KubeVirt Community, 20 Jan 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: KubeVirt community meeting 2021-01-20

Description

Recording of the KubeVirt community meeting on 2021-01-20

A

Welcome everybody to the cuber community meeting january 20th 2021 tongue twister there. uh First on the agenda. We have daniel heller, so we'll turn it over to.

A

A

You with us daniel.

A

So the question he had was um with regard to community members and inactivity. How long should we have should the status quo go before we uh reevaluate if they are active or not? Is that what the topic was.

B

Your mute just in.

A

Case so his face popped up momentarily. I wonder if he's having audio trouble.

C

Yes, I guess you're talking about me right. Yes, sir sorry, I don't know why, but my my um audio got muted completely and the volume got turned down completely. So I don't know why but yeah, um I guess you were just talking about the first agenda point.

C

Yeah, what what we discovered recently was that we have. We know someone who is uh um who hasn't responded to a pull request for a while, and therefore the um the discussion turned up, how we should handle those situations when people stop responding to poor request, pull requests when they when they get assigned a pull, request and um yeah.

C

I I think the initial uh discussion um that that we might just remove them from the reviewers section in the owner's file, but um I don't want to decide this on my own or we shouldn't decided we. This should be a community's decision. I think so, I'm asking for uh for feedback from you um how we should how we should handle this so with regards to, for example, that we also try to have a membership onboarding process. We should also have some what is the definition of a continuously active community member right.

A

Good question um ed, I I have been on and off uh distracted and you know borderline myself eligible for being removed at some point. So um I don't. I don't know what a good uh benchmark would be off the top of my head in the sense that uh things can happen. You know and people can be distracted. But what is you know? What is the period of time? That's too long, and somebody shouldn't be necessarily automatically reviewing things.

C

Yeah sure, because fabian, for example, suggested that we should follow the kubernetes activity measure, which I think is quite a bit long for our project. Kubernetes proposes, I think, one and a half years or something after someone is not regarded active anymore, and I think this would rather hurt our project in terms of well. When you get a pull request be signed and no one is responding. I think this. This will especially put off people that are first-time orders. Also, I'm not sure.

A

Well, I yeah and that's a great example for my own personal experience. If I put a push a pull request to kubernetes, I automatically seek somebody out who would be an appropriate reviewer, because I don't assume that automatically assigned reviewers are paying attention, um but then you have to wade through the sea of knowing who's who and and know you know know who to contact, and so I think, that's kind of you know underlines your point.

A

uh 18 months sounds excessive. I agree. I would wonder if six might be a better benchmark and what does the activity measure actually show us uh any participation at all or what what's the the the metric.

C

I haven't seen any exact metric on that, so I didn't. I didn't find any reference on where people are. I think this would also be some some kind of stealthy right, so just uh that people get get looked at for what they're doing in the community- and I think some people might complain about that also. So I think this should rather or this.

C

This might rather be something that that other people are complaining about people not responding to actual to active pings or something maybe- and I myself did a little bit research on how to automate this. um Looking at the github api, I saw that we only have the measure of looking at what people are doing for around 90 days or something or at the maximum of it's the events api I'm talking about, and these there are a lot a couple of 300 events that are that are always the maximum cap.

C

So I don't think that we have any good measure of automating this also so or how to look at it. I don't know.

A

Well, I mean to play devil's advocate uh I'll pick on myself, I'm a great example here of somebody who might not necessarily be incredibly responsive to reviews and I'll fully admit, there's way too many emails from github, and so, if I don't actively review my own review requests uh manually, uh I could miss one and, but I still show up as active because I do make uh pull requests commits. You know, do um do participate, you know.

A

So I don't know if you know as far as automating, I would not show up as somebody who would I I wouldn't be removed automatically because of lack of activity, but at the same time I could do better in terms of responding.

A

So I don't just as a counter example to automating this.

C

Well, I think I think um doing progress or doing doing pushes or something is some kind of activity, and also, if you don't uh commit to a pr- and you just create issues, there's also some kind of activity. I think this all would count and if even if you are responding to some question or something for example in the community chat should also be some kind of activity and to get all this under one hand, could be relatively hard. I think so yeah, but.

