Kubernetes SIG Cluster Lifecycle, 31 Aug 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: SIG Cluster Lifecycle - CAPV Office Hours - 2023-08-31

Description

No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).

A

So hi everyone um today is August 31st. um This is the cluster API vsphere provider meeting. um We follow the code of conduct which basically comes down to be nice to each other um yeah. um So maybe first it off. Are there any new folks who want to say hi I'm introduce themselves.

A

B

By Richard Vanderpool um I've, been some of the Kathy calls was my first time joining the cab vehicle so nice to meet you all nice to meet you too, and.

C

Welcome hi I'm, Punit I'm from VMware and uh I have a full request to open, which is uh the main reason I've been joining.

C

uh Not my first officer meeting, though.

D

C

Nice to meet you all asked me to.

A

Do so I think, then we can switch to the agenda um news and updates just a quick uh one. We released v1.8.1 this week, which um also replays to Cluster API v1.5.1 and includes 31.28 um kubernetes support um yeah with that I think we can also already switch to the first topic of discussion.

A

um Daniel I think you can give it enough.

E

Thanks yeah I, don't need to spend a lot of time on it. I think we we got a lot. uh We got some discussion going in the issue, but yeah it's it's a kind of a mismatch um between the the format that it can't be uh accepts and what the CPI and CSI implementations accept. So if you're it's it's kind of a pitfall for for end users, um you know if you're not aware that the CSI and CPI need um just the host name or or IP on its own and the port.

E

If you're using a non-standard port separately, um you won't find out until you deploy the cluster um as you. If you provide a URL with a scheme, you know, or a host and port to Cappy Cappy is Happy everything. You know, everything's fine, it's able to to communicate with uh with vcenter um and it seems like there's, maybe some sort of Legacy um Legacy implementation, or you know the support for for the for these separate fields and uh yeah. It would be.

E

It would be nice to to see if you know what what the sort of right path forward would be. um That's what I want to bring it up here.

A

Yeah, um so if I got to write, the action item is not on the cafe side. It's more on the provider and CSI side. um Yeah.

D

Yeah stiffness.

F

Not just any answer.

E

I think I mean I laid out. I think the options um you know, I I would let's, let's document this.

E

uh uh You know, while while you know, while it's there I I would I I would definitely like to see um a a change in the CSI and CPI to to support this, but I also understand if it you know, if that's, if that's some there's some problem doing that, and so in that case, maybe you cap V could um you know could could play play nicely with with CSI and and uh and CPI so I. It's not clear where the change needs to happen.

F

Yeah sure so yeah, if I, if I see correctly. Basically we have a few places where we can configure our sub URL in clippy I think so.

F

My take as a road-on issue is basically I think it would be better if CPI and the cloud provider would just support URL field instead of like us, adding like two additional fields in all the places where we have the URL today and I'm, trying to support them all and doing validation that you only set the right ones, um but I think it basically depends on on what um maintenance of those components are saying: I'm, definitely fine with adding documentation on whatever the current behaviors today. That's that's, absolutely not controversial. That's fine, um but otherwise.

F

uh Just from a cafe perspective, I would prefer, if you can just additionally support um like the same stuff. We supporting copy in the other two components, just configure the full URL and then we're consistent and everything is fine. I don't have much background on the other type of ones, so finger Primos depends on like another answer.

A

That makes sense.

E

Yeah and if and if there's uh thank you for um for tagging, I guess uh you know I'm not too familiar with them. You know maintainers of Cs CSI CPI. If there's you know, if there's anywhere that else that I should you know like bother bother anyone or you know, show up at some Office Harrison would just let me know um and uh I I can uh I can also see if I can I guess, dress some Doc. Just uh it's just a little note.

F

Nervous yeah, I I, don't I, also don't really have a good audio, so I know that um LeBron and Niki are definitely like the right ones to ask for glove for whatever issue.

F

um Honestly I, don't know if it's the same uh in vsphere as it was for openstack. Basically, the the CSI stuff is the same repo as the cloud provider, and it just all belongs together.

