Kubernetes SIG CLI, 8 Feb 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG CLI 20230208

Description

Kubernetes SIG CLI bi-weekly meeting on February 8, 2023.

Agenda and Notes: https://docs.google.com/document/d/1r0YElcXt6G5mOWxwZiXgGu_X6he3F--wKwg-9UBc29I/edit#bookmark=kix.8hz19v29qctm

A

I'm clicking record uh welcome to the February 8th edition of the Sig CLI bi-weekly meeting I'm Sean Sullivan I'm your host uh we'll get directly into our announcements. We have the enhancements freeze tomorrow and there's uh at least one of our caps is I, think in pretty good shape and I'll I'll leave that up to to Justin or to Katrina to discuss that when we get to sub-project and kept updates, but I hope you've been working hard on your enhancements.

A

If, uh since the freeze is tomorrow and then about four weeks later, uh maybe a little bit more than that is uh when we have our code freeze for the 127 release and we're scheduled for a release on April 11th for 127..

A

So does anybody else have any other announcements they would like to make.

A

Okay, so there's a part of our meeting we'd like to uh introduce newer members, those that haven't uh been here in a very long time or those who are brand new. uh This is an opportunity for the Sig to to meet you and you to meet the Sig. So if there's anybody who'd like to take this opportunity, please go ahead.

A

I think I've I recognized just about all the names, so I I, don't think that we have uh too many here uh that are brand new. Why don't we move on to the next item then, which is the sub project and kept updates?

A

So would anyone like to I'm gonna I'll bring up the the the prune apply set cap? uh Is that visible now.

B

Yep, yes, it is.

A

Okay, great, would anyone like to uh discuss uh the status of that.

A

Katrina or Justin.

C

Sure, uh just like it looks like it's going to be good to go for tomorrow, so we're super excited to start actually uh working on it. We kind of paired on it a little bit last week, but uh we're gonna be uh writing the feature in Earnest, uh because it looks like it's gonna go in.

A

A

Okay, so we have also would any uh Katrina. Would you like to talk about customize or anyone else? uh An update.

C

Yeah I should have said like just instead of the chat. Thank you to everyone for the excellent comments and feedback. We received quite a lot on that on that cabin, and that was really awesome.

C

um Yeah it's about customized, though uh we finally managed to get through the version 5 release, with all the features that we wanted to get in there pretty much last week, so um I believe I I put the link yeah there we go.

C

uh We tried to do very extensive and detailed release notes to help folks uh out this time. So, if you're interested in customized, please have a look at this release, give it a try. um We were really proud of all the work that went into this, uh both from the core team and all of our contributors. Wow.

A

Thank you. Everyone.

C

Who participated.

A

That's that's really big I've seen in slack a few people actually sounding very excited for this next version.

C

Yeah we have some really great stuff in there. There's performance improvements, there's a major new localized feature. uh Thank you to Anna for all of her hard work on that over the past months.

C

um There's new home fields that we added by popular support and then tons of bug fixes for um various areas of customized, um so uh yeah, and we are intending to propose this for cube cuddle in uh before code freeze as well um as usual. We bump it and keep cuddle after a period and for this one we're going to wait a few weeks before we propose that bump to see.

C

If there's more bug, reports that come in because if there are then we would, you know we would probably do uh 5.0.1 release with any fixes before actually pushing it up to to keep Kettle before the fridge. So we're waiting a little bit to do that, but we do intend to get version 5 into uh Cube cuddle, 127.

A

A

And uh so is it okay if I move on to kui? Yes, thanks Nick, would you like to talk about uh kui, 13.1.

D

Sure um and the release notes: you can see shots um some bug fixes. We had some reports of not displaying well on phones. Those are fixed, some more load time improvements, I think we're down to about 500 milliseconds for a little time, pretty happy with that and then in terms of features.

D

um Finally, Helm status and there's a container tab for pods. So that's about it.

