Kubernetes Cloud Provider Special Interest Group, 2 Feb 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SIG Cloud Provider 2022-02-02

Description

Add Approver/reviewers, pull requests, new meeting time, and more.

A

A

All right, um hi, everyone today is wednesday february 2nd, and this is the bi-weekly cloud provider sig meeting. um I will share my screen and then we can uh get started with.

A

Agenda y'all can see this.

B

I can see it.

C

Yep looks good.

A

Okay, um so yeah, I think, last week we said we're gonna try to um go through issue triage first um instead of last, and hopefully that kind of helps us manage our time a bit better.

A

So yeah, let's do that. Let me just open these up.

A

uh All right, I do think some, I think we triaged one of these last week so, like I think we can skip this one. This was we were sending this to storage. Folks um same with this, I think we redirected this.

A

Oh, I think this one needs to be.

A

So these two are done. I think, because they're storage folks look at it. I think really it's these two that are that we need to go through. So we'll start with this one uh azure load balancer not reconciled properly.

C

Yeah, I took a look at this one before the meeting and the weird thing is: um it looks like he's just getting it like can't find the ip address which okay, but he can't reproduce it. So what was leading to the ip address not being visible, I'm not sure, and apparently he doesn't have a way to reproduce it.

C

So unless that looks familiar to people, I can't even I mean it's the entry cloud provider so sorry, the ccm, the cloud controller manager so yeah, I'm open to suggestions- and I did mention to him as well that even if we can diagnose this, if the problem only exists in the version that he's using um it's going out of support in april, so like yeah, it would be nice to know if we could get it reproduced in a newer version to see if it still exists.

C

um That was kind of my initial thoughts, but maybe something in there jumps out at someone.

A

I I have seen like not specific to azure, but I I have seen similar bugs like this, where, um like there's logic in the provider, to match against an ip that was existing before, um but then like. If there's something externally that might have deleted the load balancer or the ip by accident, then the controller tries to look at the load bouncer again for that ip, because the ip is stored in the status or like ip stored in annotation or something, and then it will try to reference it back.

A

But then, when it queries the cloud provider because a new load balancer was created without that ip, then it kind of gets stuck.

C

A

I like an orphaned um service or low bouncer, um but you're right, like I hard to say, without being able to reproduce the actual issue. I.

C

Mean even if we thought we had a fix and we thought we could test it since we can't reproduce the error, I don't know how we could possibly be expected to say we solved it.

A

C

That's one of those.

D

C

I don't know, did.

D

C

Accept it, though, or should we just sure we can, you can accept it, assign it to me and I'll watch it but like unless we can reproduce it? I don't know how we can fix it, because we won't know if a fix works right, like I'm kind of we.

B

Should make, I think, I think, andrew's doing that now, but yeah. We should do that in the comments just say without if we can't reproduce it there's no way we can fix it effectively.

C

Yeah- and I did reply to the guy earlier and and thanked him for the bubble report- and I did say it would be nice to see a reproduction and a more recent version of kubernetes just because like even if we somehow come up with an answer tomorrow and get it in like 1.21. If that's the only place that exists um is entering maintenance mode very soon, so like yeah, we would. We would want to be able to reproduce it for sure.

C

I know that's such an unhelpful answer, but like what are we gonna do.

B

Well, and if if andrew's line of thinking is like correct here right- and maybe there was some out-of-band thing, that happened like a load balancer disappearing- it's like well how? How are we ever going to figure that part out? You know.

C

Do you want to include that in your comment to him andrew just be like hey, be aware of the fact that any of your things external to your your cluster? Could I don't know how you want to put that, but like be a factor.

A

Yeah, I can add a comment follow up because that might take a while for me to write.

C

Yeah yeah, but okay, thank you.

A

C

Sorry, what was that? Oh, no, I said yep this is. I want to try to reproduce it. I just I'm not sure how sounds.

A

Good all right, uh missing, annotation to specify a private static eyepiece for aws nlpe.

