Kubernetes Storage SIG, 29 Aug 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Storage 20190829

Description

Kubernetes Storage Special-Interest-Group (SIG) Meeting - 29 August 2019

Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.5ndu5uoa3ane

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

Moderator: Saad Ali (Google)

Chat Log:

None

A

So today is August 29 2009 teen. This is the meeting of the kubernetes storage special interest group. So, as a reminder, this meeting is public recorded and published on YouTube on the agenda. Today we are going to go over 1.16 planning, we're going to go over the items that we've committed to you for this quarter.

A

Important dates to remember today is code freeze by end of day, if you have anything that we've committed to for the quarter, we have to make sure that we've got it merged by end of day and then the next upcoming deadlines are going to be. Doc's must be completed by September 9th, and then the release goes out on the 16th. If there's anything else that you want to discuss during this meeting feel free to add to the agenda, the link is in your invite any PRS that you want to discuss, design, reviews or miscellaneous.

A

One item that we might want to talk about is the face-to-face meeting. What we might save that for the next meeting. Okay, let's get started. There are not as many people on the call as usual, so you may not be able to get status updates for everything, but let's give it a shot.

A

So first item is CSI migration to beta, can david chang or deep give an update on that.

A

David is not on the line. Deep is not on the line Chang. Do you want to give an update.

A

Okay, no worries so as far as I understand we're deciding to hold off on moving this to beta. The reasoning behind it is that the shims for individual drivers are still alpha.

A

And so since the dependency is still alpha, it's weird to move the feature to beta without those dependencies also being beta, the beta was going to be a opt in beta anyway, so there's not a huge impact in keeping it in alpha for now and then we'll reassess and likely move it to beta next quarter. Next item is online offline, resizing volume, expansion to beta for CSI a month. You want to give an update here, yeah.

B

We got it March, so this was done.

A

Any remaining outstanding.

B

Work here, yeah there's a book that I'm working on about fixing the resizing CSI resizing of multi attached read/write volumes, I'm working on a bill, try to get this as a bug fix, but it's not a must-have items. This.

A

Is on external resizer, no.

B

It'll go in the cubelet, actually, okay,.

A

B

The issue that Ben was talking about in the CSI call except Ben was issue is like mode. Read-Only was the problem but read read we can fix without making a change in CSI aspect. Okay,.

A

All right, thank you. Next item is moving the mount library to external common repo. Travis has been awesome here and sent out tons of PRS I.

A

Don't see him on the call, so I think the current status is lots of progress made, but I won't be complete. This quarter we'll keep that work going and hopefully finish up in the next quarter. Next item is per volume, attach limits a Mont or Yong. You want to give an update here.

B

Yeah, so in this quarter we didn't have much to do in this, except we added some e to attest to that stress. The that really test the limit up to the nodes that can support the maximum limit. So yarn I did some tests. We got it worst and yeah. That's pretty much. Actually there's some movement about like how you just can override it and I think we should look into that solving that next quarter.

A

Okay sounds good. Next item is block volume, support which we plan to keep in beta this quarter, but fix the number of bugs looks like yon was working on this as well. Last status was a you know, nice fuzzy bug was fixed and there was tests being added any updates on that I'm, neither yon Brad or anyone else. I.

C

Did see that fix go in I have yet to run my my tests on it. So this is a good reminder for me to go break out my old test and test this new fix. Okay,.

A

All right next item is scalability testing. Any updates here.

A

Okay, we will leave that one, an updated and we'll get an update next time. ah Next item is volume, namespace transfer, uh I, don't think John is on the line. I believe he was out of office this week on vacation. As far as I understand, I have not seen any progress here, but the goal here was designed for the quarter, so we still have a month left to finish up on the design.

A

Hopefully you can get some progress here before the end of the quarter.

A

Next item is new design for volume, permission handling and then hand in hand with that is a revisiting UID GID handling and file permissions for Windows any updates on any three of those.