A

Yeah so far, this has been a conversation between myself and daniel. Is there anybody else that has perspective or opinions on the matter.

D

So I mean I can share my opinion on this, so there are some projects that you will, for example, will never lose your uh your approval or your maintenance ship. This is happening in other projects, but but maybe the first thing is to differentiate between the two, the reviewer and the approval, so the reviewer, you could lose it pretty quickly. If you did, it responded for, let's say two three months you can be removed because it means that you are usually don't in the automation of sending you the review request.

D

You don't respond to it, so there is no reason it will continue, sending it to you. This is one option and uh the approval that that one is a bit tricky because someone may own some part of the code and that part may not change. So there needs to be a good reason. Someone needs to check if, if it changes enough, so it will not matter and you can set dates like you can say six months or one year or depends on you can have like.

D

Maybe a meeting the maintenance can have a meeting and decide this to clean up everything. Every six months and that's it, I.

A

Had assumed that if you're an approver you're automatically a reviewer, is that wrong? No, but there I think the.

D

Reviewer here so the approver, I think everyone is is, can do the lgtm.

D

The only difference, I think, is that the reviewers will get the notification pro will send them the notification to review it in a random manner, so it makes sense not to have the not to bid in the list. If you are not doing that, we actually I mean, I think that uh in our project I think it's also a problem in kubernetes that not a lot of people that are in that list. Usually it says two people these people are usually not not answering the review, so it is a bit odd, but.

E

I did just just to clarify uh that I got it correctly, so the difference between regular member, a review is just that a review is assigned automatically by the system, as opposed to you know any other that can just initiate html review in terms of access. They have the same capability and access and permissions.

D

Yes, in general, yes, this is what I understood, and this is what I managed to rate, but if maybe there are coaches, but that's okay,.

E

Okay got you. Thank you.

C

I always thought that the reviewer role as it is uh defined in owner's ideas is also. uh If you are part of the reviewers, then you only can then only you can. You can assign an rgtm, and if you are an approver, you can do both, and at least I thought so, but yeah.

D

That's not yeah. That's.

F

Yeah right so yeah, everyone who is in the organization can do well. It looks good to me, but reviewers are explicitly picked because it makes sense with them based on their knowledge or whatever.

D

Yes and it it actually, I mean I'm guessing this, but this is what happens in, maybe in other places that if approval will want, he wants to depend on a reviewer, then it's it's his um I mean he can request or he can wait for a pro for a primary review or something like that for uh the final arc here, final gtm, so everyone can have gtm, but the I guess.

E

The approval can.

D

Say implicitly, that he is waiting for someone specific to look at it.

E

And just a curious question: do we have statistics or or some knowledge on what is the main battle neck? Is it people waiting for review or people waiting for approve after the lgtm.

D

I think it's hard to, I think, in my opinion, it's hard one to know this, because there are many cases when you can get a gtm from someone, but the for example. The maintainer will still wait for another another person that is more familiar with that code together. Okay, I don't know if we can get this easily. I guess this. Maybe we can do like another.

D

We can add another layer that says uh that adds a specific label. If you are a reviewer and you act, maybe that can help, but I'm not that's goes into other regions here too deep.

D

So my summary to to this will say that losing the review, uh the reviewer is, makes more sense to lose it quickly. I mean, let's say two or three months, but losing the approval. It's really dependent on the code. If the code totally changed after two or three months, then I guess it's fine, but if the if this is a, if the person is approval in us in a place that didn't change the last for a long period and maybe he's still interested in being maintaining that, so I mean it's anyway.

D

It needs to be documented and just take a decision.

A

Well, so I think the the ultimate question, though you know we're all agreeing that we know somebody should be removed if they're not uh participating. What mechanism is that is that I think I heard um a committee meeting. I've heard the bot just does it. What do we want to do.

E

I think, as eddie said for reviewer, I think you can go with the bot because you actually do not remove any privilege from the user right. He can still review if he wants to, and so you just remove him from the automation, which is very good because otherwise, if he is not active- and you know, some people who are not well connected in the community can wait a long time with some inactive reviewer, not getting a review right. So it's good that we remove them.

E

Does the bot then re-add.

A

E

I have no clue.