E

That's a feature request, so I think we should um either ask them to open it on the cloud fighter. Repo uh or I can copy it over.

A

um I will assign you want to do that or join me just do that now. um I can do it okay, so then I'll leave the needs trash then, because we're gonna close it right yeah. We.

F

Do this already in the load balancer controller, so we decided not to support in the entry provider because it was deprecated anyway. So.

A

E

A

Make sense, okay, cool that just leaves uh so this one which is going to be closed and then this one which storage will get. So I think that's it for triage today.

C

I actually had one question about one that we already triaged but don't really have good answers to um 106 543..

C

uh I replied to the person and basically told them: hey go over to the kubernetes cigs azure file csi driver repo, but since we can't move issues between kk and the kubernetes org, I don't think I don't think you can move issues between the orgs and so I'm not sure like. What's our next step, if we think we've told you that this is handled by you're trying to he's trying to use, um you know some stuff in the legacy entry or sorry the the azure files provisioner that he's looking at like.

C

If you look at it, there is some stuff in kk, but everyone's telling him use. You know um the csi driver like what's what's our take on that? Maybe this is the same storage question but like what's our take on that.

A

Yeah, like it's a similar pattern to like some of the other stuff, where we're saying like all the external, like the new low bounce controller, supports it um so saying you're like kind of redirecting to csi driver.

A

um I I think that I, I guess, yeah asking folks to create an issue on the csi driver, seems.

C

Yeah well look up and see what I said in december, but I'm like I'm not sure. I just want to make sure I'm scoping this correctly. I suggested to the user. Maybe you want to close this issue and then a different user came in and was like. We have the same problem and I'm thinking I just I don't know if we want to direct people to sig storage or tell people to go to the csi drivers or something else you know what I'm saying like. I want kind of an authoritative.

C

This is where we send you now, but I'm not sure.

A

Yeah it's tricky because we still have code in kkk for specific providers. um So, like there's an I imagine, there's like an overlap of users that are like using entry and like some users using csi.

A

So I I think the hard answer is like, depending on the specific issue and the way it overlaps with the code entry and the code out of tree, it's gonna depend on that. I think.

C

Yeah and like this one I dug into it back in december and basically azure red hat openshift. So it's it's a red hat, managed thing, but on azure um does not support what they're asking for, but if they use the you know, azure file like csi driver that does have the nfs support that they want. But it's like yeah. This is kind of one of those like they're in that kind of gap of support.

A

Yeah but yeah, I agree when it starts going to like specifics to products like probably not. The best place to like a cake is probably not.

B

Okay, so what it's worth that csi driver is still getting delivered in openshift, so I imagine that I think you're talking about the rosa product, I think, is the azure or is it aero? I think.

C

Aero yeah euros.

B

The azure one yeah so like I have a feeling that, once that csi driver gets deployed to openshift, then like the next step will be it'll, make it to arrow. So it's going to be a little a couple cycles, probably before it gets there.

C

Right exactly anyway, so I just I wanted to bring this one up just because this issue, while having been triaged, is still an open issue, and when I looked at the the collection of open issues for the sega was like. Should this issue even be open, or should it be here yeah? What do you think we should do with it.

B

Given what you were saying and what andrew was saying too, like I feel like directing people to go to sig storage, is probably the best move because like if there's a question about whether whether a new feature or issue should be brought against the csi drivers or against, what's still in kk, I feel, like that's, probably a question that sig storage would be better. You know better equipped to answer. At least you know, because if every one of these has to be kind of dissected to say, okay, does it fit here? Does it fit there?

B

Maybe being closer to the experts is, you know, is better. I don't know.

C

What do you think andrew do you want to call out to stick storage on the on the issue and just see if they want to weigh in just because it's still sitting here tightly or sega club right or so.

A

So from historical context, I think six storage has made it clear that they don't own, um like provider, specific drivers and they that's. Why.

E

Like you could go.