D

F

They might also be the right ones to ask for CSI, um but I I literally, don't know no.

G

It's it's separate so.

F

G

The lead and she's she's, the chairs uh storage tag, so I, don't know if they have a separate meeting for researchers. So if it's just in the CSI meeting itself.

F

Should remember start off just additionally taking them on? Did you understand what I think.

H

F

Okay, can you uh can maybe just see them.

F

ah Of course, oh my God yeah I should have known. Will.

A

Do thank you. Thank you. Thank you. Folks,.

A

If there are no additional topics, we could go to the next one um Richard.

B

Hi, yes, um I've been um working on a ipam feature with an openshift um specifically for for vsphere, and one of the concerns that we've had is the experimental status of this API. We would like to continue using it and to and and in order to do that, we would like to promote this to V1 beta1 and there's been some discussion in the issue, um but it's it's been a little bit since we've talked about it. I just wanted to kind of put the question out there.

B

Are there any concerns with that and if there, if there aren't, how do we proceed with this promotion? I'm somewhat new to contributing in this space? So.

A

I

I, remember correctly: Jacob had a few open things in the current implementation, mostly around determining host names or DNS entries that might require some changes to the uh to the contract.

I

um Not here, sadly, so no Pokemon are select later on to see very uh where that work actually ended up.

B

Okay, that's that's awesome. Thank you. So much for for doing that. Let's see there are a few questions here: I'm not sure who popped their hand up first.

F

uh Yep yeah, it seems like it makes sense to definitely follow up in slack, um I'm, not sure how it was done in the past year and how much like of the ipam provider was handled on this meeting.

F

But at least since I'm, showing like the last one two months, we basically um I, don't know we never talked about it provider, so I'm not sure how much we can make a call for live and provider in the Clapping meeting, um but I think, let's, let's follow up in slack or on the issue and um let's try to at least get um some discussion going around like what would be like roughly the remaining things that you want to do before.

B

Yeah, thank you for letting me raise it here. I wasn't sure either about the proper form, but yeah.

F

But same for me, I mean I, know that uh not IPM provider is pretty strongly correlated to cap V.

H

F

Yeah yeah: okay, that's probably a huge overlap in the past.

D

And I also noted a lot of like.

F

Vmware people uh we're in more of them after.

B

Cool. Thank you. Max.

B

I

I just forgot the low end.

A

Okay, I would just ask the question so that one is the current implementation I linked here. That's just right: okay, I'm, just not that I at the wrong reference here; um okay with that, if there are no other ad hoc topics coming up or questions, if.

B

You don't mind if I can just come come back next week and just follow up on that. I would really appreciate it.

F

Yep, just one note um uh just so you know this meeting is bi-weekly. Oh.

B

That's right, yes,.

F

B

You so much I appreciate it.

A

Yeah perfect: okay, with that I think we could do another round of going over the open requests.

C

I have a adult question: I I did not put it in, though.

D

C

Document I wanted to understand the uh the way pull requests, get merged to release branches. uh I have a pull request open for me, but the tests have been failing and I've been getting emails about the release branches. So is there something I need to do about it or.

A

Do you have a reference to that um which per request maybe request.

C

F

Just a question: were they like recent emails or whether from your previous PR, that you had against the release? Brunch.

C

No, this is a recent one. This is I created a new one and uh yeah I've been getting emails about uh the pull request. Failing for the uh the branches, uh the release branches.

F

You shouldn't be getting anything to be honest, um maybe if you want we can, we can maybe follow up async and you can show me the email and we can try to track that back. But just I can tell you what an upstream it was. What should happen. Basically, you have this PR, the jobs are running there. You might see some figures there and only after this PR is merged, and if you think we should backport it, then we start doing anything against release branches I'm, just based on this PR. Nothing should happen.

F

So whatever is going wrong is not correct. Okay,.

C

Okay, then I'll share it with you guys offline, and uh you can go over that.

F

Yep yep, maybe it's I, don't know. Maybe it's something happening on your fork or something uh I mean it should be unlikely. What who knows yeah.