C

Actually about what what are we talking, so I don't have a bot ready. So um I was just thinking about uh automating this and looking at the uh at the gita back there like I said before, and I think we could automate it and if, if we would conclude that um that reviewers would be removed when, after a certain period of time whatever that would be, this, I guess could be automated and would make sense, at least so that we wouldn't have to handle this case manually and for other things regarding community membership.

C

I think this should be discussed in a broader audience. This shouldn't be handled by a bot automatically, or at least that's my opinion on that.

A

All right, so I will take a note, so it sounds like the consensus is that if it can be automated, it should be.

B

So one one comment here: um so the the actual removal implies a pr is- is sent to the owner's file or aliases right. um That could be the opportunity to review that proposal. So whether the pr comes from an automated process or not, uh it's like it provides a chance for uh well the affected people. Let's say to to review the decision or the proposal.

E

That's a good idea: you can put the actual people as reviewer for that piano.

C

Yeah or ping them at least um that that that's a good idea right.

A

Okay, I think that uh the discussion has kind of died off. If there's any follow on, should we refer to the mailing list or how do we want to proceed with this conversation because it doesn't sound like there's a resolution per se?

A

Where is there.

A

All right, if we have any follow-on discussion, let's do that on the mailing list, just so that we don't lose it and next topic we have uh edward haas I'll turn it over to you, sir.

D

So I'm sorry, I didn't add the link. We have a proposal to to introduce srv live migration uh feature and we started with investigating in that regard and had some talks until this document was created, a little bit of work has been started on on pr's, but there is nothing merged or even approved. So I'm just want to share this this work and I hope we can, if, if we have people that will want to comment back, it would be appreciated. My only problem at the moment is a little technical.

D

Is that, like some some reason, I cannot share it with outside the company, which is very odd. Well.

A

The the shortcut there by the way is to share it with the google groups, uh email alias, because anybody that shows up at this meeting has access to it. Then.

D

So I can share the the group meeting and that's it: okay,.

A

Yeah I mean it's, it's not as great as having it just public, but uh it's it's at least open. uh You know, nothing's being hidden, you can anybody that wants can opt in.

D

Okay, so just uh just add it to the to the document as this and I'll I'll share it I'll send a share link.

D

This is the link and then.

D

I'll do it.

A

Thank you, okay. So the upshot is this is a proposal for how to do live migrations if sirov is uh attached to the vm.

D

um It yes, the so if the srv devices are attached to the vm and we start the migration, the proposal is I'll. Summarize it quickly is detaching. The srv devices before the migration is started and reattaching them on the target.

D

After the migration finished it, it mimics the solutions that exist today in openstack and an overt.

A

Okay, so at this point, we're basically looking for people to review and provide feedback.

G

H

I I had a question. Basically, I had submitted a ticket with the issue. My name is rupesh and I think we discussed the same issue last week. Also, I wanted to see if I can get the support on this, maybe like with a donation to the community or with somebody giving me some um time for helping me fix the issue.

H

Is that possible somehow.

A

uh I uh am recall the the proposal from last week. uh Could you refresh your memory.

H

So I'm trying to create a vm and uh I'm using k3s and um with k3s and qbot we were able to install successfully all the pods are running fine. But when I'm trying to create a vm, it's not going and communicating with the k3s api server.

H

So that is where the issue is coming.

F

ah So yeah, due to the the webhook not being resolved by the area.

I

F

H

F

Cert manager, they have the same issue and it's also not resolved so they what they do. There is disabling dessert manager. I guess we're always doing. K3S should have a look there, so we should probably open an issue there yeah and ask them.

J

By the way I mentioned last time, I'm gonna try my k3s and I tried two days ago and I couldn't install keyboard because I'm running on a3os, which is a very small operating system, they provide and it doesn't support kvm.

F

Okay but yeah, then it's at least clear what needs to be fixed, yeah.

J

But that's specific to the operating system I use. I I I I'm going to try on a through k3 is when I have time. Yeah.

F

Sounds great but still I mean there there seem to be issues with web hooks in general on k3s, so I don't know. Maybe we should check the repos there and see if they have something open already. Regarding today,.