A

To the kubernetes community, repo and stuff, like the ownership of the various like azure csi and like aws, like they're, technically under rsig, um because there's experts yeah, like the extraordinary, has to manage, like all these other, like general purpose, csi drivers too, um but yeah. I agree like the expertise, kind of falls, both ways um I think for this. One like we couldn't just close it: okay, but I'll yeah, bridgette I'll. Let you do that.

C

Yeah yeah, I mean I'll uh I'll, just give them another reply and be like hey. You know, as we pointed out, please go look over here if this doesn't solve your stuff. Here's the link again to the arrow like support and.

B

C

Slash close, I I just I don't want to leave it just sitting there but like. I don't think that I guess what I'm saying is. I don't think kk is going to give them an answer here.

D

G

Yeah as background the way this came about is because I was involved with the original csi.

G

The drafting of it felt that it was at the time it was done, orchestrator agnostic, so it was done for apache, mesos, docker, swarm and kubernetes, and the ownership is over in the cncf under the cncf storage sig. But even they kicked the csi things kind of sort of out of tree, but if there's any org that is closer to or owning csi, it's that cncf storage sake and not anything under kubernetes.

C

Right right, absolutely, okay! I will just write it back to the user and explain this again and uh close it. Thank you.

B

uh I was gonna offer it to you. Did you want me to try and poke a red header about dropping a comment here from one of our storage or arrow folks, like do you think.

C

If you, if you want, I mean like the like, I looked at the documentation as you can see where I linked them. On december 9th I looked at the documentation and, like arrow, doesn't support the azure file. Csi driver right now. Will it someday yeah? Probably it doesn't right now so yeah, okay,.

B

I'll see I'll see if I can get someone to make a comment on this too, just to just to help. I.

C

Mean if somebody wants to give them like road map stuff, that would be fine. I don't have that at least myself, but yeah.

B

Me neither, but I know like I know some people who probably do have it and they could just like say: yeah. It's coming. You're gonna have to wait six months or something you know.

C

Yeah yeah, okay, thank you.

A

I do think um you did raise a good point, bridgette, that, like we probably have a ton of triage issues that are assigned that they're still yeah um yeah. I.

C

Don't know I don't want to open uh getting worms, I suppose we can get to the rest of the agenda and then come back. If we have time.

B

What was that issue number just out of curiosity? Oh.

C

This was um 106 543. I'll drop you the chapter.

A

Thank you. Thank you yeah. Maybe we can have a discussion about these another time. um Okay,.

C

Yeah, anyway, sorry for the digression I was just like this is still sitting here.

A

uh Okay, cool, uh I guess we'll go through the project- updates quickly: um okay, cool! I don't see anyone from these two here, so we'll go straight to aws any updates or not. Here I.

E

Don't think I have anything um uh he showed you everything.

F

No, I don't either for this. Thank.

A

You uh bridget anything on the azure side.

C

No nothing specific! Since last time we chatted um happy lunar new year, cool.

A

uh Gcp, am I the only yeah I I don't have anything I'm still on boarding. To be honest, so I don't, I don't, have anything sad um ibm, no update. Actually, you know what I realized. um Michael I'm wondering, if, like I, I know like red hat, doesn't have or like openshift doesn't have a provider but like given there's like all these issues that are crossing openshift and cloud providers. Like I'm wondering, do you think it'd be worthwhile to add a sub project update here for openshift or do you think not even.

B

Well, um probably not, um but like I mean I'm happy to bring, I try to bring issues that we're working on because you're right we work across many of the providers here, like alibaba aws, azure gcp, ibm cloud, openstack vsphere, I mean so aside from baidu and huawei.

B

You know yeah we're touching all those things, but usually we don't. Unless we find a bug, there's usually not much interaction for us or no real updates. We do have requests for features every once in a while, but like they haven't, come in really heavy on the cloud providers. Yet so I'm not I'm not sure that it makes sense to have an update from the red hat or from openshift.

