that got.

A

B

A

If yeah I may have misstated I think uh the tests that we were looking at were the conformance tests are here. So there was somebody from New, Zealand I believe who was who was looking.

B

A

B

Don't know I no I'm, not aware of that at all. If someone is working on this one, this person definitely have not talked to me about this. Okay.

A

uh That's fine I think we can follow up separately, I think it came.

B

A

In one of the meetings, but I don't think it was part of uh of we are planning spreadsheet if I remember right so so that may be something else, I think that's.

B

Yes, probably not, for this feature, conformance I, don't think, because this one doesn't really have any API changes so right right.

A

Right yeah, so okay sounds good. Yeah.

B

We can yeah, we can follow up offline on this one yeah.

A

B

Thank you. Okay, thanks.

B

And then next one change block tracking, uh so I know that we had a meeting the CSI Community Center meeting to discuss this and the core is supposed to update the cap. But I didn't join yesterday. So deep in working group, we're.

A

Still in process on that, we're still up.

B

There, okay, still working on that: okay, yep, okay, thanks.

B

And next one enable privileged containers for Windows, uh so menu yeah.

A

Yes, so we are still interested in doing this. uh I think we we need to figure out how we are going to prioritize this work, uh so I'm I'm in the process of trying to determine exactly when and how we can start contributing on this. uh There was a discussion last week uh where we talked a little bit about. You know uh what the approach should look like and Mauricio asked us to add some comments to the cap.

A

uh Based on that so uh I think, once we once we have figured out what, when we can actually start work on, this I will come back to this forum and provide an update. uh We. We are definitely interested in doing this. It's just a question of when that's it so.

B

Okay, thank you for that update this one secure Linux, 3 leveling, using Mount options, uh so I think Young's um PTO, so I don't know. If there's any update on this one Michelle do you know if there's any update.

C

uh Sorry, which one is this secure, Linux? Oh, so this one is dependent on the Reconstruction one and so um yarn has a fix out for the Reconstruction issue and that's being reviewed.

B

B

And then okay, we have a season: migration, remove entry, GCE.

C

So the um plugin has been removed. I think that PR went in a couple days ago.

C

um The overall GCE provider I think, has not been removed yet, but that's being tracked by different people that it is not here, it's it's.

B

C

Being handled by said, Sid cloud provider, okay,.

B

So that's not targeting 1.28 right.

C

Yeah I'm not sure what the status is. Okay,.

B

All right thanks.

B

And and then we have I think those are just to keep track, and then we have the Seth RBD deprecation I believe in there is a PR out. I, don't remember if that's merged yet Michelle. Do you remember Seth RBD one is that merged.

C

I think so the deprecation notice.

B

C

B

And then the next one is uh Seth fs and there's also a PR uh I is.

C

I believe so: okay.

B

A

uh Can we Mark both of these as completed or we need to keep tracking some additional work.

B

We probably want to keep it just to make sure if we have any documentation needs to be done. I don't know if there's additional documentation that that is needed.

C

B

C

Want to do a docs update.

B

C

B

Okay: let's go.

A

Ahead and Mark both of these as started.

B

Okay and let's see why it's okay, better default story class, so uh so we know that the the cap was merged, I, don't know if yeah I'm, not sure if there was any PRS for this yet Michelle. Do you know this? Is the default storage class one.

C

um I think the um the ga promotion PR merged Okay, so um this is done besides. Docs updates.

B

Okay, so there's no additional coding that I needed right, okay tests or or down the merge. Okay.

C

B

That's it so this one is almost complete. Okay, thanks and next one is a quality of service for volumes.

C

Yeah so I think I saw um the CSI spec changes are under review and I believe calorie um started, a PR for the API changes and then um Sunny is working on a a like a dock, to just um divide up a bunch of tasks so that we can kind of figure out amongst the group. If multiple people want to help implement this feature, we can sort of divide up the tasks that way.

A

Yeah, that's great to hear uh I I just wanted to confirm a couple of things regarding this. uh We missed the CSI Community sync meeting, uh but I uh I read through the the uh the summary and just wanted to confirm a couple of things. One was that it seemed like uh it was decided in the meeting that the resizer sidecar will essentially uh now watch for notifications for the Qs attribute changes as well and, in addition to any changes to the side to the size uh of the volume so uh is.

A

That is that is that something that's been finalized? Is that has that been decided.

C

um I mean I think that's the most likely path that we want to take. um Did you have concerns about that? No.

A

I think that's actually the right approach. uh We are fully aligned with that. I just wanted to make sure that you know uh that that is the parts that that we are headed uh down so.

C

A

The current plan excellent and uh the other question that I had was: are we targeting 128 for for the implementation for this? At this point or.

C

um We're trying to uh I mean realistically there's a lot of work that needs to be done so I, don't know if everything's going to be completed in the 128 time frame, but I think we're definitely going to actively work on it. Yeah.

