Kubernetes Storage SIG, 12 Mar 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Storage 20200312

Description

Meeting of Kubernetes Storage Special-Interest-Group (SIG) Workgroup - 13 March 2020

Meeting Notes/Agenda: -
https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.ollcun95m49j

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

Moderator: Xing Yang (VMware)

A

All right, hello, everyone today is March 12th 2020. This is the kubernetes in six doors. Meeting I'm, Shin, Young I'll be moderating the meeting today inside. Please show me if I missed anything.

A

Thank you. The meeting will be recorded and uploaded to YouTube. So let's get started. I have shot my screen here, I put the window 18 newly scheduled here. Just as a reminder. We just had code freeze much fifth last week and the next deadline is the doctor deadline. Next Monday, March, 16 and then GA.

A

The release date is March 24th. Let's take a look at the planning spreadsheet. The first one is raw block. Do we have Yun here I.

B

Am just waiting for.

A

Everything else is done. Okay, great feature.

B

Is yeah? Okay,.

A

A

Next one is a cloning TJ Yong.

B

Yong, so this is Jenny Burton at ease. There is one remaining issued issue in external provisioner and average working on it and again, documentation is under review.

A

Thanks young and the next one, it's a CSI migration. Next steps. We have deep chanted here to give an update for someone else not giving update on this one. So.

C

I wanted to give a quick update about David. He ended up leaving last week was his last week with our team and he's gone to a fin tech company, so he's probably not going to be working on kubernetes storage.

C

We're gonna have to sync up with Chang and deep to see what the status of this is and get someone else to help out here for the next quarter. Okay,.

D

Hello yeah, oh yeah, I, think I have yeah those contacts currently on this from a from a framework point of view, I think there are no outstanding issues that neatly there might be one the minor outstanding issue, but for the most part, what were the phase were in right now? Is we're just trying to get all the cloud providers to the start, implementing and getting testing and getting to beta? Okay.

A

A

Next, once this I online offline resizing issues kurmond its come on here,.

B

A

Update okay, oh.

A

I seen her as a I think come on submitted a PR in extra resizer to support CSI 1.2 I. Think it's just looks like just wait for him to update that PR.

A

Okay, well get some more update from comment on that. Next one is the support, continued Rises and note plugging in Windows with storage proxy. You can't even give update on this one D. This.

E

Is the this is KK? Okay,.

A

E

Your disk changes are much right now. So there's a few, you know error paths, kind of like scenarios which we are still fixing, but in general it's working, GCPD work currently is getting tested and we are going through that process of getting to a point where everything is working. There are a few more patches which got merged in CSI proxy, but in general we are at a phase where things are all working. It's just up to the point where one rotating release happens, we'll keep making debugging and other fixes. Okay,.

A

A

Snapshot, okay snapshot, so we continue to make back fixes so I'm still working on the there's, a great snapshot. Time out, you should Adam still, you know it still testing and fixing, and then we are also updating the talk for the was the hostess driver stop date.

A

The next one is a non-recursive rolling ownership. Do we get an update on this.

B

Man, so the feature has been merged as.

B

A

B

A

Great so this one is complete, then that's.

B

D

I think the only thing outstanding is the docs for this one. Okay, that's good.

A

Secure Linux across information handling. Do we have an update on this one.

A

Mission handling for Windows. Do you have any update from deep.

A

Shadow, you know anything about this, sorry which item this is the file permission, handling for Windows, I. Think.

D

It has not started and probably won't start this Lee in the face. Okay, we'll have to maybe maybe KK. Can you potentially bring this up in the next Windows CSI meeting? Yes,.

D

E

A

This one is CSI inch. We read only handling. Do you have any update on this one Chris.

A

A

See the driver escaped 14 for GA young Christian did.

F

B

Yes and I think you also have to documentation this time. Yes, okay,.

F

G

Have already been merged, Denny's website PR is outstanding, but that's it.

A

Website but that's that's a doc right website. That's the doc! Another dog.

G

Si si o si.

A

Si dog you, sis, okay, all right.

G

Sorry, the CSI documentation has been updated emerged in the county's website. Really something release notes have has not been merged. Yet. Okay,.

A

G

A

Need is most obviously.

A

Feature itself smudged great mix, and next one you prove matrix, I, think that's it done next. One issues related to listen volumes are lock points except there's. No one working on this is there an update. Michelle, do you know anything.

D

About this one, so Andy actually has a PR out to remove the check-in, node, publish and Stage four.

E

D

Volumes are now points we were hoping to get that into once 18, but I think there were some potentially backwards compatibility concerns. So we decided it for 117 we're going to add a deprecation notice to the release, notes for this behavior and then make the change in 118 or sorry 119.

A

Thanks next 130 pause, so this is Patrick. It's Patrick, you know call.

A

Okay, you show: do you know if there's any update.

D

On this one I haven't heard any I know. um Patrick has been mostly focusing on fixing some of the bugs in the current code base and I think with respect to like revisiting the design proposal. I, don't think he's taking a look at that.