C

But yeah I think I get an email every day, yeah early in the morning. It.

A

Could be that these are GitHub actions running on your fork? Maybe are.

F

A

From GitHub, or are they coming from Pro.

F

Okay, yeah. Maybe so the question is um what like? What um Chopper action? Is it complaining about? Because um there's one thing: we have a few periodics running on our repository um and nowadays, if you Fork the repository, they are disabled by default, but in the past they were enabled pretty fault.

F

And if you run, for example, our security skin on your fork or other stuff and your release branches are out of date and you still have findings there and whenever the periodic runs, you get some sort of notification and, depending on your setting, it's either a GitHub notification or an email.

F

um So maybe that explains it somehow, depending on the exactly yeah.

C

Yeah I think these are. uh These are coming from my phone I missed that part and uh but yeah they are from GitHub, um so I guess uh maybe what Killian mentioned on chart uh I do not I haven't pulled any of the release branches and that might be it yeah, yeah I, don't know uh so so the solution I'm hearing is to kind of update the release branches on my own, but is there a way to disable this.

F

Yeah, you should be able to just save the forks.

A

Yeah go to your fork into settings and under actions General there is a knob to disable all actions on your fork. Okay, yeah I'll. Do that. Thank you. You're welcome, okay. Let me close that one.

A

And thanks to Nadir for taking notes.

A

Okay, um let's Maybe opening that around of requests to discuss okay, um intimate test for iPhone support. um It's the latest State here.

A

Cigar was working on a dock um yeah. Basically in CI we need the mechanism to yeah reserve a block of IP addresses, so you can Implement that um yeah that's to be done. I think. Is there anything you we could add here.

F

No I think it didn't change much I mean. Basically, we have to work a bit more on the environment and then we can build something in there um should be realistic, but we'll need a bit more time to do that.

J

uh Specifying different.

A

Credential vsphere machine I think there was some message here.

F

I think the plan was that you want to do something um by Reviving, a PR on core copy around multi-tenancy.

G

Yeah I'll start throwing out my yeah.

F

Do you maybe want to give a disclaimer on this PR that um I don't know at least a hint that that's not something that we're planning to support based on that doesn't realign to what we would want to do just because, before someone is investing more time.

G

Yeah I did well I need to check because I think.

F

G

Think this one we might have to allow for it, because it's to do with like more TVC stuff, okay, so we'll need I'll need to be brilliant.

F

Yeah then, let's wait for that and and then after that.

A

Maybe a question because all the pull requests written out would take a look, are add an update, I think at least at the last month, or so I know not really. Okay, look at what I said. um Okay, then continue to the next right.

A

And transistor verify, link clone I. Think that's. That was something we already talked about.

H

um Yeah I got it on my list.

F

A

F

Just a question: do we also have an issue for that which is also signed to Killian, because I think, then you can just close the pr and only have the issue, I'm pretty sure I assigned the issued Killian when I saw it the last time because he already had a PR. So we don't have to keep both yeah yeah.

H

F

So, let's toss the issue just with a note that um clearly open my new one or something once you he gets there.

F

Something like it might never happen, so we can just close this one already.

A

Okay make the cleaning of clusters identity secrets configurable with Default. True, is.

C

A

F

Was a select discussion around that yeah that was probably not linked, but, to be honest, I mean if you're not getting a response since April um yeah I mean it. It might be fine to add this feature if someone actually wants it. But if someone doesn't answer for almost half a year, then I'm leaning towards not adding another Knob I.

D

F

The bot closes it, that's fine. If the bot just goes at some point, then I'm not sure if it's worth adding another flag.

A

Yeah, um so we could either wait for getting closed automatically or we.

F

I'm slightly wondering if we don't have a state board running in or repository to be honest uh that should be well. It should already qualify.

A

Well, Nadira assigned himself in July. So maybe that's because that's the reason why the state boat didn't kick in here.

G

Yeah I think this is definitely related to the thing that I need to revive in capping.

