KubeVirt Community, 13 Jan 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: KubeVirt community meeting 2021-01-13

Description

Recording of the KubeVirt community meeting held on 2021-01-13

A

Okay welcome everyone. This is thecube community meeting on 13th of january 2021. um The first item on today's agenda uh is, I believe, from david okay.

B

Yeah, um okay, so I'm trying to cut the version 0.37 release and when I did that, I hit a error publishing to the quay application registry, and I didn't even know uh exactly what this was, what we were doing with it. But it's um it's related to the olam bundle.

B

I don't know who's actually using this. It might not be anyone, but uh we are in a situation where I don't think we can continue publishing there. Now that we've incremented to our version, one due to some kind of strange publishing, artifacts that are persistent from previous releases that invalidate our csvs.

C

Can we somehow verify if this even works? I mean, I suspect, also that no one is using it. I don't think that we ever updated. It then. Actually, I think we're just pushing this, but not updating it, then, on the visible end or something so, I'm not sure if this can't even consume, but it will be pushed.

B

I don't even know how it's consumed. I don't see it advertised anywhere, I'm not sure what it is to be honest, uh so I'm kind of reaching out here to see if anyone else knows what's being done with this or if it's being used.

B

C

One does it comes from a time where, where where hco did not do this, so I think it's a left over kind of um yeah mark sleuth should know a little bit more about it. He.

B

Did post about it, but there was still no indication of how somebody would consume this, and if anyone is so that's what I'm bringing up here, uh if anyone has any details about who might be using this or if you're using it yourself bring that up. Otherwise, I'm proposing that we deprecate this, because I don't think anyone wants to maintain it.

B

I don't want to maintain it, especially if there's no value here so there's no value, I'm proposing that we get rid of it and we just continue with the publishing of the artifacts. The way we've we've done, the past minus this application registry that I don't think is being used.

B

um If somebody is using it, then we'll we'll have a further discussion on what we think we can do to resolve this. If, if anything, does that sound reasonable to everyone sounds good to me? Okay, all right that! That's it for my topic. I just wanted to raise that and see. If there's any objections.

A

No objections, but just one question, so how do we make sure that no one is using it? Should we give like a a time after this after today.

D

For anyone to speak up well or I.

A

B

A message on the mailing.

D

B

Through the thread about the release, blocker being blocked, I think what I said there was: I'm gonna give it 24 hours. If we remove the publishing logic we can, we can always add it back. If we need to, we can make a new release. Just to you know, add a new uh version, 37.1 uh publish to this application registry if it turns out somebody's using it. So I guess what I'm getting at there's, not a lot of risk that I see in not publishing unless somebody's got a critical use case.

B

That's just leveraging this and then we'll find out and we'll reintroduce it.

C

I see no risk either.

B

Okay, that's it from my side thanks everyone.

A

Thanks, okay, so the next topic that we have uh the regular agenda- I don't know if it's actually a regular agenda, the bug scrub. So maybe I I will jump yeah.

E

A

E

uh Please announce the summit, but before we jump into this.

A

Yeah yeah I okay yeah. I will just add a quick note on on the qbit summit preparation um so in the notes, you'll find a a link. This is a reminder.

A

You know there was a follow-up on on on david's message um before the you know last year already um that we are preparing the scuba summit event uh scheduled to happen february, 9th and 10th um half day each day, there's a link there to the working document for about preparation right now um we have a few ideas for content on on the two tracks: container: uh sorry, container, uh contributor and user focus tracks.

A

um However, at this point we still don't have uh any any like complete proposal, so this is a kind of an encouragement to uh anyone who might be interested to propose topics- and I don't know david. If you want to add uh something more, we don't have any. I mean we're working also on the details on the logistics. This will be announced as soon as possible, but obviously this will be open to anyone to attend um again.

A

Details on how to attend will be um made available as soon as possible, but in terms of content, uh we would like to you know, start moving on on a schedule on a concrete schedule. Topics that are on the list at the moment or on the head like for the contributor track is a discussion on uh you know, continue the discussion that was started a while ago on on the road to v1.