A

Okay, I was just taking a note there, um so I don't know if the we can make any immediate promises. Other than to you know, look at the the api they have and see. If there's some sort of insight, we can help with so.

H

If, if somebody can help me with the with like a few hours of their time, um I would like to compensate also with with the community donation or something or directly to the person. How do you want to do it? But if that is possible, I would ever love that. Please.

B

This was especially you you, I if I remember correctly, you raised an issue: do you have a link that you could share.

H

Ok, let me give you that.

B

A

Okay, um in the meantime, should we move on with other.

A

Discussions all right. Next up, we have chris to uh propose uh status of version branches in the user guide.

K

Hi everyone um I was thinking about enabling uh versioning in the the make docs migration for uh cooper documentation and we notice there are a number of uh old release branches in the user guide repo.

K

I was just wondering uh who creates those branches, and uh when do we get rid of them and are any of these existing branches still relevant.

F

That's a good question, so I've reviewed a lot of user guide prs and then I haven't been aware of this. Even so, I'm not sure if someone is really using that.

A

Yeah uh so upstream documentation for the project in general, uh I was, I didn't know we were versioning it. That's a very good.

F

0 14 is the last one. I know 0 19.. You can delete all of them. Chris yeah just delete them. They have no relevance.

A

Yeah, I believe that's from two years ago, at this point so excellent.

K

Like I, like the sound of that answer,.

A

C

A

But that's a good point going forward, um so we let's we're on release 37, if I recall right now, is that this month's release, uh if we were to start versioning the documentation from this point forward, how long should we have them go back.

L

We shouldn't, I don't think we should version it. I I can't think of a reason to version it, because everything is forwards or backwards compatible like what would be the things aren't really changing uh as we move forward like we're. Not gonna have.

A

Well, I can play devil's advocate and say that you can at least know when something was introduced. Yeah even.

K

If things have changed, things have changed and uh and the things that were changed weren't fully changed um in particularly uh the host disk to empty disk migration.

L

There's multiple dimensions to the user guide as well: it's not just cubert, so I don't know what we would version uh I mean. So when we look at storage, for example, there's cdi features that might not correlate directly to the cube, really, I'm not sure how to version it. What would we anchor it to.

A

Good point, because if we did it by api version that comes back to what you just said, you know we're still on the same api version. We were a year ago because it's backwards compatible.

A

And if you anchored it on um bran, you know whatever the stable branches. Once again now now we run into a problem with what what extra information did we actually share by versioning it? I don't know.

K

Well, I'm totally fine not having versioning, because that's a lot less work on my plate.

B

Yeah and and if if something happens, that something changes that let's say for whatever reason breaks compatibility, this should actually become part of the documentation. This should be documented and not you know hidden in our branch. So um so it should become a new item either well somewhere in the documentation. So there wouldn't be a point to keep a version with and without the break, make sense.

A

Yes, I I'm conflicted, because I I can sort of see a use for it, but I can't think of how we would sanely do that.

A

So the consensus that I heard at this point, though, was no versioning necessary for now.

K

Yeah, that's what I gather stu, who.

M

Thinks who is going to use versioning of the dogs.

K

um I see many other projects on uh on the internet that do versioning, um so that was just a proposed feature with the user guide migration to make docs.

A

Okay, um I guess the yeah, the the consensus is that we won't version for now and we can always revisit if it turns out. We can find a good reason to do so there by.

D

D

The example can be rel or rev everyone, they have every version they have. There is a document and it's under the version. It's like a branch.

K

D

Yes, I don't know, I just know the one I usually reach real at some point, but yes, maybe everyone has because because like if, if someone has, I think the reasoning is here. That is, if you have version x then then you have this feature and if you move to version x plus one you have additional feature or some features that were taken out and stood, there needs to be totally different. The other approaches is that for each feature that you add, you can say deprecated inversion or added inversion.

D

So it's there are two approaches here.

J

I I know the kubernetes stocks are versioned, but it's because stuff changes quite a lot in the docks and people want to go back. Sometimes I think.

J

But I don't know if it's necessary right now.

A

Okay, I think that the the consensus is not necessary at this time, but we can. We can change if we need so with that. Are there any other topics uh for the agenda today before we move to the open floor all right pep the floor is your sir.