A

Great, thank you very much for that and the customize function catalog. Let me move to that particular link.

E

A

I just wanted to.

E

Mention that uh I'm getting started working on this, so if people have opinions, let me know there is a PR open updating the cap itself, uh but probably don't look at it. Yet wait till Katrina does and.

A

E

Look at it so it'll some stuff will be out of the way, but if you have ideas or anything, please let me know because I will start digging into this soon.

A

C

Thanks Caitlin, that's a super exciting feature for customize, um so I'm really really happy to see it moving forward.

A

Excellent is there any other sub project or cap updates we would like to address.

A

If not, we can move on to the open discussion.

A

Okay, so the uh it doesn't look like we hit. We, it looks like we have about three items here. uh The first item is one I put on uh there's a recently Resurrected um issue. That's been around for a long time and is uh is starting to get more attraction, um which is websocket support for uh Coupe, cuddle, exec and attach, and also port forward.

A

So I'll give it just a little bit of background on what this is. There's nothing that the Sig needs to do, but I'd like to raise its visibility, because I think that there's going to be some work in this area.

A

So there's a couple Coupe control commands that would significantly benefit from Full duplex communication over a single TCP socket and those are the three Coupe cuddle commands that I just described, and so initially uh the way that they were this protocol was implemented for these commands was was speedy and Speedy was a Google proprietary protocol that eventually became HTTP 2.0 and there's some slight differences between those two, but because of that, Speedy has been deprecated.

A

It's been deprecated for probably eight years when HTTP 2.0 was uh was ratified uh by the ietf in back I, think in 2015, and so we there there's been many different uh uh complaints about exec, not working in certain situations um or not just exec, but these these three Coupe control commands because of Speedy and because of the deprecation and because this particular uh protocol has been deprecated for so long. So the API server already supports websockets, and this is another protocol which implements this full duplex communication over a single socket.

A

And so there is an effort to get these commands to understand or to speak. This websockets um I think eventually we want to get to http 2.0, but uh I I. To be honest, I, don't think that that's supported in the API server, but we do know that the API server supports websockets. Now, although uh Daniel Smith was saying that you know there may be some bugs, it's um and we'd probably want to test that out uh in order to get a quick win.

A

I'm, pretty sure that we're going to be implementing uh websockets in these commands and the way that this works is you send an HTTP request, and one of the headers is an upgrade saying: I want to upgrade to websockets, so so I just wanted to raise that.

A

Raise that visibility to everybody that um anybody I think actually a couple of us uh myself included, might already have been given.

A

uh There may be some consensus that some of my time could be worked on this, and so I might, uh and there may be others as well, but whoever would like to to be involved in this, uh please. Let me know.

B

I remember: there were some earlier approaches at websockets, I, I kind of remember what happened with those vrs, but over the past years.

B

um I do remember a couple but I'm guessing that they didn't went through the reviews or yeah.

A

Right so if you look at this uh issue, it started about three years ago, uh dim started it and it's actually quite involved, and so it goes back and forth with a bunch of people, and there was actually one person, uh Mark Borstein, who I think got most of the way there and it it kind of just fell apart. It looks like um and it started the the issue itself started, becoming stale and rotten, it's been frozen by uh by Daniel Smith just this last week.

A

um But but if you have a look throughout this issue, you could see the history of where we've gone with it.

A

B

Like I said, I I found the pr that I have in my notes, uh which was trying to touch the exact and touch.

B

I wonder if that's limp in the original, oh.

F

B

It is but that's good, and that was a that's not all done from a year ago,.

A

Yeah I know and I also heard that Chao uh had started working on this, but um I think that also fell by the wayside.

A

So I'm gonna I'm gonna, try to get this pushed over the finish line and was hoping to to bring this to your guys attention and to uh benefit from because I I think that there's been dozens of people who have been involved in this and um I'd like to kind of Corral it and get it over the finish line.

A

Thank you. Okay,.