A

Yeah, we still have time until the end of the quarter. Since these are design items. Hopefully we can get some.

D

A

No worries next item is volume. Expansion for stateful sets last update here was the clip was almost merged. Any updates on that month. Yeah.

B

So the yeah the cap has a GTM, approve, I think it's. It needs little view from API side because we're relaxing a requirement that the volume claim templates in Stateville set objects could be edited like they were immutable before. So we need someone from API team that Jordan someone to a GTM it, but it's been approved by Sega ABS. It winds up reversal storage, it's ready for implementation, ok,.

A

That sounds good. Next item is CSI volume, health, seeing any updates honest.

E

Being well, okay,.

A

So folks, please take a look if you're interested in this design and if not, maybe we can help us separate meeting to discuss it or something like that to get some progress on it or we get punted to next quarter. Okay,.

E

A

Item is CSI entry, read-only handling, I, think John did not get a chance to work on this seminal Marcus. There's no updates.

A

Next item is alpha beta G, a designation for CSI spec. This is same as last time. Their CSI expect features that are currently pending and we're going to decide on a case-by-case basis, which one of them will go through this process.

A

Next item is the storage proxy for Windows for CSI I, don't know if deep, joint or someone who can give an update here.

A

Anyone on the line able to give an update on the windows work.

A

Okay, so I know they were able to start a new workgroup for the window, storage work and they've been meeting regularly. Hopefully there has been good progress, but I know, update in this meeting can follow up with them. Offline next item is snapshots to beta so Shane. Do you want to talk about this.

E

So we have made a lot of progress, but there are just so many things need to get in and there were dependencies. We couldn't get a everything in by coda. Freeze, so I think this one will be delayed until next quarter. But since we have done a lot of work, hopefully you can't get in merged early next quarter. Yeah.

A

And I want to thank Shing and Sean and Jing for all their work on this. This is extremely impressive. The amount of work that they got done, it's kind of upsetting that we weren't able to land at this quarter. But you know this will we'll deal with that and and we'll get it in next quarter as soon as possible. So thank you. Shing Thanks.

E

A

Item is volume, cloning, which is what John was working on, that did get merged and so I think the follow up here is going to be documentation and I need to ask about verify. If the tests are there I believe they're already there, but I need to double-check.

B

Well, the APA was started for the volume source. Yes,.

A

So the volumes the data source discussion, we suffer those who don't have context on this. There is a discussion earlier this week, both snapshots and volume. Cloning use a field in the PVC called data source which specifies what source to use to populate that volume in the future.

A

We're planning to open that to generic data, populate errs and potentially making those data populate is independent of the storage system that provisions the volume and the proposal on the table was to split out the definition of the data source from the PVC into a separate object for a number of reasons, and.

A

The problem with that was, if we decided to do that, then we would have to hold off on moving both snapshots and cloning to beta until we got that proposal approved built as alpha and move to beta. So we're looking at a six to nine month, delay on these other features, and so after lots of discussion, what we decided to land on was, it doesn't make sense to block all the hard work.

A

That's been going on here, we're going to pursue that as a third type, in addition to volume, cloning and and snapshots on the existing inline data source and the implications of that are, if it ends up going to GA that we would have two ways to be able to specify volume, cloning and volume snapshots basically inline, as well as a separate object, which is not the best solution. But it's the pragmatic approach, us I, don't think it makes sense to block these existing designs.

A

While we figure out whether we even want to do that or not. So that was where we landed there, and that meant that both cloning, as well as snapshots, were unblocked.

F

Right people said they're still yet to be resolved right.

A

So stateful set and generic data populate errs as well as volume source. All three of those discussions are going to happen in conjunction with each other they're very related, so we decided to punt on all three of those yeah.

F

My question is actually for the current model for both Pannonian and the snapshot which storing Worcester is there any documentation, painting people how to do recovery for state faucet from a snapshot? The.