B

Thanks yeah, I.

M

Just wanted to uh maybe.

B

M

It before, before I joined like five weeks ago, a stu shared a draft about the role or the membership in the community, a document and there there was some discussion about it. But many like.

B

Yes, and that that was sorry so that that that was actually discussed as part of the first topic that daniel raised on one of the aspects of this document, which is how to kind of promote. uh But um your question was like.

M

When can we make it official, maybe maybe like who owns it, and when can we make it official.

B

That's a very good question and slightly related to this and to the topic that I wanted to mention, which is the keyboard summit. I was uh going to say that I was trying to uh add a session. uh You know we have to keep it something coming I'll talk about this, but what about uh suggesting a topic on how you know how to contribute to kubert and cover topics like the process of you know becoming a contributor and staying a contributor?

B

I support the discussion before and maybe we get said that you know that discussion has a deadline for reaching an agreement uh like after the the summit. Does that make sense.

A

Did you mean to then discuss membership as a topic in the summit.

B

That could be an option. I mean as a potential topic to discuss. um It seems.

M

A bit boring to me.

B

Well, contributing to the project is not boring. No.

M

No contributing is interesting, they're setting up like who are the different roles and like to meet something that we should just sit finish publish I wouldn't like to spend time on it in the summit about this way. I want to hear about interesting technical. uh I.

B

Then then yeah I mean, um should we say that, like let's say before the summit, so this can be kind of presented at the summit as uh yeah.

E

I I yeah, I think that presenting- or that is a nice nothing, so you can announce and and describe the roles and how you can move from world to all. um So so what? What do you suggest just adding comments to the document? What is the procedure to kind of discuss? It.

A

E

A

Quite a few, and since we've brought it up, let me find the document. uh There are quite a few uh comments already on it and I don't have it. Unfortunately, here I don't know what happened to I closed the tab, uh so what we basically need to do is take ownership of it and um you know and commit to some of the comments that have been made at this point.

M

Isn't it used to that they own this document.

A

Yes, uh there was some uh talk of uh chris getting involved, but I don't know if that came to fruition, so I can keep owning it.

A

So yes, dan, I will take the document sometime between now and the next community meeting and incorporate comments have been made and then we can bring forth a proposal again next week.

M

Can declare it as golden and then everybody comply with it sure. Maybe we can translate it to md put it in a.

H

M

Or something or a link.

J

M

What do you say what what's the proper way to publish it like just say it is now uh golden or to push it as commit to become yeah, contributing.

B

Page, this should be in the in the repos. uh It should be also mentioned in the in the user guide, there's a section for contributors, so that should be highlighted there.

A

Okay, so we need to convert, we need to incorporate comments made and make a pr out of it pr to the community side or pr to the code base.

B

I would say in the community side and also on the user guide, at least a pointer to it. So.

C

Oh just just uh not not completely related, but I was just I forgot an update. We just implemented um on the dev stats. We we implemented repository grouping and I added a I added a comment: a commit to the qubit community, where I uh created a six yama. um Maybe if any one of you is interested um can have a look at that just um just for a quick heads up, sorry just for for overloading everything here but yeah. I think this was just worth a mention. Probably.

C

Awesome. Thank you.

A

Okay and with that, I think that we have concluded the discussion that turning it back over to you pep to talk about keyword, summit.

B

Yeah thanks daniel for raising that, because I actually saw that commit- and I was honestly wondering the details. But anyway.

H

B

Moving forward because um yeah the tuber summit now has a landing page. uh It's linked there from the notes. We are still accepting uh proposals for topics to contribute. uh There are. There have been four submissions so far and a few more people in showing interest and working on proposals, but uh don't be shy. You you're still on time um just mentioned. I think we will submit a new topic uh to talk about uh contributing to the project and well um I just wanted to to to mention this and that's it basically.

D

The the deadline is, is.

B

This weekend this week, yes, uh we aim to have uh the proposals this week, so we can work on a schedule through the you know by the end of the month and and have it uh available. So people can prepare their agendas. Let's say and.

B

We also got access to the online platform um yesterday and we're starting to work on. Let's say the logistics of it and more details on on the actual logistics will come soon.