A

Or maybe another way to put it is like I think anyone should be. Anyone should feel free to like add things under the provider like you, don't have to be a maintainer of the provider to add notes so like if you run into bugs with whichever one just like drop a note um and we'll just take it like that.

B

All right cool yeah- that makes sense- I mean I remember- I've brought up gcp stuff here before and I think I see you've already got this uh 107 750, which has gotten a lot of attention inside red hat recently. So yeah I mean I'll just keep bringing up issues as I see them, but happy to put them under the provider-specific columns.

A

Good, okay, this year, I'm guessing that was lebron. Do you want to give an update on the 123 release yeah by.

G

The way I put it later in the agenda, but I just thought that uh andrew, since you moved to google vmware, has recruited a couple other meeting participants for this meeting. So maybe you want to invite lebron to introduce himself and then there might be another person who intends to start attending these meetings as well.

A

Yeah good call uh lebron: do you want to just introduce yourself.

H

Yeah, hello, everyone. uh This is lubron. I used to work with andrew in vmware in the same team, uh so we are both maintainers of the cloud provider. Vsphere.

B

Welcome aboard loubrow.

H

Welcome nice to meet.

B

D

H

Yeah updates will be just uh we got a 1.23 release next week.

A

Awesome on the extraction migration, uh I think there's just um just reviewing cats. I think there's not much aside from that. um So if you have okay well, actually we'll just talk about caps later, because that's kind of a bigger topic, um so yeah, let's go straight into the agenda. um I did want to follow up with one action item that we had about adding more approvers to legacy cloud providers.

A

So last week we added nick in there, so hopefully he can help us um go through the backlog of cherry picks that need to be approved.

A

um And then yeah the next item, I put two uh they're, probably the same uh topic, but there was one pr I think from someone from red hat. I think.

B

And then there's a.

A

B

A

And there's another pr from someone from aws. I think where? hmm Where do I start? Okay, maybe I'll start with this one.

A

So so this one is fixing a bug where the ipv6 address, like we weren't doing a proper uh equality check on the v6 address, which resulted in like addresses like this, that get truncated to to not be equal, because we did a string check and then there was some top there's some talk about, like uh always putting the canonical ipv6 string inside the status- and I was wondering like that seems correct, but if we did that we could end up.

A

Breaking clients somehow, and I was wondering if anyone had thoughts about that like if we forced all these six yields to use the canonical like representation. Would that result in, like some controllers, trying to go into some kind of loop to like update them?.

E

That would be a bug in the controller. Arguably.

G

Yeah, well, it seems like outputting the canonical representation in logs or whatever seems smart with ipv6 there's a number of valid input formats, and it strikes me we should take them all so long as they're valid.

A

Yeah, that's a valid point too, like we should always compare them using the same.

G

Yeah, when you do the compare, I think the only way to have it be reliable is to take both sides of the compare. If you're going to do a string, compare you'd have to convert them to canonical to have that be valid. If you're going to use a string, compare.

A

Okay, yeah, I think that's why we opted to keep this pr just to update this, um but okay yeah. But if anyone has opinions of like what the field should actually be set to um yeah, we should. We should talk about that, but this is also a heads up that this bug seems like if, if you're running, v6 and you're, one of the entry cloud providers you'll probably run into this book.

A

There's also like this bug about um what the expected behavior is. If you set the node ip flag in the cubelet and you're running an external cloud provider and I'm on the fence about what we should do here so like this pr is proposing that we shuffle the address sorting so that the provided node ip is always first, which I think is safe. But I I wanted to hear what people thought.

E

On the call, so I'm interested to hear to understand what you think would be dangerous about that and if we don't do that, then is that flag worthless and should it be essentially deprecated.

B

And just and just for a little background here in case you know not everybody's up to it like up to speed like the problem that we're seeing is that when we bring up a node and like start the kubelet, we have one I this happens a lot on openstack, but it also happens on. um I think we've seen it happen in vmware as well.