A

Another topic that is on on the list uh is um actually a follow-up of some of the items that were mentioned in last week's community call um on how to contribute, or in other previous calls by steward. You know how to contribute formalizing, let's say the process of contributing to cuba. How you know, explain how to and discuss how to become a member uh et cetera. So that's a topic on how to contribute to kubert.

A

um Anyway, I will not uh talk too much more. There are links to to to the proposal. Tracker um feel free to follow up and david any additional comments or anyone comments. Questions.

B

No, that sounds good. um Do we have any sort of guideline regarding the deadline that we would like these submissions to be submitted by.

A

Right yeah, because we didn't actually do a formal call for proposals. We didn't have a matching deadline for the cfp. Yeah great question um yeah. I know it's not not a formal date, but it's as soon as possible, let's say and or yeah it's it's it's a very informal call for contributions, so um the criteria will be a bit. You know the the proposals are to be made uh in the open. As for requests to the community, repo and basically acceptance acceptance will be on.

A

You know. First come first serve so to some extent I mean we need to after everyone agrees that it's a valid topic so and we don't, we only have so. The total span of the event is eight hours. So that's it's not a huge amount. I mean, there's not a lot of room for many many topics, so I I know this is not a real answer to your questions. For the reason we haven't set a deadline, yet we can agree on one it needed.

A

We obviously need to start populating the agenda as soon as possible.

A

uh The hope at the moment is we have a few ideas and proposals again, not for my not any of them like formal yet, but I believe that there there are enough ideas already to cover the whole agenda to populate the whole agenda. It's just about getting them in and decided. Yeah agreed upon soon.

B

Make sense that makes sense to me. I've been telling people try to get in by the end of the month. I think there is risk that some of these topics are potentially overlapping and given the short amount of time that we have, um it would be best.

B

I think, for the contributors sake to get their their abstracts in as soon as possible, because that allows us to see where the overlap occurs and help guide uh perhaps like how we can take different directions with different content. If there is overlap, so people can can still get uh their chance to present, but just kind of make sure that we're doing it in the most efficient way, so that that's the.

D

Only risk- and I only saw a few instances.

B

Of where that might be possible, so it's not a huge risk, but it should be a motivating factor. I think, to go and get those in early if you're thinking about it.

A

Yeah as soon as possible, and also not only overlap, but I also see some of the topics. I believe there are potential opportunities to collaborate between other people like uh well. It might, you can see it as overlap, but like uh things like snapchat, snapchat and migrations, might I don't know, uh have some word up or I don't um another one is like.

A

In the use cases front, so nested clusters um can also have some over I mean have can have a different take on the session, can be about discussing a use case and might expand to other use cases, or it might focus on on a technicalized much more on a technical aspect and therefore we could talk. Also about you know, keep it. As uh I mean there was interest for a while ago in the community by another member on a terraform provider for cube beard. This could be a mixed topic anyway.

A

um The conclusion is as soon as possible get those proposals in and therefore we can get things in shape.

E

I'm from I can't speak for speak for others, but for me, when it comes to call for uh proposals the as soon as possible, possible means like two days before the uh deadline, so maybe something one would help would uh help some people uh getting them submitted. Yeah.

B

That's important just setting a date makes people actually target something. Otherwise, if it's nebulous, we that's that's reasonable.

B

We can just throw out an arbitrary date and then but yeah. I want to say what I'm thinking but yeah.

A

Okay, um yeah, that that makes sense is okay. um End of the month uh could be a kind of a natural target.

A

Maybe you can real status and let's, let's do something: let's uh chat about it after the call a bit and- and we will send because the the message you send was you know again like almost one month ago, we will send a reminder to the list and include a.

A

E

Okay sounds good.

A

Okay, anything else: uh if not, we can jump to the box crop header.

E

Sure so, uh let's just take one for our convert: slash, convert backstroke, I'm sharing the link to the issues in the chat uh yeah. Let me share the screen that may make this more appealing. Where is the.

E

E

Can you see a list of issues?

E

Yes all right, all right, so um I'm not hundred percent sure how to do this, but we have uh the best firm possible to handle this. So, let's just start: uh unless there is somebody who wants to erase their own issue, I would take them from the newest one to the oldest one. The goal is to either close the issue or make sure that it's being addressed or make this make it addressable.

E

So, let's start from the very top there's issue downgrading from 036.