C

On that PR says, Uh there's a like at the very end right before it closes there's. It says it is clear that this change will require cap to move forward. Is that still the thinking.

A

uh That that'll be part of what I'm gonna dig into, um but I'll report back um back to the Sig on where we are on that so yeah. If that's the case, then it won't get into 128.

A

Because yeah not probably not going to have a cap within the the next day, yeah.

C

That would be challenging yeah.

G

That's it um were there any powers.

D

G

And I saw that somebody mentioned that they were trying to implement this using gorilla's, websocket library and.

A

That was yeah that was Borstein, who mentioned that um and that that actually did sound like the correct path. Again yeah you, you disagree, I, gorillas, it's archives; okay, so that might have been a while ago, yeah.

G

um Your entire, like HTTP Associated libraries last year at some point, which is very unfortunate because they're really nice and easy to work with, um but yeah.

A

Okay, I'll put that as a note, yeah I think that there's just it's a sprawling it. If you start digging into these issues, you'll end up with you know, 100 tabs in your browser, so so I think a big part of the effort is going to be just to kind of Corral. It all organize it all and uh see what what gaps need to be filled in order to get it across the finish line.

B

A

Don't okay, we.

B

Are on the same topic from my notes, uh it's also linked in the original PR yeah. There has been multiple or just let.

H

G

B

Supportive of the uh of the episode of the approach in general trying to push this forward.

A

Cool excellent.

A

Can we can we move to arda's issue.

A

Okay, so uh let me share this particular tab. Ardo. Would you like to speak about improvements around Coupe control, delete minus minus all right.

F

Yes sure and that's an interesting case because simple mistyping caused a traveling clusters. Users wanted to delete PVCs in one namespace, but he or she wrote persistent volume and keep control, discuss the namespace and delete all the person's tutorials in the Clusters and I. Think we can.

F

This I think we can add some kind of validation or even block uh deleting cluster-wide resources if the namespace is explicitly set and that's why I wanted to bring that topic and because of bad foods, compatible issues. I, don't know, I, wonder your opinions about it.

C

Isn't this the same issue that inspired the keybar CCAP? Ultimately because it kept coming up and we decided we couldn't do it because it was breaking.

B

Yeah I'm just trying to find uh uh the original discussion that Eddie kicked off some time ago. Okay, I have the link, uh so Eddie had an entire conversation about the topic.

B

Almost two years ago, it's like May 2021.

B

There has been multiple uh uh argument in both directions.

B

um About the flow, how when should we change it? uh The the general outcome is that cluster admins are free to add um validating web hooks if they wish to prevent accidental delays, uh which would work whatever process that they want to have around uh the confirmation.

B

Most likely Cube cuddle will not be changed uh due to backwards credibility, but that's where we came up with the cube RC such that users are able to change the defaults for a certain flag um and another one was the introduction of the interactive mode or delete not sure if the interaction was ever implemented in delete, but the interactive option in combination with QRC, where you would set explicitly the default to always be on in case for some users, uh this could give you that kind of uh uh guarantees, but in general,.

E

F

In terms of interactivity I understand we can use trip RC, but the command is wrong. We are trying to delete TV by also allowing settings. Namespace um I think we should just prevent this. If this is a cluster right resource and namespace is passed, we should block that.

B

Yes, and no, because you can always uh articles the biggest issue, is the pattern of all uh if he, if the author would not pass with that at all, the original issue would not happen in the first place.

B

uh uh In parallel to that, you can always pass a mix of uh cluster and namespace scope, resources, in which case you do expect that it will work just fine and figure out what you care about and correctly identify the cluster's code and the namespace scope and apply the provided namespace to the ones that you have and unfortunately, by the fact that the lead is one of the commands that has been out there, the longest, the the backwards compatibility and the potential Fallout of changing the defaults is bigger than uh and the cost, especially if we look through over the past.