F

Yeah, oh okay, that might be fun yeah, but, to be honest, I mean we cannot at this point after almost five months uh waiting for an answer. To be honest: let's go just close based on. We didn't get feedback, uh feel free to reopen or open an issue if you're still interested in that I think that's fair, and if we get into another discussion, then we kind of see where that ends up. But you don't have to keep if you're surround forever.

F

Just write, if you are still interested in this feature or something I, don't want to say it's not important or something like that.

F

Yeah, you can definitely write your lack of feedback.

C

A

Quantum Library support proposal I think we discussed that one in the last.

F

A

F

Going to say this pretty much that we get feedback um I, don't know, Ricardo is not here so I'm, pretty sure that Adam is busy with a lot of other stuff because they know what he's working on yeah.

F

um So what we probably need is at this point, either wait or if someone is interested, they should take it over which might be Ricardo.

A

F

Think but I mean they have to talk to each other, essentially.

A

Or something like that, it just gave feedback last week so.

A

G

Think that's, okay, I'm might be inside baseball, but I'm not sure Adam, Adam's gonna have time.

F

Yeah, it's just based on what's working only probably doesn't have time like until November or something or December, so we can definitely give the hint to like recorded. Maybe maybe let's say that Ricardo, um if he maybe can, can contact Adam directly or something and ask um if it would make sense for him to take it over or something like that, just to give a pointer and then Ricardo can decide what he wants to do.

F

Are they configured out together.

F

If it would make sense, did you take it over or something.

H

F

D

F

Busy pending the discussion on a document on the issue, do.

D

A

Think that one we also discussed last week uh for two weeks ago, I think there was an update here.

A

A

So we need some someone needs to invest time here to I think clarify the situation or help clarifying the situation.

A

um I, don't know the details. I think this needs to get the laser engagement in here.

G

Assigned to me and I'll chat with Andrew, if I I don't understand something.

A

Okay, I, don't think we could skip up. We could progress protocols right.

F

A

F

Can skip that one I mean it's working purpose um yeah and if it's done, we just review it. Yeah.

A

It's own a reference resiliency test from clue, yeah.

H

This is active, I think we can move.

D

H

It's just failing in an annoying way. Yep.

A

um That's the request: yeah.

F

That's that's almost that's almost done basically yeah.

A

F

um I uh I want to spend like another 10 or 20 minutes clicking at uh the links that uh that you provided and then just approving it yeah. So that's good, let's just asking for more details around like how does this actually work.

F

um Huge might be just good to just ask a high level question like: um are we in general fine? If the additional assertion does it break anything or not? I figure I discussed it at some point, but just to double check.

D

E

Yeah otherwise,.

D

F

Just comes down to like uh revenue for correctness, yeah.

A

Exactly so, um one thing for me here is yeah: I will take.

D

A look at it and add.

A

Some test coverage for defined VM function if possible, um yeah. We discussed this additional check in the past. um I now implemented it in some way, which this one will add. More calls to vcenter when find VM is called, um but from discussion. I think that's okay or accepted to be more secure. Instead, um yeah happy to create some feedback on that. If that goes, the right direction.

F

So I think I have two parts, one for this I think I, probably wouldn't already try to optimize vcenter calls if we don't really have scale tests and data to um so we can tell if you have a problem with too many calls and we have to literally optimize for digital calls or add caching for that sort of stuff um and the other one is just visit.

F

A high level question like what we do here now is we calculate the expected inventory path and then we make sure that whatever VM we find actually has that path and I just assume. That's totally fine, because the inventory path will never change right. I mean if we do be motion on whatever else might be a totally valid thing that just happens. This path should always stay the same, correct.

G

uh Storage remotion because.

F

Yeah because otherwise we really Break Stuff, that's right with that.

G

Yeah, what happens in the case of storage or emotion, because that might move the inventory path? I'm not sure, though, um and you can also drag and drop stuff in the uh UI, so yeah I think there might be some things you want to check there.

A

Yeah so in the case of manually moving VMS um and then.

F

Eventually paid right, I mean that now that might break.