B

If I'm remembering that correctly, but, like you know, when we start the kubelet we'll have one ip address, that's kind of like the main ip address we want to use and then, when the ccm comes in, it gets multiple ip addresses and it chooses a different one right, and so that's causing this dissonance. When the you know when the node object comes up and then when we try to when we're trying to link the ip addresses, so that's kind of like I know that the folks internally at red hat, have been working on this they're.

B

Look. They want some way to like authoritatively know which ip addresses are going to be supplied there, and I think what uh stefan is proposing or maybe matt's proposing in this one. Is you know, yeah the ability to more positively know that when we inject that ip at the kublet using the flag, that that will also appear for the ccm so that we don't have this dissonance between the two places.

A

Yeah, I think that, like basic yeah, to make it like more deterministic.

B

Right and and again this like from the controller side from the operator side like we know which ip address should be right on the openstack platform, because we know like the ordering of the networks and whatnot, but it just seems like you know. Sometimes the ccm comes up and just chooses the other one or it's based on the host operating system, how it's ordering you know the nics when they get listed by that you know by that process. So it's like it's not always deterministic the way. We'd like it to be.

A

Yeah, so so going back to nick's question, um I I think it's only so so generally, like the pattern is that when you set the node address, you set like one internal ip, one, external ip and one uh host name and generally the ordering in that sense doesn't matter because the cube has its own, like ordering, where it checks internal ip for or like when you or like in cube api server. You can set the preferred like address type order, but I think the problem is like.

A

If you have multiple internal ips, then you can't the order is then random from cubelet and that causes issues, and so that's why the node ip being first is useful. But then I'm I don't actually know why it could be unsafe. But it is like a change in the api and I'm just trying to think through how it could be unsafe. Before we go ahead and just converge, it.

B

I I feel like if it's unsafe, it's because someone is doing something unsafe already. You know they're relying on a specific ordering to appear when it may not be predictable. You know, and it might have just been working for them so far, but like these things should be very explicit at that level in terms of what nick you're grabbing and everything.

E

What's the state, what was the um like? I don't know original description of the note, ipflag um or original intent of it. Was it just to.

A

E

A

Ensure that, like one of the addresses matches the note ip, but it had no no semantics around ordering.

C

The smacks of load-bearing bug.

B

I mean yeah, it's bringing down our house for some deployments right now, so yeah for sure.

B

There was another question about that label about that um that flag too- and this is something that came up during our discussion- I'm just gonna- see if I can find it quickly. There was a a well-known label that used to be uh that was marked as alpha and I've never seen it marked as deprecated um the alpha.kubernetes slash provided node ip.

A

That's what it is yeah same thing.

B

Right so like this is one of the questions that I had when we were talking about. This internally is like: is that annotation or is that label still in use, and how does it match up with that node ip uh command line, flag.

A

Yeah, so the note ip flag is one of those weird things where it is a cubelet flag, but when, like during the external provider transition, we sort of had to we're kind of forced to port over some of that functionality into extreme providers.

A

um So, basically, if you set node ip and you're, not if you're not sending a cloud provider, then the cubelet uses that ip when it registers itself.

A

um If you are using external cloud provider, it sets that alpha annotation on the node object and when the node controller sees a node object. It can then read that ip through the annotation and then it'll validate the returned addresses from the provider and ensure that that ip exists. Otherwise it it uh errors. The initialization and the node never gets registered. So it's kind of like a way to force to ensure that a node has the right id before it's actually accepted into the cluster.

B

Right, but I mean isn't that like to me, this label was that seems like the mechanism that we should be using to synchronize between the kubelet start and the ccm ordering right, because, like the ccm loads up and gets some list of ip addresses, and if one of those ip addresses matches what this label is that's the ip address it should choose, because that's the ip address that the kubelet was set with the flag so like when we were talking about this internally and no one could answer what that label.

B

That label was used, like my thought, was like this label is effectively the communication between these two processes.