B

A little bit over but I would say five six years. I would say we had like four or five issues. Eddie uh Eddie addressed an issue a while back. That's what trigger taste, uh email uh threat. I was trying to approach a similar problem even earlier. I think it's. It's also mentioned in that email, but they're just jumping into something similar and for now we're pointing people to the coupon C effort that should allow uh to prevent them from similar cases and other than that.

B

There are options how you can prevent uh from those kind of situations on the question level.

F

Okay thanks, probably the most thing for me is I thought you can only pass PV or PVC, uh but if it accepts multiple resources, one time yeah it.

A

F

A

So a couple quick questions should we uh should we say triage accepted on this issue?

A

B

That sound, reasonable I would probably try to find the other the.

A

B

uh Either the one from Eddie I think he has um a couple of issues: 1056 he's pointing out to 1056., uh so maybe just not sure what happened with the 1056. Let me open it up uh that got closed.

B

Yeah, because it has a uh with the entire explanation, so I would just link Eddie's comment sometime uh 56., specifically, this comments.

B

And close the issue.

A

Okay, so we have a conclusion on on where we want to go with. This sounds like and.

H

One more thing on this like another single way to look at it, maybe is that the all flag is a problem. Is that I mean instead of like for giving people away to not accidentally delete everything it seems like even the existence of the all flag is not even a thing that we should have.

B

Good question that was probably a good question, um but I'm guessing you already know the answer that um we can't change it at this point in time.

B

uh I I know why it if I remember correctly, why I always added it was added for convenience.

B

uh Unfortunately, as with every convenience School, uh it has an equally uh high probability to hurt yourself as much as healthy yeah.

B

That that's the problem, but yeah interactively basically the the simplest most straightforward way to fix it.

B

um If anyone is interested in adding interactive to Duluth, if it's not there yet um I'm happy to uh to review that kind of a PR and then help with the QRC uh work.

A

Thanks so Arda, do you think that you have everything you need on this? Yes,.

F

Thank you uh I understand to use the point of view if you create a temporary namespace, and do you just want to delete all of it all of the resource in it, you can use dash dash all but yeah I understand the problem here. Thank you, foreign.

A

That one I'm going to move on to the server-side apply this default and share that tab.

A

um There's a brand new cap uh with mache, or maybe even Antoine like to talk about this server-side.

B

Obliged if Antoine is on the line, I'll probably pass it on over to him because uh he authored the cap uh I stumble upon it. Literally yesterday or the day before yesterday, uh yesterday's and um I'm open. There will be a discussion about the topic a little bit later today during the prr review. uh Jean had some concerns about changing the default and we need to figure out how to approach the topic.

B

I'm, not sure if Antoine is planning to Target 127. With this.

I

B

I

I mean the the consensus. Is that uh it? The timeline is not, you know very well. We don't have a lot of time to discuss that so I'm open to postpone that until the next release I. If does anything, we can do these release to prepare, though I think, would be a good idea.

I

um Oh, it's been delayed a long time already. So if we can move with that, I'd be happy.

B

I'm, pretty sure that, uh whatever the outcome of today's this uh discussion will be, we can actually within 127 start adding some warnings around the user usage of the uh of those flags to prepare the users that in the next version, we will be doing whatever change. We will decide to do having that extra time will be definitely beneficial for for the users.

B

Thank you, oh so Antoine. Do you want to quickly discuss the the idea behind the cap just to make sure everyone are on the same page.

I

So I got to find a different room. What did you say my chair.

B

Can you quickly describe the idea behind the cap uh so that everyone in your room are on the same page.

I

Okay, so the idea we had with this cap specifically was: um we had many discussions, so I can't even remember what I proposed in the camp, but um I think we wanted to switch. The default of server side apply to being Auto and auto being a new flag, the flag meant, uh if you have some resources, your clients are applied in the past, who are going to continue client-side applying these resources, but new resources are going to be solo set applied and uh the resources are going to continue being server-side applied.