E

I see this being addressed, but do we support downgrading at.

B

E

B

We should so what's happening. I think when I read into this, was that the downgrade there's two ways to downgrade there's. uh The first way is to simply install the old operator, and that will, if you don't, have a target version set in your q, vert cr. It's just going to match the operator's release version.

B

The problem with that is, the install manifests for the each individual operator aren't necessarily compatible with a newer version of qvert. So the operator has to have our back that allows to deploy the newest version at qfr and if you downgrade an r back changed between the newest release of cuvert and the previous release, if you just downgrade the operator what's happening, is we're stripping away. Those are back permissions from the operator and it can't actually influence the downgrade of the operand at that point.

B

So the only way that I see it as possible to support this and we do support this- is to remain on the latest operator and you have to actually set the target version or the image tag in the keyboard cr to the previous version, and that's going to allow the newest operator to downgrade the operand to the previous version.

B

That sounds really complex. It just means use the latest operator and when you define your keyboard, cr set the version to the older version that you want to deploy and that should actually work.

E

So there is no bug fixing needed. We are just waiting for a confirmation, yeah.

B

So I I made the comment tried to describe how how that's done. I I want to verify it should work. We actually kind of test this. So this is the the workflow that's actually in our functional test suite. I, I suspect that it's going to be fine. I just kind of want to get some feedback before we close it.

E

Sounds good? Okay, thanks david all right, let's move on, then this is getting addressed. The next one is enable in vm xml for reclaiming space when using seth, rbd images.

E

Stef rbd images are finally provisioned. After deleting files from the images, the image size remains the same and thus ends up being filling up, the storage pool.

E

um Do we have anybody from storage or anybody who follows this? What issue is this.

E

All right I take the snow. I will try to follow up offline with somebody uh who handles you.

C

Let's ensure that assign someone from search like michael hendrickson or okay, or is this a sign uh yeah it's an issue you have to use slasher sign cc, all doesn't.

E

E

Okay, let's move on.

E

It smells like storage again image upload file using raw images.

C

This sounds like the target. Pvc is too small.

C

Or something yeah assigned to search again.

F

Could it be give give gigabytes versus skippy bites or something like that? It can be something.

C

Like this yeah, it looks like this right. It's pretty close yeah.

F

Let's look at the.

C

Size difference.

E

Yes um well at least michael has something.

C

Yeah yeah the image size is a little bit more than 40 gigabyte and it's very very good. um We can try.

E

Yeah but you can.

C

Just write here that it's probably yet he probably needs a bigger just mr.

C

E

Great okay, awesome and now we can use the sick label. I guess first storage.

E

To make things still organized okay, uh moving on crd arrest coverage to stop working and artifacts are no longer uploaded. It was reported by merchant and he's working on it.

C

Okay yeah, he self-assigned it. So nothing needs to do. Okay,.

E

We should be good, then um report data from guest agent to vms this field again from marcin, and he got answered from eddie just two days ago, so we can consider this uh being tackled. I guess and coming.

C

From so yeah listen, nothing! We need to immediately do here.

E

uh Some um vm continues. Five minutes then fast with dirty after transfer shutdown and it hasn't been addressed, um brings any belts.

C

We have to ask some questions here like if the vm reacts to acpi shutdowns. If not, it will just wait until the timeout is over and then it will just kill itself um mind if.

E

I stand your mind on this one.

C

Yeah sure give it to me.

G

Hey guys, uh good morning, oh hey, go ahead, are you late, but uh so I had the issue. I don't know if you guys have already discussed because I joined it. I just wanted to uh bring it up, so I'm using uh k3s with keyboard so have uh have like. Have you guys uh used it before with k3s or have you seen any issues with k3s, because when I'm trying to create a vm using a keyboard with k3s, it's actually going to.

G

Yeah, it's actually going to internet to validate the vm rather than going to api server on the k3s.

B

What enter the internet? What.

G

So I don't understand so uh for validating when it's creating the vm, it's going to keyboard dot, io.

G

G

So it goes to http uh somewhere in cube.org.

C

Where do you see that it goes to cupid.org.

G

um You see that post http.

B

Oh, that's, that's a service account yeah.

G

Service account not not like sorry, not keyword, dot io, but what whatever this is. Yeah.