A

Yeah and and almost always, it does do that um because it ensures one of the addresses is the node ip1, but yeah like there are certain cases, especially like with dual staff, being more common, where, if you have multiple internal ips, then the order matters, and so then we need to like change the current node. The way the current node ip behaves and also ensure that the order is first.

B

A

B

On this label right because like we wouldn't have to care about the ordering, if the label is always correct for what the flag was said as because we could always check if that, if the flat, if the label matches one of the ip addresses so to.

A

B

A

We already checked the annotation for an address and- and we already translate that into the status which is starting but, like you have to update the status, though, because the status is read by api server and the cubelet and whatever else.

B

Okay, maybe I'm not understanding that.

E

Part of it so with with cubelet today, if you set note ip, does that ip um show up first in the list of addresses.

A

It it could show up for us, it's not deterministic.

A

Yeah, so so, michael, like maybe another way to put it, is the note at this the address list, the list of addresses in the node status.

A

That is the actual api for how the entire system, like knows like what ip to use to connect to node the alpha annotation is just a way for this for the cubelet to pass the preferred node ip into the ccm.

A

So the ccm knows which ip to put in the status eventually, but like it's the status that is really like the actual api and like the annotation. It feels kind of like a hack that we're using to just like pass that preferred ip down.

B

Okay, so like, given that then yeah I'll flip back to the notion that the status block then should be ordered the way we want it right yeah, so that that does make sense to me.

A

Yeah and then this pr is basically saying like if we have the annotation, uh which is like here somewhere here um so yeah like this function, ensure node is checking this annotation, and this is just instead of just checking that it exists. It's saying shuffle it so that it's always first, which I think is right.

A

I think that's right. I just I'm just trying to think if it could break anything, but maybe it's one of those changes where, like it's worth it to maybe break some like some of those like weird clients that are doing weird things, so that.

B

I think so yeah, I think it's worth it to break it and see who breaks and then just fix those places.

A

Okay, anyways, um I think we, okay yeah so um yeah. I guess action item. Please review this pr and uh provide feedback, but um yeah I'll also review it soon, because I know it's done breaking some things.

A

Okay, uh bridget, I think you have the next item.

C

Yes, I wanted to follow up because we had had a topic of oh hey. uh I believe the gcp folks had noted that there was a dependency deep deep in kubernetes that seemed to be pointing at azure, and that was sort of weird I followed up. I asked internally as like you know, following the chain of who maintains this specific thing and they said. Oh we're going to be cycling that out and I'm like. Okay, because I said you know: do we want to go through like cncf processes or some other foundation?

C

Is this a project that should stand on its own over in some other? You know governorship and the the short version of what I got is: oh we're actually going to be moving away from that into something else. So I'm not sure if that's a complete solution or a complete answer, but it's a we probably aren't going to move that specific github repo at any point in the near to distant future.

C

So I don't know if that gives you enough answers. um Andrew was it you who was asking about that? I forgot.

A

I don't think it was me: okay,.

C

It might have been walter.

A

C

We can follow up with walter when he's back as well, but um anyway, yeah.

C

And this is uh this was surfaced because of how um the I think it's the transit of dependencies show up now, whereas they didn't before so.

A

Yeah makes sense.

A

Okay, okay, cool yeah, we'll follow up with walter on that nick new meeting times.

E

Yes, um so I I had a poll in the city cloud provider, slack channel, it looks like um a few. People have have answered it. So if anybody here hasn't yet um go ahead and do that right now looks like the morning times are winning uh so the uh I believe 9 a.m and 10 a.m. uh And it's all for wednesday, I didn't propose changing the day. um I think 9 a.m and 10 a.m are are winning right now so um yeah, if you strongly disagree with either of those times, then be sure to vote.

E

If you haven't already sorry.

C

Is this in kubernetes standard time, which I assume is pacific.

E

C

E

Yeah yeah right kubernetes standard time exactly um it is specific and yeah. um So I think I don't know we can leave it uh up until I don't know end of the day today um sure or something along those lines and then we'll we'll see what's what's got the most votes and we can reschedule next meeting for that time.