I

One time one we can have today is that if you have some resources that are the inside applied and some that are so also reply, it's pretty hard to. You have to only apply the the one you want with the right flag and then another set with a different flag which can be confusing uh and we've seen. People go back and forth between client-side and server side by accident, because they don't specify the right value and uh we are doing our best to get a good migration from 1pdlr.

I

But it's not completely perfect, so we'd like to avoid people going from client-side to server side to client-side constantly by accident, so Auto detecting uh was an idea to help people to prevent people from uh doing the wrong thing.

I

um We were arguing on what do we do with new resources?

I

um We could continue to plan cytoplay them or we could show us how to play them, which would go toward the goal of like having our knees or outside of play eventually, uh but that's going to break a lot of what what most small people I think, mostly because of conflicts and I, don't think it was. The ICD systems are necessarily ready for conflicts.

I

um We also just got discussed uh with Katrina the idea of like forcing conflicts by default.

I

It would basically uh it would migrate people to server-side reply, but most people would not realize I would say the vast majority of people would not realize uh so that would allow us to migrate people to Southern server play without them noticing, um which is good. The problem is, we lose a lot of the value of conflicts um and it sounds hard to switch back the value of false conflicts. Later um so, we quite don't know what to do.

I

To be honest, the proposal was one of these ideas, but we have multiple ideas and we're still trying to weigh the trade-offs, I think, and we don't have time.

C

Yeah I've been I've, been to Swift and Alexa slack a bunch and going throwing ideas around because we want. We have these goals, but we wanted to make it not disruptive for users and make them benefit from the value and there's a lot of trade-offs there. So it's tough.

B

Thank you I'm personally, supportive of the change. My only.

B

Objections about The Silence, which was the one that I first voiced in the cap where, um because you rely on the fact that, if something exists then you use server-side, uh you use client-side and it has an annotation. You use client side. If it's something new and doesn't exist in the cluster, then you switch automatically to server side.

B

My main objection is that if we have a one directory or one invocation of people that apply, you might end up with all the resources applied with client-side and your resources applied with the server side.

B

That's something that I would like to understand uh other than that I I, think that is reasonable and, um although maybe I would put some some kind of information that, oh yes, we actually picked the server side just so that you are aware that this is what you should be doing in case. You figure out that you you do want to.

I

How how do you feel about uh cicd system that are going to break if we sort of start to play some resources by some price.

B

That's that's the that's one of the use cases if we start doing half on server side and half on the client side, everything else should be just fine. If it's like brand new, you should be fine using the video approach are.

I

You are you saying: if we had several set of play, equals true, meaning we have no mixed uh mixed objects. uh You'd be fine with that, because that's still going to break systems, but the system is easy to fix. You just add dash dash server side, your portfolks, if it's a system, it's very easy to update I, think it's it's going to break.

B

I in case you have a mixed, I'm, not 100 sure what would be the best approach. Yeah.

E

B

Next case and the ability to prevent the next case is something that I would be using.

I

Yes, I, don't think we have a lot of mixes today because it's hard to have mixes anyway. I, don't know if you have an opinion about that. Katrina I'm with that exactly.

C

I am I, don't have a an opinion about how common it is. I have no idea to be honest, but I uh I guess I want to dig into like why that's bad, like apply does actually is on a per object basis, so um I get. It can be confusing if the default like, if auto mode, does client-side on one hand and then server side with Force conflicts um disabled on the other, then you're gonna get very different behaviors there, because client side behaves as though it forces conflicts.

C

So that would be like what the heck is going on. But if you had, if the combo were instead, like auto mode means client-side for existing and server side with conflicts, Force, which is very similar Behavior, just like upgraded so to speak, to server-side, why?

C

um What would the risks to the end users be? If that's, what auto mode meant.

I

But so do you want the value of false conflict to change based on the the value of server side? Is that what you're saying.