A

So I do you have a cap out that has not got many review comments. I promise you to schedule a review meeting for that for that uniform.

A

Next, one CSI out of tree move in a fast driver, Michelle.

D

um So the NFS CSI driver has actually been making a lot of good progress. There have been two people that have been getting the repo back up to state and testing it out. I think we should be very close to actually cutting a release for it in the next month or so.

D

On the ice cozy front, there was one person that was that did some initial work on it. I need to follow up with them to figure out the rest of the steps to be able to get get to it. Release.

A

D

Both of these two, we did not really get volunteers to work on it. Although I did see, one person make some activity on the Flex repo, so I will try to follow up with them offline to see if okay, it's something they're interested in working on, but at least from the fiber channel point of view. There is nobody working on this yeah.

A

Next, one is duplication of the external storage repo that being started.

D

Yeah I think this one is on Matt's plate, but he might not be aware of that.

F

D

B

D

There was actually someone today that expressed interest in continuing to maintain the NFS provisioner, so we should think.

C

Of iran's message saying that he wanted to make sure that this NFS provisioner was not deleted. Did somebody volunteer to pick it up? I think.

D

He might be volunteering to help maintain it. There's.

B

D

Thread going on right now, without it, okay.

B

B

D

Should move it yep.

E

D

I think that's the first like regardless this whole repo is getting deprecated, but if there's anything in there that we want to actually continue to maintain, then we should pull it out and create a new Bebo for it.

F

H

Are cert sets are Darley I'm.

A

A

You're asking who is the owning it? Who is it working on it? Let me show, do you have a name.

D

No, oh, so I think we need to that's what we need to figure out like if we do want to continue to maintain this. Who are the people that are volunteering to be the maintainer --zz? Okay, hey.

H

Michelle, maybe you and I can sync up after this call. This is John terpstra del okay,.

D

Sounds good yeah just send me a message on slack.

A

Thanks June XY is willing snap Shani transfer.

I

I

Again, I just have one or two high-level questions: I'd like to get feedback on I put down at the end meeting. Today, that's cool yeah shouldn't be too.

F

Super high-level questions: okay, thanks.

A

Next is what, in a house so Nick has started doing development. He has a PR that get all the release to set up he's working on the controller and the agent side changes the sitter's back. That is being reviewed. I think we just have one last issue that we need to get resolved have a meeting later today, 10:00 Pacific 10:00 today to review that yep.

J

K

A

Yeah thanks and next one is generic data populate er. It's been.

A

So I, let me show you want to give an update on this one yeah.

D

So the feature gate work is in.

L

Yet this feature is: is merged, the docks are done, michelle has reviewed them and I, don't know what it's gonna take to get the docks completely merged Michelle. Do we need a second reviewer or someone special I.

D

Think Doc's team will get around to it. Okay,.

L

Yeah, so so we decided to do was to not because the feature is basically experimental not to provide a lot of detail and user adoption just to basically mention the feature Gabe and explain what it does, but not explain how to use it, because we don't really want end users trying to use this feature until we have actual data populate errs to do things with mm.

A

L

Very minimal Doc's, okay.

A

Yeah, okay and then you are still count the POC populate er. Yes,.

L

Yes, well, we need a design that addresses the the known shortcomings of this approach and that's on me. Okay,.

A

L

A

A

Next, why is the object? Storage, API? You have Jeff Oh John.

J

Hi John here we have a review of the cap yesterday, Jeff and I are making updates in the background on a working branch that will push up today. I posted a recording in the first meeting to the six storage, Google, Group and I've got the second recording more dating a pony.

J

I've also sent out a.

A

J

A

The next one, the one expansion for a full set, see guys need to find an owner. I, don't know if we have over units. Does anyone know okay, Kay? Do you know anything about this? One yeah.

E

So I think the update from the last meeting has sort of disappeared. The thing is see: gaps has shown interest in reviewing it. Somebody needs to own it and proceed. The idea was that after 1.18 code, freeze, I'll have look at it and tried to revive the cap, so.

A

You will be voluntary.

E

Right now, I don't see anything.

A

Easily: okay, great thanks.

A

Okay: next one address issues PVC created by stiffest I will not be altered. Mood is in a call today. I was just checking with him the other day he said he said, he's still interested in Wakanda, so he's still looking at putting together cap. So.

A

And next one execution hook, so this one we did review it in the data protection good meeting in the last professional meetings, so I updated the the new design based on the feedback. I think we need to check with the API reviewer just to make sure that this new approach you know adding this Hotei in the container is the one that they wanted, because in addition to adding this new container, we also added a new API object.

A

Entry, which will be, which is usually difficult, so well need to ask them if this is the way they want to go, because if we don't, if we don't do that, then there's no way to tell people. It went you wrong this hook.

A

There is another alternative which is kind of like a hybrid approach, which is not introduce a entry object with just adding that using the the handler adding that in a country in the container. But then we need to trigger that call for an external controller, so either way, there's pros and cons whichever we choose to do so. I think you need to get some get some guidance from that. You can reviewers on this one I think that's all we have in this spreadsheet.

A

Okay, Mike, you have this item here.