A

It but I think from the yeah. Sorry.

F

um I think that was intention essentially of this PR, that if people are doing weird stuff um but just modifying things that it should actually break but I mean, if we now say um something like drag and dropping VMS around or something that um you should be able to just handle, and it's totally fine VM goes out of sync with whatever people manually did in vcenter and they don't want to fail on that. Then it's I mean then we can just can do that in this version. Yeah.

G

Yeah, it's been like that for a long time, so that would be quite a significant change.

G

Max's faces and I'll shut.

I

Up- and we actually rely on that in a few cases, so we did a few moving classes from one recent or from one uh cluster to another or it's exciting pool from to another and then just rolling it out afterwards, at a more convenient time.

F

Yeah yeah I mean um we can still just change our opinion on that one and say: okay, hey, that's not something that we want to break um and just close it essentially before you invest more time. I think there's nothing which is strictly requires on the meditation which is fought like it might be good to do, because we don't break anything and safely.

F

Okay and I mean if you're, in order to break use cases for totally valid, and we can just close it now and move on yeah. To be honest,.

A

Yeah so I'm, okay, with that um it was just a follow-up and a from a discussion we had and we thought it would be good, but it turns out it's not good. So.

A

What's that one.

D

A

That's also close this one.

A

Yes, back at feed, GPU implementation.

C

A

C

Need to take care of all the review comments. I haven't uh had the chance to work on them. Yet.

A

C

Good um I did have the question about the duplicate code path. So is there a preference on which one to be used because I was going by the existing code? So at the last comment that you made.

A

Yeah so yeah. The point is here.

A

A

Yeah, the point is, we already have some PCI pass-through logic which lives at a different package which is here, which is for the PCI pass-through mode and also the way it's used. It's different, like your implementation here, is directly in the clone of the vcrm I. Don't know how to show it now. The.

C

A

And the other, the old way or the the current way of adding PCR devices is after cloning there's a reconcile, PCI devices function, which runs before the VM, gets started.

A

Okay, so so I think when we do the GPU the V GPU thing, we should do it the same way as for the password devices in that function instead or at the same at the second function. For for that, and do it the same way to stay in the same pattern?

A

um The reason for that is for the PCI devices for the past, through the PCI devices or for the old way. um This also updates a condition which would not get updated if we just do it in the clone.go.

C

Okay, I'll, okay, I'll make the change I guess the issue happened, because my initial pull request was based out of 1.6 and uh I. Don't think this implementation was in place, then so I just kind of re-based it on. So maybe that's the reason. I'll fix the output request.

A

That could totally be the case yeah. Thank you.

A

Okay, uh yeah I think this is just getting it over the finish line and there's some open ings.

F

A

Last ones, but yeah I think this.

F

Is yeah? That's just spinning.

A

Yep, that's a draft.

F

Or maybe you should talk about it just for a second and then the previous one, because relatively big change, so um the large effort that we're currently doing is we enable a bunch of lenders and we see what it defined and try to improve stuff, so we're not trying to plan default lenders, but it's just trying to see what they tell us and then actually calling me um so that change is busy doing I, don't know who's yeah, how familiar you all are of Capricorn.

F

But essentially you have this nice pattern of building our own context, embedding the goal context and then just wrapping another condition, another context and passing that context around like crazy.

F

um So what what we deliver that triggered this here is basically that there's something called contained CTX which basically says, or you shouldn't like embed the context into a struct and then just do weird stuff.

F

um So what you're planning to do is part of this effort now is um we want to make sure that we have a totally normal, regular goal, runtime context that we pass around as a first parameter and split it out of that magic? Other context that we have so that we can use it like everyone else, too I don't know, get a locker that controller on time did put into it. We can take the context that, or the controller runtime gives us with.

F

Oh yeah I mean sorry with the locker stuff is in there, but we can also like just propagate it down to whatever we want um Putlockers in there add more key value pairs over time propagated around all that sort of stuff. You just want to end up with, like the normal pattern here, reverse having this magic, CTX thingy that you pass around and nobody knows what's in there and you don't really know what what parts of it certain functions are using.