C

I I guess that's what it comes down to I'm, more thinking from a behavioral perspective like if I'm an end user. What do I expect and how do I get consistent, Behavior here, like um the the consistent flow Behavior like because, if I'm thinking about like the average user, they have no idea what server side, client side means right and to them like the meaningful difference is: is that difference in Conflict handling, so the transparent upgrade path that comes to Optimum to um forced conflict? Server-Side apply?

C

So that's that's sort of my train of thought like that. That's what we could do automatically so auto mode could potentially do that which yes, you're right in in practice, means toggling a flag value based on another flag value which, maybe is, is not ideal. Arguably that already happens because um Force conflicts um behaviorally. You know like what? What does that sound like to the user um for uh server side apply? It's it's defaulted on right or sorry, defaulted off.

C

The polarity confuses me when I'm talking about this I've default it off, um whereas a force client-side apply behaviorally, it's defaulted on. Yes,.

I

Yes, I think for people who have no idea what's going on you're right, the the default is actually not the same, uh but as soon as you've started using the server side of like like flag. Sorry, then it's very weird.

C

I

C

I'm saying like if the auto mode uh is sort of behaving like the client side mode of that server-side flag, then it's actually not any weirder than what we're already doing.

I

If we accept that um yeah I can see that I don't know if it's going to solve the plan, though I would, in six months eight months a year from now will be like. How do we get from Auto to something else and um we'll have the same discussion again and that's gonna. Make me sad, uh yeah.

C

I think Otto's just one facet of the plan that other conversation we're having about like what is the actual transition of these flags is, is important to have, and we don't have like a solid plan for what that can look like and be non-destructive, but Ottawa would be like I a tool to help people actually do this and and not be disrupted in the process. What.

I

If we actually gave people eight months and be disruptive, what if we told them we instead of wanting somewhere for eight months and in eight months we flip the solo side, driver and people will get conflicts, they'll be material. If they have tools, they can upgrade it in the eight months. If they haven't, they can update it since Sunday, it's not like the migration is actually expensive. Right by then we might have the QRC file.

C

I I think the the QRC is not as helpful because like from a safety angle, what matters is what we're talking about in auto mode that, like a given object, gets a consistent strategy. So you don't flip back and forth, and QRC is a user preference. So it's a per individual per actor preference, which is probably not the level you want it on, um but yeah I mean we do. We do have to do a lot of warning.

C

The timeline is super long because we're talking about GA Flags so like what is that, like a year and a half or something um I'm, just trying to think of like we want to give? If we're going to tell users that they have to do something, um we want to give them the tools to be able to do it as safely and disruption free as possible.

I

um So yeah we basically what you're saying is we have you all know how to make sure the migration is working perfectly, but all they have to do is just switch the flag if well, if they want to keep the old way right. Oh they, oh they integrate whatever they have to integrate the server side thing.

I

If they apply manually, it's probably better on your way. If they apply with an automation system, they can either specify the fly and keep the old thing all they have to. They have a young hugs to update the cicd system.

C

G

I

G

C

Disruptive that I think we need to find some way that, like they could not possibly miss the um whatever.

I

It is the warning: I I agree with you. They're gonna miss it 100, but then what the problem is, what do they have to do when they realize they messed it up? If it's like? Oh you have to rewrite all your controllers and it's going to take months. Then that's definitely the problematic. If it's like. Oh you have to change this one script to change this one flag.

I

I could live with that.

C

I guess the um there's also a question of like what is level of confidence and say we flip from clients of our default to server side by default with conflicts enabled you could say like okay, on the one hand, maybe the only thing they need to do is add. Force conflicts equals true. If they had a problem with this, they didn't notice the app Auto upgraded and they had the errors yeah yeah they had that flight. But how confident are we in just that transition itself?

C

Like you said we fixed a bug in it last release, it's not perfect. uh Do we do we actually like the secondary questions? Do we actually need something like auto mode, because that provides sort of an extra safety layer for people who are doing this transition because it makes them never actually transition any single object from one to the other.