B

Yeah, so that should be the dns entry, that's local to your cluster. It should be.

C

Okay, um yeah, and so so there is no. This is actually the this word api keyboard. Just svc is actually inside the namespace, where kubrick is installed. So, okay, okay, after that, as we see, you could then also see something like dot, cuberd and then dot cluster dot, local. That would be when the most fully one.

D

Right but this should be resolved.

C

So um you're not the first one reporting exactly such an issue on k3s. I know that people somehow always manage to make it running, but I have no clue what they didn't always did that it starts to work. um I think one thing would be to find out if the the network provider is working. So basically, what we require to work is that uh that one part can reach through a service in other parts, not more.

G

C

B

Further than that, uh dns needs to work, so you might be able to reach a pod to another pod using the pod ip, but this is requiring dns to be enabled within the cluster which might be uh which might be the problem uh context deadline exceeded it's unclear if it can um resolve an ip address or or really what I don't really know what that would.

C

What I have seen that is what this normally means is that it could resolve their p address and it is uh connecting to the service, but nothing is listening on the service or nothing is routed forward to a port behind the service.

C

B

So a really easy way to test this would be to uh try to just start the pod and send some sort of request to the api server, okay and and see if it uh if you can at least establish a connection.

C

And even simpler, also try to just do something like cube: ctl get ports, there should be all namespaces and just see if something is scratch-looping or really everything is there.

G

Yeah everything is up and running um there. I don't see any issues when I do that everything everything seems to be running no issues and I actually tried actually having the cni also changed on the k3s to calico, and maybe maybe I was thinking that was the issue, but when, when we work when, when we install this with k8s with the native kubernetes, it works fine without any issue, but uh with k3s only we are running into issues.

B

G

B

To look at is the vert. Sorry, not the verb. The k3s api server logs and seeing if, for some reason so what's happening, is in the validation of this object, our you're talking to the kubernetes or the k3s api server, and then it's calling back out to our service to validate that object, and then it's going back to the k3s api server to make a decision. The decision on what to do next.

B

So if the connection between the k3s api server and our qvert service, that does the validation uh is broken or it can't resolve or whatever is happening there. That could be another explanation.

H

Are you? Are you running it on one k3s note or multiple ones? Oh one, okay, yeah! I I have a. I recently switched my my own single node to k3s and I had a similar problem with the serp management workbook. But I don't remember how I resolved it, but I wanted to try kubert. So I can't try on my system on my k3s, that's very fresh and see if I can replicate it.

G

G

That will be helpful. Thank you.

E

So um process wise, what do we do about this issue? I mean it doesn't look like an issue in convert per se more like an issue of the underlying infrastructure.

E

So shall we just close it or wait for a confirmation? What should we do in these cases.

B

I think it's useful to leave it open for discussion. I think we know that it's not likely a bug in cuba, but if somebody hits this in the future they're going to google for it and it's helpful to see how it's resolved, even if it's not related to like a code change, that's my opinion, but.

I

A

Yeah and process-wise, I see uh I was searching for issues while you were talking about the issue, and I saw an apparently duplicate one. This is 4732. I think for 731 might be a duplicate of this one. Raj.

G

Yeah you're right.

A

Right, okay, so maybe the other one can be close as not.

E

This one was commented on and is older, so I will just.

E

Comment on this one and close the other one in favor of it.

E

E

um If you could confirm that the connectivity works fine and if you managed to make it function with convert, it would be cool if you uh commented the process uh on the issue. So, like david said here, people can I'll find it. If I hit the same problem.

J

J

um This is chris caligari. um I also created um issue 673 under the coovert.github.io repo to have a quick start page created for k3s. We've been talking about that.

F

J

Some time and seems like it's about time to to get that page created.

E

What's the coincidence called, um uh would you find it useful if I length absolutely.

J

Yep, that's why I'm mentioning.

E

E

All right, okay, so, uh let's see how this resolves, but now we can move on um all right. Kubernetes 120 fails on ipv6 test, um so the issue with this one is that it's not anything horrible. The the problem is that our vert ctl allows people to create services for uh ipv6 or dual side clusters as well.