F

So just give some context on this thing, and that's just the very first step which basically comes down to moving the context out of the uh or contact structs and having just a separate normal context, parameter throughout the entire code base.

A

F

Wants a object or something.

A

Yeah, so that's basically, this step here is the pull request which we did just see and it gets followed up by addressing The Next Step afterwards,.

A

Okay, um no, that's a quick one eye open today. um If anybody wants to run Cube, stick metrics with custom resource metrics form powered by cap V.

A

This work in progress per request is based on something in Cube, State metrics, um and this basically has or contains the first configuration for getting metrics for cavivia um CRS um yeah.

A

This was Steve work in progress until the original Prairie quest in cubic metrics gets merged, which is a generator similar to controller gen. But just for creating this Matrix configuration from you know markers in the go code.

A

um So everyone could ignore that one for now.

F

The same place, just one just one question question before anyone from disappear, so we have a very unique situation in Cappy, which is basically mean, which is basically that we have like multiple CRS with the same names but in different API groups. I'm just wondering when we extend this to also cover the supervisor series.

F

um Will that go nicely in some way or will crash badly. That.

A

Will go very nicely because um the kind and API version should be part of the metric as a label.

F

Okay, so we can basically decide either. We have like one resulting configuration for cube symmetrics, which is like both set of series, or we can just generate like one.

H

F

Supervisor from non-solutions.

A

Yeah so overlap. One point is overlapping names won't be an issue because of the label. The other thing is, we could prefix the metrics of supervisor differently. So let me find that one I.

F

Mean if it's already via Dimension, then we don't have to write.

I

In your configuration for cube, State metrics and not having them in cluster um will just result in no metrics. Basically, for that specific resource. There's no problem of defining stuff for resources that you don't have it complaints but doesn't break.

A

I I didn't get that sorry.

I

I think because it doesn't have the supervisor clgs installed, but you have the metrics um keeps the state. Metrics will complain that it can't find those resources or in the worst case, doesn't have uh the the are back for it, but it will not break Cube State metrics.

I

A

F

A

um And with regards to Rio lab there's a prefix, we could Define So. Currently it would be Camp V, underscore vsphere cluster and 40 VM operator versions. We could have a different prefix if we want that, but we don't necessarily have to because of the GVK being included as links to.

D

F

Thanks and the two cases are included in every single metric right, um yes, not just in symmetry. Okay, all right sounds good.

A

Okay, um I think we're here over the list of the requests, so we have another 18 minutes left. So maybe we can take a look at the issues too.

A

And I should have mentioned it before we started a proquest on we're out of the normal Legend and feel free to uh stay for this discussions.

F

Oh yeah, that's that's what's unintentional, but it's sort of fitted um just in case. If someone wants to talk about any specific topics or issues, we can definitely do this first before we just randomly go over whatever issue we find yeah.

A

J

A

Me yeah hi hi and welcome. First, um we had this kind of question already in the last meeting and Stefan I think you went ahead and added some good first issue. Labels.

F

Or something like that appointment.

J

J

But one issue like industry and iPhone, a person herself in the 2016.

F

um I'm not I'm, not selling you 100, because the audio quality is not that great. It's. The first part of question was: if you can just I, saw yourself uh 100.

J

A

Also feel free to- maybe just so in the slack Channel and we can have an Asic discussion there. If you would like to.

J

F

Can you open that one Chris he.

C

F

Issue in the channel I can work on this. One I would just take a quick look at what issue is about and if we still think it makes sense because I'm probably nobody looked at it for a while.

A

Yeah, okay, so this one would be about the current default templates when creating a cluster, don't include a cni um and the idea would be to provide a flavor which includes um cni, I. Think.

F

I'm not sure if it actually means to include a flavor which is generally like that or if it just means modifying the the generation tools, so that you can produce a template with flavor. But we don't necessarily like publish it.

A

F

Not sure I mean it could we could interpret it both ways. I guess that was during talking about template generation.