A

I'm cool with making it affect the next meeting cool.

E

Okay sounds good, that's it.

A

A

Okay, uh steve, do you mind if we talk about the kubecon stuff last and then we can.

G

And then you already talked about the intro one thing related to that: lubron got a bad zoom link. He joined the google group and then you get a welcome to the group uh email with it. I think an old zoom link, so I don't know where that's recorded the one in the kubernetes community calendar is good and, of course, the link in this agenda note stock is good, but apparently somebody new who joins the group gets sent a bad meeting link.

A

Okay uh lebron: do you mind following up with me on slack and pointing me to where that is.

H

Yeah sure, thanks for watching.

A

Okay, um yeah bridget. Do you want to talk about the mix protocol cap.

C

um So I think everyone uh or most people on this call already know this, because I brought it up a bunch of times and everyone's kind of weighed in and put their own changes and fixes in, but just giving everyone a hey, excellent. We are moving this forward so for all the work that you've been doing to make sure that you handle. um You know mixed protocol on services correctly.

C

Thank you and it looks like we are going to move from alpha the beta and just the reminder that moving something from alpha to beta means it's on by default. So there'll still be a feature. Gate and people can still turn it off, but we are moving forward with this. Thank you so much.

A

Awesome- and I guess, like the romanian action items is like if there are any places where we're not validating or handling mixed protocol, we need to just make sure that we.

C

Yeah, I actually think google is the one provider that was kind of like well, our users might get into a weird state, but they, you know, I have signed off from tim hawkins saying: go ahead, go for it, but presumably for your out of tree provider you're going to make whatever changes you need to make make your users get the experience you want them to have.

A

Yeah sounds good: okay,.

C

And you have a little time because this will be in 1.24, so the very soonest that this would be turned on by default for your users would be april when 1.24 comes out.

A

Yep makes sense awesome. Thank you, okay. um Steve! Do you want to talk about the.

G

If you want, you can click that link and just open it, so we can look at it while we talk, but this is just kind of boilerplate for that. Cfp um deadlines coming up on the 14th, as indicated here they changed some of the language and rules on this. But when I went to look at the form so they're specifically calling out that submitting it doesn't guarantee you're going to get a session. I think they've had a large number of these as they add projects. So you know in early days.

G

I think it was viewed as semi-automatic for sigs to get a session, but apparently no longer the case. They also cut the size of the abstract, from 750 characters down to 500..

G

So I put a tentative title there uh I'm not strongly attached to that in any way it's just kind of generic, but if anybody else wants to take a crack, just that's an added link to that document. So go replace it if you like, as far as I'm concerned uh the detailed description, there's a couple things that I was putting there look at that last line of we'll also discuss plans to handle tbd.

G

So this would be things that maybe we didn't cover in the last one that are new on the agenda. I think walter threw out a few in our last meeting, but I didn't keep it in the notes and I thought maybe I'd go back and catch the video, but it's not there yet. So I think walter actually had some ideas for this, but I don't recall what they were. So we just have room. We don't have a lot of characters.

G

Maybe this these two tbd's, if we can blow like 30 characters to describe a few other line items and as far as I'm concerned, I think this is probably good enough uh with 500 characters. You know you can't you can't do much other than just a broad description, and I think the only thing maybe open for discussion is I personally like that effort.

G

We did last time of kind of these um lightning talks for individual cloud providers that we took as recordings, and I think it might be nice to do that again, but maybe others feel differently and that's a good way to have the whole community involved in this session, and I think I learned something from other cloud providers by watching those lightning talks myself. So I'd like to do that again, but let's open it up for discussion. If somebody's against that idea, I think we'll. We don't need that in our cfp proposal.

G

We can worry about exactly. What's there, we'll just call out that we're going to have lightning talks in the abstract and get it in the schedule, so people know there's a slop for them.