E

Just by setting minus minus ipv6 um issue is that between vertices 119 and 120, the api for ipv6 or dual stick on services has changed and we have to adjust to this change and it's already assigned and should be worked on in the next week or two. I hope big um so in case there's anybody out there using vert ctl with services. You may hit this on ipv6 if you're, just creating the service manual using yaml. You shouldn't have this issue at all.

E

The next one add volume and remove volume. Have nobody in our api reference.

E

um Ron, do you are, do you know what your alexander is aware of this one, or should we do some nudging? There.

C

Herman uh sorry, I was, I was talking to myself at least.

F

C

Got the opportunities over here, so maybe alexander missed it.

C

um uh He didn't reach out to me or anything, but what we need. It's like. I had one guy who wanted to try this out and um the only way he could even see what to do was going to our unit tests or to our functional tests, because we had the documentation, isn't complete so yeah.

C

So this needs to be fixed.

E

Mind thinking uh alexander just so he confirms that he's aware of it, but.

E

All right, okay, flick, ci.

E

Report it 20 days ago.

E

This is no longer like it does. It fails every time for final. Oh, that's! Okay! It got me in the first part.

C

uh So I guess this is something on our test, which doesn't work with some shift assumptions.

K

E

Okay, um we'll look into this one.

C

uh I just yeah, I can just say that on on the keyboard, ci part it's not failing, eject click, finder and so on. It's not failing.

E

There great um I'll, assign it to somebody um after you go all right um feature request something interesting, live update of vmi labels and annotations, uh it's 20 days old and there's no reaction. So, let's go through it.

E

Annotations get carried from vms to the vmi and to the invert launcher, but same history from the labels, uh but this would be nice to live. I did update the vmi labels and annotations. So when I change label of vm it gets propagated down.

C

E

C

Would even go as far as they did. This is a bug not a feature request.

K

E

E

Do we have any good attempts to work on this? One.

C

I don't know with some spare time.

E

Okay, I mean okay, so this is another situation, often scrubbing we can get into for sure. So.

C

Just to send it to me or david, we'll find.

E

Someone, okay, um I mean we don't need to tackle all the issues right away. I guess it's like. We need to make sure that the bug the issue is trashed and we know what do we want to address it or not?.

B

E

B

This is a bug, roman.

C

Well, I know it's a bug because I implemented all this and I wanted to have it work like this and everybody. Let's say.

B

Let's talk about it for a second. If uh we're saying that labels from a vm, let me make sure I'm getting this right. Labels from a vm should be carried it's vmi dynamically. Oh.

C

It's from a vm, I thought he changed it on a vmi and it's not propagated.

B

No uh character: okay, no! No.

I

Let me, let me make sure I'm reading this right: okay, yeah being science, adding annotation pod.

C

ah Yeah no, this works as expected. Actually.

E

B

Live update. The demo has labels, we're not gonna. Do that so there's the conflict here is that if somebody uh um we don't really know what to do. uh That's correct, because because if somebody said something directly on the vmi, then we'd clobber it potentially with a vm annotation, and I don't.

H

Think that would be.

C

Expected yeah, what I thought this is about is about which we we really change labels and annotations on the vmi directly and they're not propagated to the pod. I think that's also not happening, but this is a bug, but this one here is something which I would prefer to not implement.

B

I I care very little about the vmi to pod one. If somebody found that useful, then.

C

Yeah, you normally need it when you have controllers and you want to take out a vmi from the controller so that you can investigate. What's going on with it, because when you change the labels, the services stop routing traffic to it and you can investigate it. That's the use case for that, which you also have on parts.

H

B

I

That makes that.

D

B

A lot of sense, I'm not sure if we're there yet well, I guess we are there yet, with the replica yeah all right.

C

We might we need that, therefore, but this should not be implemented here. Yeah.

B

Different discussion makes sense.

E

So uh david, sir, would you mind commenting there with explanation? Why we don't want this.

B

Sure uh was that four seven one one yes yeah of.

E

E

All right, uh I think that the 20 minutes have passed, and I don't want to drive this too long, so we could get sick of it. So unless somebody wants to raise their issue, I would call it a day and continue with this done next week.

C

Sanskrit was was really good session, thanks for thank you for having that thanks a lot.

B

A

B

K

A

Okay, see you guys next week.