A

All right, thank you and roman. You had a note about uh merging in github.

F

uh Yeah we are already for one or two weeks in a pretty bad shape, due to a new random operator bug which we ran into, and I created a pr for it to fix it to suddenly to finally make it possible again to merge prs automatically.

F

uh I guess you have all been bugged by the failure. um Also to keep you deaf about it and then the 0 37 0 release came out and some more operator related things broke. I have it all fixed now and as we speak, all tests were green, but something else got merged in between so it's re-testing now, but but it will be fixed uh soon. The pr fixes it now because the last test turned green yeah. I will update you on kubrick devil once it's should be within a few hours today.

A

Okay, um as far as the bug scrub uh is anything going on with that currently peter.

N

Yeah we had the first station last week and I think it went quite quite well. So if nobody minds, we can continue with that for the next 14 minutes before we disambar.

A

N

All right all right thanks! uh Let me just share my screen.

N

All right, we got uh plenty of like ci reports. If nobody minds, I would skip them and go to the issues.

N

That may be more interesting for the little time that we have left. So, let's start with this one configmap an emulation variable is incorrect. Question mark it was opened yesterday and the issue is.

N

That a vm but stay spending and we fail to console into it.

N

That's not much to work with.

N

N

And it speaks about.

N

Emulation um shall we just ask for uh more logs there, because uh I can't figure out uh what's wrong just how interesting.

M

N

M

Been changing the variable names in data got the test vm running.

F

uh Where do you read it.

M

ah Okay, they just enabled the emulation okay. So probably the host does not support uh virtualization.

L

H

They saying that emulation.

L

Didn't get enabled sorry, I kind of was.

L

N

I think it's just maybe misleading. So as you say, then it did not work without the emulation it works with it. So I guess the question would be.

N

N

N

And let's leave it there.

N

All right: okay, let's see what they say, I'm moving on to this um drop liver demon, driver storage for the bird 690.

N

As of the slipper, the camo driver no longer answer depends on the storage driver and it should be possible to stop including the package and containers once a new revert is imported.

N

um So we are just waiting for the newer version of flipboard.

N

um And I recall there were some changes related to the libert shipping and studio version. I could anybody shed some light.

L

N

uh Are we already on with verse 6.9 or are we still waiting for it.

A

You know this brings up a different question uh upstream. We have a funny uh way of consuming libert in the sense that the images are custom built for us. If I recall- um and this is a topic that I was probably going to defer till next week- to actually propose, but it may be that we want to explore moving to centos stream uh instead as a base image uh rationale being that that's what livford is actually basing their work on upstream. So it would help us uh as a community to to consume that.

A

However, there's some technical issues under you know the surface of that that make it easier said than done and there's some side effects. So I know we've talked about it in the past couple months, but uh no, I don't think we're using six nine. Yet I I don't recall which we are, though we're on 6.6 right now.

L

Yeah, the discussion about centos dreams is something we we need to follow up on. I'm not sure if we have the right people in this uh meeting to really move forward with that, because we need to.

L

We really need the people that are maintaining those packages to kind of give us uh guidance on when they think that they could start supporting that and giving us what we needed. The cadence that we needed.

N

Okay, so let's treat this as a tricker and and a reminder that once we switch to six nine, we can address.

N

N

All right um next, one.

N

Admission blocks integers and cpu and memory resources.

L

I thought we fixed that uh let's see what version they're using.

A

um We did not fix it, um so the issue here is that we broke it inadvertently by consuming a newer version of the open api spec. If I recall lubo, who is his name just popped up on screen, uh does have a pr for it. That's in the works.

N

Okay, awesome, so it's uh yeah, it seems it's being at rest, all right and then let's skip.

O

N

N

ah This one is from ezra uh ezra. Do you want to uh present this one.

E

uh Nothing much to introduce it's just that I update some document it's under the docs directory and I had to go through all I talked with daniel about it. If there is anything, I think there should be an easy way to exclude directories and even in the future file types from just running all the tests, it also consume a lot of testing resources from all of us. So there is no reason if someone is changing a readme file that you run all the tests right.