I

Yeah, that's going to create a lot of it. I can see that I don't know. Do you have an opinion lives.

J

This sounds like I guess: I didn't understand that question. It sounds like we don't trust. Server-Side apply.

I

The transition are, we, do we not trust, solicit, apply or the migration code from client side to South Side reply. All both you can say both.

C

Migration, how much are we going to expect our users to trust the migration like if we're? If we're not confident in saying this is perfect, it doesn't go wrong. Then um Autumn, something like auto mode, sounds like super useful to the end user, because then we don't actually Force anybody down that path. Even when we flip the default.

J

I'm using the course complex I think is a pretty certain way for a migration like yeah.

I

The people are going to have some client-side applied object. If we flip the switch to the solar side, then we're gonna have to migrate. Everyone's objects, all the objects right at the time. They do results there to play the first time so when they update, so we need to make sure we don't mess up the config at that time. I guess that's the uploading Katrina.

I

C

Automotive could be litigation for that if we have a default to auto mode in the middle for like a long period, then then maybe we eliminate one of the failure modes that that users could potentially encounter, which is transitioning a single object from one strategy to the other.

J

Well, I, like the idea of switching server side apply on in the near term, but having forced conflicts Beyond in the auto mode she's talking about, but then in the future like we're warning everyone we're going to turn Force conflicts off eventually, so we have all that time. A year to transition, people onto server side apply and then at the end of it we can just as easy as turning Force conflicts off.

I

So what does this look like? It looks like go ahead. My chair.

B

So maybe we could literally use 127. to start putting the warning already that oh you're, using uh server side uh you're using apply. We are going to change it. You should already start explicitly setting the flag to true or false such that users will be prepared to use one or the other before we even start making the switch in the next release and um I'm positive that there will be a multitude of users who will not pick one or the other and we'll go with the migration.

B

But at least we can prepare them ahead of time and make a noise about it because, like that's literally what we can do not to break them.

I

I, don't know if we know what auto mode means right now, so I could turn out. I, don't know if we know what Auto means right means right now. um Does it mean we have? We keep client-side apply as it is, and we each other players on the side, police sources, and we do something about new resources.

I

We could first do Auto apply. The existing resources continue to create with client side, have a warning to say we stopped server side of playing your resources soon then start doing that with the first conflict, maybe in the order mode and then stop doing false conflict, one at one point or another, and then someone sort of like everything. How does that sound?

I

Trying to ease the transition I feel like we're, cutting the that's, maybe a French saying I, don't know how appropriate it is, but I feel like it's going to be a lot of pain points every every four months. People are gonna, have a little bit of pain as opposed to uh having a lot of pain at one point, but then forget about it.

B

Oh forget about it at whatever we do doesn't matter how long we will extend it. There will be people who will come after us and they will be complaining, but I I, I I'm, very supportive of that kind of episode, where we would gradually switch from one to the other over a few releases. So.

I

What do we say? I switched the cap to I, keep the auto mode. We keep it the way it is uh for existing resources, as in currently client-side applied. Resources are continuing to be clear inside so outside of play. The resources continue to be suicide. New resources are client side in the first place, maybe with the warning.

I

And we see what material face looks like you, don't like it.

B

um I'm I'm so for uh I'm, still leaning towards that I I might be wrong, but I'm still leaning towards the approach where, if you're, processing entire directory and nothing was touched on, the directory before was apply. I.

G

B

Automatically switch to servers that I apply because that's literally a new, uh uh a new package.

B

That has been already applied and it's already been applied for the clients that apply where you would end in a weird mistake or half of it. uh uh Server-Side applied and half of the clients that apply. That's, probably something that I would prefer to avoid.

I

Okay, so when you're saying directory, can we say it's uh the set of resources being applied, and so basically, if, if any of the resources in the set that you're applying has ever been client-side applied, then new resources, our client-side applied? Yes, what if you have a mix.