A

Closest thing that we have is Yin ons application snapshot and restore kept and in that cap, basically, what he proposes is backing up the PVC object and then manipulating it when it's being recreated to modify the data source. I, don't think that's a very clean solution, but that's currently what's on the table that.

F

Actually merged or yes,.

A

As far as I can tell that cap is merged.

A

Okay, I'm, so ephemeral, inline volume, support, Patrick or if Michelle happens, to be on the line. Can you give an update.

G

Yes, all of the PRS are in the queue.

A

All right so good progress. There next item is CSI topology, uh sorry uh from where are we CSI topology prep for GA.

G

Yeah, we did not get anything in for 116, but we've started on all of the items, so we should be enriching. Okay,.

A

Alright, next item is CSI driver, crutch, GA, skip attach and pod. In quote this is the CSI driver object, no.

G

Work was done for 116. It sounds like there are some people plan to work on it for 117, though okay.

B

Yeah I'll take a look at this item and we have some people who want to work on this folding 70 good.

A

All right next item is improved: CSI metrics, Sean Michelle. It.

F

Isn't still under review, there were some comments led by Michelle and Rama.

F

Basically, it's a big change, so Kumar's ideas probably to go back and revisit the idea, maybe filepath put that and they are still concerns around how the cache should be managed. Right now is a kind of a little bit message, since this is into n matrix to be recorded. The P is still under review. I. Think this item will just sleep to 1.16 117 during the fact that I was booked, another snapshot, okay,.

A

Sounds good sounds like progress is being made slow and steady, but we're making progress. Next item is the execution hooks kept, which should be managed by sake apps? Does anyone have an update on that.

E

A

Okay, nice is addressing this issue.

A

We had nobody currently looking at it. I don't think that's changed yet. Has it Michelle? No.

G

That's correct, I'm gonna. Add these items to the backlog for the CSI ephemeral volumes, work.

A

Well, next, item is the volume group API I don't believe we made progress here either.

E

So we discussed about it in one of our what a snapshot wasn't good meeting. Okay,.

G

E

We only focused on the first use case, snapshot I, think we need to some discussions on other use cases spreading, maybe even like replication.

A

So this was also designed, which means that we have two.

F

A

A quarter so that sounds good to me. Generic populated data source is kind of in the same boat. No conclusion on this yet needs further discussion and.

A

Will be tied to a discussion around PVC dinosaurs.

A

All right next item are the CSI drivers, NFS I, scuzzy and fiber channel any update on any of these three items.

D

There's no no resource for this. Currently, we need a voucher and.

A

Then Michelle imagines same thing with NFS, correct.

G

A

G

A

Quarter and hopefully we can get volunteers to pick up on these generic CSI drivers and helping them move to move to something. That's much more. Usable next item is post GA and entry topology work, we decided, we won't make progress on this. This quarter. I. Imagine there's no changes. There, Michelle correct.

F

A

So we'll punt that 1.17.

A

Next item is CSI out of tree move flux, volume, remove flux, driver to a separate repo last status update was that the code was migrated over but needed to set up CI. We needed a volunteer for this work. Any updates on that.

A

Okay, I'm, assuming we.

D

A

D

Still need a volunteer for okay.

A

Sounds good and then last item we have is address an issue with PVC created by stateful set not being Auto removed when the PVC is to earth, when the stateful side is deleted, I don't believe we made updates. We had updates here, Gary you're, not on the line. Are you okay, we'll follow up with that?

A

That is all that we have for status, updates going back to the agenda doc, so reminder for any of the features that we are landing. Please make sure to get the docs in as soon as possible.

A

They have to be completed and reviewed by the 9th which is about a week or so, and if you have features that did not land, please make sure to update the enhancement issue to indicate that it did not land and that we're moving into 1.17 that lets the release team know that this is not something that they have to track for the release. So that would be the only follow-up work if a feature did not land, so that seems to be it is there anything else folks wanted to discuss today.

A

All right, then, we're nice and quick. You get half an hour back in your day. Thank you very much and we'll see you in two weeks.

A

D