C

I think the simplest approach on that would just be the using the run if change with a with an expression that just matches markdown, but I'm not sure if that would work.

N

Don't we need the the opposite of it too much everything but smart jump files.

C

Yeah exactly and the problem is with that- I'm not an expert on regular expressions, but I I found always um writing in a regular expression that matches everything, but something is hard right. So I'm not sure if we can easily do this.

E

I basically just wanted to raise the flag. You know it's not like. It seems like very crucial. It's just you know it it first. You know the impression is not so good that you know I'm changing your admin. Suddenly I see that all the tests are running and second, is that we are consuming a lot of resources which are scarce. I think because tests are taking a lot of time these days.

N

It would make a lot of sense. Is there anybody who would uh like to give it a try? I mean. Maybe we see that it really produces a huge red x and it doesn't it's not really worth it, but maybe it won't be that difficult.

C

So if you know just just assign it to me so it's I will, I will have a look at it.

N

Thanks thanks daniel.

N

Yeah- and I mean yeah, if you see that it's too much work, we can just market this one's resolved. I guess okay um issue with extra disks behaving differently after o36 update it was opened a week ago after an update from 35-46. I noticed issues with my v: I'm not able to properly format month, primary and secondary.

N

N

This some series, although I don't know much about storage, uh do we have anybody uh that place, runs with storage and convert here.

N

There we go, do we not.

N

I think that michael wouldn't be mad. If we market him here, would he.

N

G

M

O

M

A way to assign it to a sub group like for storage, organizing, a topic.

N

um We we do have uh sig labels, uh but I'm not sure if they are being used for scrumming by the interest groups. So.

M

But maybe we should like to to make it easier to for people who want to say I I want to fix a storage bag today. How can he easily oh yeah.

N

We can label it uh for sure, um I'm not sure if it would make it easier for uh somebody that works on the storage like it won't make easier for us to find somebody, because they.

B

N

um I think it's more efficient if we just connect them directly but yeah. Let's mark.

M

It, of course it's not being said it's just like if somebody wants to look at all the storage stuff or the network stuff.

N

Okay, the next one was the enability set. The error policy. Alibert has the field error policy, and this defines how qmo is going to treat io errors, more details and deliver.

I

Them so if you are interested, I can summarize the issue.

N

Oh yeah that'd be great. Let's go ahead.

I

um Yeah, so actually I was working on a feature keyboard feature and to generate kubernetes events on io error, and actually I tried to reproduce an io error on keyword and I was not successful and then actually that turned it out to be because kuber doesn't set error policy to stop, but it's by default set to report.

I

So basically the io error are just reported on the gas corner, so this depends. How do you want to manage this kind of error um so link to this uh issue? That is also my work in progress. Pr.

I

If the error policy set to stop, then the the virtual machine is paused and with the pr reopen you can get a kubernetes event. So basically, that's the source of this issue.

L

How did you reproduce the I o errors like well, it sounds.

I

Like yeah, you can see, I just create a buggy uh disk and this is with vm setup. You have that there, so I just simply have one one node cluster and then I just create a pvc that reference this. You have the yaml down in the issue and then I was able to to get the vm post. So that's in the second comment: if you want to see the setup.

N

um I wonder if this would make sense to have exposed in the convert api, or is it more like a backdoor.

G

N

Can do through the um through the hook? Maybe.

I

So I actually had a offline discussion with roman, I'm not sure if he's still there, but anyway, it's just suggest to set the other policy to stop and then continue with the implementation of the kubernetes event. So but yeah that could be configurable or.

M

Would you like to configure it pair a disk or on the entire vm.

I

So in liberty is per disk, but I think it makes sense just to do it per vm.

I

So right now it just hardcoded. So.

I

Yeah another issue I have is actually to associate the right so the disc, to the pvc- um that's something I haven't succeeded yet so so yeah that that's basically what what it's about and if you scroll down you can see there your event that got generated so basically yeah right. There.

L

I

L

Events, uh but we don't stop the the vm is that accurate, yeah.

I

Yes, so basically is the user that get the io error, so it's the gas kernel that report you error and that could be, for example, okay, in my case I had a faulty device, but that could be also not enough space and the vm is not paused right now, because the error policy set to report.