I

Sure so, basically for those they haven't. Look at this. Yet there's this namespace transfer, API high-level. There is a transfer request, object that is made in the target. Namespace may transfer approval object source the way things are currently in the cabinet once a controller kind of matches, a request with approval. It will begin the process of you know: updating the PVE, creating the new PVC, deleting the old PVC.

I

John Griffith is suggesting leveraging the data volume I mean the data source field and using the namespace transfer request as the the kind there, so that basically user would do the request for approval thing and then, in addition to that, create percent.

I

The main reason I didn't I didn't do that in the first place was just because not all probationary supported the datasource, but obviously that's probably maybe just a temporary concern I totally get that it could fit. So it's what I'm wondering is what people think about leveraging the data source feel for this.

I

For this feature, so.

A

Making like us so making a third data source like, in addition to able to snap shot and the PBC. This is.

I

A

Would be another one? Is it you.

I

Do the transfer request the approval and then, when you create the user, would create the target PVC with the data for the data source field and the kind the request? And then the controller would see that, and you know, update the PV to be basically update the PV and delete the source PC.

M

And part of that is because keep in mind you know, there's been ease PR to remove the restrictions on data source, so there won't be just three things that you can do anymore.

M

He just from a from a workflow perspective, I thought it was I thought it would be a good, consistent, workflow in terms of creating volumes from other things, to follow that same pattern. But if I'm, the only one who feels that way, I definitely wouldn't.

M

I

It totally makes sense to me it's just you know, you know. Another thing, I think mention in the PR was that you can get the user can get feedback on like a quota violation right away. Otherwise it will happen kind of in the background.

I

Nit yeah I think you know it's basically do you want the user to create, create the PC I? Think just do you just do one more thing and when the new PVC that they create also kind of I, don't know the spec has to match the source. So I wonder you have to go with that.

I

The you know, the spec of the target PVC would have to match the spec of the source. Well,.

M

Yeah, that's good point. I was actually thinking there. I guess it might not be that easy. I was actually thinking you could. You could just implicitly.

M

Use that but I guess since the since the object is going to be you're not going to be able to update that object. So that could be a problem.

M

A

And this will be handled by a new controller or the provisioner. The sort of provisioner which I'm.

I

Mentioning a new controller like similar to I guess maybe like the common snapshot.

A

Okay, this is an e handle, the PVCs transfer and not snapshots right was it also for snapshots yeah.

I

It could you know the API is for any, you know, storage related stuff, so it could be for snapshots or I think it could be. You know the API should support and snapshots as well. Okay,.

A

I

Haven't tested out actually the implementation of that yet I assume it's smaller yeah.

A

M

Yeah I'd be curious to get thoughts from Michelle on saw and others.

D

It might be worth can we summarize this in an issue somewhere, and maybe we can have a meeting to discuss this.

I

I

Sure why, but is supposed to be like a hub issue or a like just a Google Doc I can send that to the group.

K

It's not what kind of issue are you envisioning.

D

Maybe a github issue, I.

D

Think not everyone may have yeah or I mean I, don't have well I. Guess it's easier to comment on stuff in Google. Docs is.

C

There, an enhancement issue already open.

I

A

I mean here Yuli should have something right. Well, maybe this one hasn't attached to that stage, a game transfer. Listen! Is this an issue? 6:43 I, hope? Oh, that's, a cool doesn't issue or it's a look at yourself. Maybe it's a tap! Oh, that's the chance craft. Not you probably don't, have a issue yet.

C

But uh yeah I agree with Michelle: let's have a meeting to kind of go through this design and and figure out what the next steps are.

I

Okay, thank you.

A

Okay, thanks Mike Jung.

A

Okay, do you have any other issues to talk today.

E

This is sticky here. I had one question really so as part of fun rotating release when we are planning to do CSA proxy binaries themself getting released. What's the general process, we have seen documentation on sidecar releases and stuff from CSI, but a binary release watch the process, and is there something similar to that this that we have done in the past.

D

Sure and sorry you're asking about releasing the CSI proxy. Yes,.

F

D

So I think Jing actually asked me about that too, and I sent.

E

D

A link under CSI release tools. We there's a page they're called sidecar release process yep.

E

D

It documents, like all of the steps that we usually need to do, to prepare and then actually make a release.

E

Okay, so it's so, we should go with the equivalent sidecar kind of release, even though it's like a binary release. Yeah.

D

E

D

Things are going to apply because this is a binary and not a not a container and I I. Don't are we actually hosting any binaries anywhere? Or is this just like a code release so individual.

E

Cloud providers will have to host the binaries, no matter what we could think about github itself, having a release which has binaries, but that's one part of the process which I wanted to clarify with it. Yeah.

D

So I think for now, because in general, just building like Windows, binaries and containers is not a really well supported thing in kubernetes, we'll probably just have to do code release for now, but longer term. We should work with the kubernetes work. What is it called infra work group to figure out how to solve this cool thanks thanks.

A

Thank you that there are other issues to discuss.

A

If not, let's see that is for today thank.

J

A

H