F

Yeah I mean the normal thing to do. I think is just the templates that we probably should publish and the line to look fixed artists that they don't contain. Calico, because the the clicks are literally like deploy all the stuff and Addie and deploy Calico I. Don't have any objection to make it possible um I think it is um that someone can generate a little bit like that I'm just not sure useful.

F

It is because I mean, basically, you have to download the Calico yummy from somewhere and then embed it in a cluster resource set right.

F

Yeah I mean I would be fine with it if someone um wants it or if you specifically wanted uh wanted feature.

J

I can work on any, but I am beautiful about it.

F

Yeah yeah sure.

H

F

So the problem is definitely not like um that. Someone else would want to work on it's just a question of. Do we actually is there? Anyone who wants this feature right, because we just don't want to maintain stuff that nobody wants to use um and I'm, not really sure if um to be honest, I'm, not too sure.

J

That it just doesn't make.

F

F

Yeah I think I think that makes more sense. Probably I, don't know if anyone else has some opinions on that, but I mean to me it sounds like based on a quick start. It doesn't make sense, um it also doesn't seem like anyone was asking for it really.

B

F

Mean it was an idea, I guess, but since then I would probably close that one with the comment of like hey. If someone would be interested, please reopen, but otherwise we assume that nobody needs it, because the regular quick start deploys Calico separately.

A

I always hear some plus.

F

Ones but uh general, what definitely makes sense is just take a look at the issue and sign yourself ask for more guidance or something like that.

F

um Yeah I think that one makes sense, but it's not necessarily an easy one, but um yeah feel free. Take a closer look at the issue. um Just ask if you need help and uh try to provide more info, um definitely feel free to just assign yourself the other one. That's fine.

J

Yeah, okay, I I, will take a look.

A

And definitely thanks for trying to step up and join and help the community I'm always happy to see new people, absolutely okay, um so with another 10 minutes left, let's take a quick look at issues um and if I click again what happens? Okay,.

A

Okay, Network requirements.

A

Yeah I think there was a yeah I. Think LeBron already asked for more details or follow-up questions, so um I think one one thing to see is: we should take a look for The Stables. um If this is running on a repo.

F

Yep, that makes sense, because it really looks suspicious in general.

D

A

F

State on this issue so.

A

Yeah so maybe it works, you know: VM computers, resize same data store.

A

Yeah, that sounds like a storage policy question which may be related to the issue or pull request. Ricardo um also must add.

K

Yeah I think this one I'm back sorry I think this one. It's uh it's that yeah I've been jumping between meetings the whole day, but this is this. One was fixed that I I uh I just didn't follow this issue, but uh that's that's fixed by by that pull request that got measured and released it on 181.

A

Yeah, that's a very good answer.

K

And I have just tested that today, so I can confirm it's working.

A

um I, don't know how we do this, usually, if I ask for confirmation I'll just directly close, it.

K

I I think so, if I have added this like fixes something on my pull request once this gots merged the issue, it's closer as well right, so it I I I'm in favor, actually to doing closing and saying that if, if it's not working to reopen yeah.

F

H

D

F

A

And it didn't completely get your question. It was what kind of Gola golang library to reuse or.

J

Like like for contributing to this product of what wooden Library I, have to learn.

A

What goal in library um there are plenty of libraries for use, it's the controller runtime, there's Govi Momi, which is for interacting with the vsphere API, for example, but also for VM operator. It's a different thing, so I had to give a simple answer to that.

J

A

Yeah, thank you, um I, don't know if you have any any had any sessions at some Cube Quint or something like that like deep Dives with Capri I doubt so, but maybe there's something around that I don't know.

A

Okay, um I think we don't have to comment this one, that's the proquest, which is open.

A

Multiple clusters presume.

F

Yeah, that's an interesting one.

A

It's a specified, multiple compute passes for Zone.

A

Okay, the one crashed there. There are some some apps so.

A

I seem to be use cases for that. um The question is if it makes sense from the point of view of the project.