G

Yeah sounds good and then oh, the other thing is they're, allowing up to four people designated as speakers. In my experience, you can actually shift those roles out later.

G

If you want, you know, you just need a couple people to anchor it down and they are still indicating that there's almost like seems to me like a preference for this being an in-person event, but they will entertain remotes and we could potentially do something hybrid, like we did last time of you know having at least one person on the stage uh playing the recordings of others and kind of roll. Our own hybrid, the cncf platform itself does not support hybrid per se.

G

You know where a person comes in last minute because the stuff has to get digitized and they can't guarantee equal latency from the conference site versus coming in from some random place in the world, so they are willing to do that, but we can sort of pull it off ourselves. If the live person just plays recordings- and I think last time they were able to support q a afterward if people joined remotely over slack, you know it's not quite the same as a zoom q a but it it worked.

G

I think we took a few questions like that in the last one.

A

Yeah make sense- I I don't. I'm gonna be going on pat leave in a couple months, so I won't be able to attend, but I'm happy to help review this and um get in good shape.

G

Yeah, so the whole group can be there and I think um either you walter nick is going to have to submit so the speakers have no particular requirements other than I guess they should be sig members.

G

But the person who fills out the form has to be they're going to compare to make sure they're the sig chair or tech lead. So you or walter are going to have to send it in, but I'd be happy to do kind of the work of maintaining the draft and get on that. But your one of you is going to have to be around to actually send it in on the deadline.

A

Yep, I can do that, but yeah. I think I also like the idea of like doing um giving like every provider like a slice of the time, I think the biggest hurdle there is like coordinating like more people to like put the put the content together. But aside from that, like yeah.

G

I kind of pieced it together and did the combo edit last time and the feedback I'd have on. That is that essentially, everybody was late. It's hard.

G

I declared a deadline.

G

You know sort of like walter's philosophy is you lie about the real deadline, but we managed to get it done, but I'd appreciate if it would have been easier if they had come in a little early because it got it's kind of treacherous doing a video edit on a laptop versus I have a gpu on a home system that would make that a lot easier than getting these literally.

G

You know I. I am unwilling to take them the night before, uh because too much can go wrong, but I don't really want to do it. Even you know, I'd like to get it done a week before if possible, and then the other thing is there were people who cheated on some that were too long.

G

You know we it's 35 minutes. So if and it's harder to make a short one than a long one, I know that, but some of them I ended up having to speed up and edit and things like that that maybe sort of moved them towards chipmunk voices and things, but would be better if people really did hit the three-minute cap on these. If we're going to get them from a lot of cloud providers,.

C

You know what you could tell people, and I know that I'm one of the people who just turned in the slide, because I had no bandwidth last time.

F

C

What you could tell people is, I will play the first three minutes of whatever you send and then I will cut it off, so you choose. If you would like to send me extra bonus content that I'm not going to use go for it. I will use the first three minutes of what you send me the end and that way it's in their hands.

G

Yeah, I don't know, and then the other thing is of course it. I didn't even realize it last time, but you're expected to publish a deck of your presentation and upload it to scad. So I really wanted the decks that people presented and I managed to get them, but it was extra work and this time I'm going to be smart enough to ask for that up front so that I expected.

C

That is skippable fyi.

G

Okay, but anyway, that's it for my agenda item.

A

Awesome, thank you. um So, given the next sick meeting, it's going to be before this deadline, um yeah it'd be good. If maybe like next week, maybe on slack in the slack channel, we can touch base and make sure we're all on the same page of.

G

Who's going to leave comments in the dock itself or in the slack channel, I I think either one would work.

A

Yeah or maybe um oh I'm not sharing my screen anymore, but I'll leave a comment inside steve's um document uh I'll write comment here: if you're interested in contributing content.

A

And then um folks can, just like add their names to that thread, and that way steve knows who's interested.

A

All right should we end 10 minutes early, then anything else folks want to discuss.

A

All right I'll see you on two weeks in.

B

Okay, bye bye, later bye.