A

So, if you're looking for any help, please let us know we would be more than happy to contribute to the implementation from outside.

C

Yeah awesome, yeah yeah is back I'm, going to ask her to um share the the doc that she has. She started a doc to kind of break out all the implementation into different parts, um with the purpose that we can. You know kind of spread out the work amongst folks.

A

Got it uh yeah, that's great. uh We look forward to that. uh I also wanted to mention one other thing, uh and this may not be relevant to other storage providers, but it is.

A

It is essentially a problem that we found when we were looking at our analysis of of this issue, uh so uh right now uh on the AWS side, uh if you do multiple uh modifications for a volume, uh then there's typically a cooldown period of about six hours uh that is enforced by the uh by the uh by our backend uh before uh subsequent modifications can happen.

A

So the consequence of that is that if we do a modification of the size and the Qs parameters, one after the other, uh the second, uh what the second update is going to have to wait for six hours so uh and uh what the challenge that we have on our side is that it's possible that uh multiple requests can land on uh different side cards uh and and when they come to the CSI driver, they may not be able to come. They may not come at the same time.

A

So so, because of the way the the leader election is set up for the sidecars each side, car picking its own leader, uh there is a possibility that you know uh we get the requests at different times and, and that causes one of the requests to get dated.

A

So to solve this problem, uh we are actually implementing uh a request Coalition capability on our side, which will essentially take the two requests and and kind of merge them and make a single request on the back end in order to uh to allow both the operations to happen simultaneously. So uh so we're going to be implementing that from our side, I think everything is still in line with what uh the cap is proposing. uh So I don't think there are any concerns from that side.

A

I just wanted to mention this, because this may be a problem that you may hit in some other form, because, uh depending on uh how the leader election happens for each of the sidecars, uh there may be some scenarios like that that you may have to think about as well.

C

Yeah yeah that that's definitely good and good to consider I think from um what I know of the the gcp API. It's two different calls like um resize. The capacity is a different call than um the iops, and so it doesn't have that issue right.

A

C

I think definitely um other systems might have that issue, but I would also say like if more, if more people have a similar issue, maybe one thing we can consider is like because I think right now in the cap, um the capacity and the attributes are separate.

C

um We could think about trying to combine that into like a single modify volume call yeah so but I think right now with the cup currently proposed. It doesn't have that, um but maybe it's something we can consider as an addition. Yeah.

A

Actually, uh we we thought about that. I I, understand that it's it's a special case scenario for us, so I'm! Okay, if you know we decide not to move forward on that front, but it would really simplify our life uh in terms of if, if there was a single uh API call that came into the driver with both both of those attributes we uh passed in at the same time. So instead of two grpc calls.

A

If it was one, then you know we don't have to worry about solving that problem from RN, so uh I mean just just something that I think. uh If there is some flexibility in that side, that would be really awesome. If we can incorporate that, but as as things stand, we can still Implement uh uh this operation from our side with with some changes at Target. So.

C

Yeah I think it'll be good to see like once. um People start implementing this if, if they also come up with the same problems,.

A

Okay sounds good yeah, so uh so yeah so I think uh just in terms of uh implementation. Yeah do let us know uh where how and when we can help uh in this respect and uh we'll be happy to contribute to that and uh uh we we will probably start our implementation of this request, policing capability at some point uh within the next week or so and uh we'll let your folks know if we have any concerns on that, but as of now, this is more Just for information than anything else.

C

All right, thanks, yeah.

B

Thank you, okay, so the next one, robust model manager, reconstruction uh I, believe we got an update earlier from Michelle. Well, yeah.

C

That's right: um Jan has a bug fix out for review.

B

B

And next one is TV. Last phase transition time. uh Yes of this one. We also have the cap merged I'm sure you know do we have any PR submitted for this one already.

C

Yeah, um the pr is under active review. um I made some comments a couple days ago.

B

B

um So the next one Moon extension for C4 sets um so it looks like Haman- is not here today, I'm, not sure. If there's any update on the design or the cap there I know there are some reviews going on I. Don't like Michelle did you did you uh take a look of that.

C

um I haven't seen any latest updates I think Hamot has been paying more attention to it than I have.

B

B

And next one is uh numbers for no Shadow. uh Yes, this one, we also have the cap merged and uh ashmash has submitted a working progress between test PR.

B

Okay and the last one is the storage capacity scoring. This is a a design in 1.28.

B

um I, don't know Michelle. Are there any PRS out for review.

B

C

B

Yeah this capacity score.

C

Yeah um I I have not had a chance to look at the cap yet, but the the cap is out for review. Okay.

B

B

So that's all we have in this brush sheet and we don't have any PR so design review requests uh are there anything else folks want to discuss today.

B

Okay, if not, then you have more than 30 minutes back thanks. Everyone bye.

A

You take care all. Thank you.