F

I mean what looks a bit weird is that in that example, all zones have the same compute clusters, I'm, not sure if I missed something in a text, but would it be actually a separate failure domain if you use the same compute clusters or maybe the example is wrong and they're actually separate clusters in each Zone.

A

Yeah, that's a very good question.

G

A lot of people seem to like it yeah.

F

But maybe maybe let's just assume that, like you know, we have like Sone a cluster AV zombie separate cluster, a b, not the same um I mean we can definitely ask that um I mean I, I, don't know. I, probably don't know enough about this underground. Channel since sounds like a fine-ish idea. It's just that it's probably breaking change through API to I, don't know I mean not necessarily breaking change. um We can add a new field. We can deprecate the old one, etc, etc, and what what's like a supervisor's? Take on something like that?

F

G

I I think it's a no no.

F

It doesn't work there yeah and because you always have like three availability zones and it's pretty clear for that. That's right.

G

F

Yeah I think I mean um might be worth asking like if those uh complete cluster APR, separate ones person or if they're meant to be the same. Apart from that I think I mean no. If you don't necessarily have to do anything with that issue right, it's there. It's fine.

H

Regardless of what the example says, we could allow this anyway right.

H

If somebody wants three zones with the same clusters in every zone or.

F

The like what I don't know is if the tagging works: oh yeah. No, that works sure yeah I mean if someone wants to do it, um I guess why not.

F

I mean I'm I'm, really not sure about our apis and how they would have to change to make that work.

F

I mean the weird thing is basically, if I, remember correctly, if you, so, if you scroll is slightly up so the entire first half of the crd is just to make sure that attacks are correct on the region and understand. So, basically, we ensure, like okay, the data centers set correctly and the zone is attacked correctly, and it can only do this like for one data center on foreign um and everything else then basically actually defines like what is this failure domain.

F

So, um while it's fine to make to add a slice down there, um it doesn't really fit to. We can only make the attacking for one compute cluster, we'll make uh replicated thirds correctly.

F

So that thing up there probably also have to has a change in some.

H

F

Mean ideally because I mean I, don't really like the API at all. To be honest,.

A

But what is it possible to select the data center also at the Zone level,.

F

um That entire section there is not used for selection.

A

F

D

You've been Data Center, so.

F

um I think, basically, you can do whatever you want, and region and soul and doesn't have any influence on anything else. It's literally just validating sounds so I'm, not sure if you have a validate in webhook, which ensures that you don't use the same type in region and Zone. um It's also fine. If you have multiple compute testers that you use the Zone section only to validate the text on the first compute cluster. It's just a super super weird API, but that's already today, in my opinion,.

D

A

So I'm thinking of if this is not already possible, but just don't use compute clusters in here.

F

um I think the computer here doesn't make any problems, because it literally just looks at the commute cluster and says all the tag is fine and that's no, no impact on anything. What I don't know is if that compute clusters feel that there's a slice here, if that is actually a slice today,.

A

I I, don't think so.

F

Exactly and I think that's why it doesn't work today um and sure you can change this to a slice or add a slice and make it work that way. But it's the super view to the Zone thing, which is only one: it.

A

Would still does not work up here, yeah yeah.

F

I mean it would work because you, you would only validate one.

A

Of your clusters.

F

And nobody cares about not validating the other one, but the API looks even stranger after that change.

F

Maybe you should get rid of the stacking thing.

F

Yeah, maybe we just fold all of this, like in a Ricardo's document and figure out like how this API should actually look like.

A

uh We have an issue for that right. It makes sense.

F

Yeah makes 100 sensor linger.

D

That simple drawing.

F

Yep but, but maybe that's maybe that's a good main point, maybe we should just figure out like the requirements we already have and what else makes sense and then try to get to some sort of.

A

That's this one right.

F

um Yeah, basically, to get to some like okay. This is how this API is supposed to look like and so like final step, and then maybe something in in between that you can all do with your own better one to enable more use cases.

A

Well, I think with that we're already two minutes over time, so any last words from anybody, otherwise I would feel close to call and stop the recording thanks. Everyone for participating and hopefully see you in two weeks.

D