M

I believe that uh in over the default was to stop yes,.

I

That's correct.

M

Yes, that's correct, and- uh and I wonder why do we need the knob I mean can't we just flip to stop uh I'm just trying to to think. Where should we expose it to the benefit of our users.

I

I'm not sure that's why roman suggested just simply decide to stop. The thing is today: cuber doesn't set it, and the river behavior is to report.

M

Okay, okay, I misunderstood. They thought that you are exposing a new knob you're, just not.

I

M

I

I'm just sleeping yeah.

L

I would only expose the knob if we have a use case where somebody would want to continue through an I o error. So if we can't think of a reason to do that, then we should just set the stop.

M

All right it works for me and then, if somebody really wants they can continue through the keyboard.

K

M

And do restart. I.

I

Mean you can have a look to my first commits and that simply said your policy to stop. That's just something, and the second commit just include a generation of io error. So.

I

That's the reason for the issue. Sorry, I took too long, probably.

M

Not at all, excuse me if the vm has stopped. How do we con continue? It is the revealed cattle command to continue it.

I

I haven't tried one pulse because my my setup just is a faulty device. It will be interested to try to probably get a full device and then try to decrease and free some space and see what is the behavior I haven't tried. Yet I can try.

M

I I suspect that it would stop again if, if it's still broken and it tries if the guest tries again to to do another io operation,.

I

Yeah, so the best practice would be that some somebody externally fixed the issue with the device, because rivert is not able to do that and once the e issue has been fixed, then you can just unpost the vm, but that that's something I haven't tried. I just tried to generate some io error, but probably it's interesting too.

M

I'm just thinking that for for the complete story that we are giving the user, uh I would say that your pr should say hey if I'm introducing new behavior that gets vms to stop. If your vm stopped fix the io and do this weird cattle command to to unblock yourself.

O

Yeah I can, I can try, I can try it with weird cattle. I haven't tried yet.

L

One thought here: if somebody has their run strategy set to always they could potentially get into a boot loop. I guess where, if it's at iowa, uh they would just stop the virtual machine and then it would immediately get recreated like back into a running state and then it would happen again and then we could pause.

I

But yet so currently, that's exactly what has been described so I'm basically booting a virtual machine with a buggy device. So it's not reported that is paused. You can just see that in the log of launcher, but you really need to check it, and so it's not so easy to to to catch that. That problem with the uh kubernetes event, at least if something is going wrong. It's reported by the kubernetes event. So that's exactly what the pr is trying to do.

A

Okay, uh we are two minutes over time at this point,.

C

A

I know thank you for bringing it up um peter. Do you have anything else? We should cover.

N

um No, I mean we are, we have over 100 issues described with, but we want.

H

To do all of them today can we can we discuss on.

N

N

N

um Yeah I found uh that one of the other issues uh it's slightly older, reported a workaround for this. um Did you get a chance to try it.

H

Okay, okay, I will, I will look into it. Sorry, I might have missed it, but I will. I will look into that um suggestion.

N

Thanks great, um I think we are good one and uh just maybe one last thing ls. The is this. ah This is not ready to review so once it has uh police being the assignees, I guess uh to move it forward.

I

Yeah the the thing is, um that would be nice if you add the hint how to associate the pvc to the to the disc because yeah, that is in the description. So I don't want to bother everybody, um but I have a. I would like to associate the disk with the pvc and that's actually a bit tricky in the code to get the reference, because the vmi is not up updated with the information, so not sure if it's worthy or if mentioning the device is enough.

I

So if you have any suggestion that will be very useful.

N

um Is there anybody who could help with it.

A

H

A

Forward may end up being well the way it's written right now. I think it would end up being a synchronous call to look up that association, but that's not great.

I

Yeah and even querying kubernetes, my client, I even don't think we are launcher as the privilege to do that.

A

Oh you're, absolutely launcher does not have the privilege to do that. Yeah.

I

That's exactly what I suspected anyway, that's the special note to describe what my current problem.

N

um Can we follow up on the convert.

I

Yeah, that would be a great story and just take too long.

N

No worries all right: I'm stopping the screen share.

N

Thanks all for participating have a great week.

B

Yeah, thank you.

B