B

I would warn and and I don't know, that's probably the the point where I would start complaining that oh you're using both and we will break you soon.

C

Dig into that like why? Why are why are we afraid of that situation like it? The applies that is kind of arbitrary. In most cases, unless you have pruning enable you can apply random sets and and have uh like the cube control generally speaking, doesn't have, and the cluster certainly has no concept of sets across these objects, so why? Why do we actually care um about creating that scenario.

B

I'm being paranoid as usual, I didn't have any strong arguments and I can easily be persuaded otherwise um I'm just trying to come up with the worst possible scenarios. That could happen with that um there. If, if you put 399, can tell me that I'm wrong and I'm making this up I'm I'll be fine with the I.

I

Mean I agree with Katrina, it's very concerning, if you start doing say: oh I'm, just applying this one resource I know I just changed you're going to mess up the entire thing.

C

I

C

Worried we're going to make people care about the wrong thing like that.

C

If, if we don't make the right plan here, we're going to cause a lot of confusion like we need to be clear on, what's actually dangerous to the users and what we need them to pay attention to, because they have limited attention for to give us and like, for example, if we can make them not care about the client-side, server-side distinction and only care about like what to them is the feature in this migration, so that they're not like what even client-side server side. What like.

C

If we can frame the changes that they need to make in terms of things that are going to like individually make sense to them. I think that that will be better for for the ux um I.

I

Can totally see the bug we would have if this guy is like oh I'm, trying to apply this resource and I get this behavior and when I apply these two resources I have a different behavior and would be like yeah. We created this for ourselves, yeah.

C

So like, if it's a visual level, then as I said, because the sets aren't a real thing.

A

And it- and it seems like one of the biggest problems- is going to be okay, we moved somebody to server side and they may or may not need forced conflicts right, so that's gonna so like it could be that this is completely. You know the one case where people are going to trip up you know for sure, is we moved them to server side and oh no?

A

Now they have conflicts over some of the fields and something just started failing, which sounds like the cicd case, uh especially um if we focus in on that case that very very common pain case um that might help us move forward. What do you think? What does everybody think.

C

Yeah I I think the the force con Concepts that can't force conflicts issue is kind of the the Crux of the problem. I agree like because of the CI CD just disruption like tons of people use this in in that context, and um we're always telling people to also so we don't want to make their lives hard when they do what we tell them to, um and one idea that I had around this is that, like that, forced conflicts flag?

C

If, if we're, if we're hesitant to do a migration where we just turn on server-side, apply with conflicts forced by default, because we didn't want them to miss out on that functionality, then maybe that's where we should force the user's attention is in making that choice like make that flag required or some equivalent flag required so that they have to say no I actually didn't want conflict detection. Thank you very much like because that's that's a meaningful choice of them.

C

Do I want conflicts enabled or not, and if I'm in a CI CD system, maybe I I, don't want to detect conflicts if I'm outside it I. Do um they don't really care? If the operation happens, oversider client side, but they do care about that distinction and we want them to care about it. So maybe that's what we look at.

I

So what do we do now? Should we do Otto and new resources outside of black response conflict.

C

That's what's making sense to me, but I'm, like really nervous about like tomorrow's last day, to get an enhancement in, and this has to make sense as part of the plan. I don't know.

I

It's gonna be fine.

B

I would I would try. I would use 127 to figure out the details and write it down if something pops up. We will have enough time to address the issues accordingly and put warnings already in 127 in the code in place, uh maybe ideally put out a blog post, explaining and and preparing users for the um for the switch and their warning could actually point to that blog post already and then I, don't know something like that and then actually start with the changes in 128.

C

One one concern I have about that is that I think it's not even clear what the transition will look like for users right now, so we don't know what to warn them about.

I

Good trees through right, no longing yeah yeah.

B

But that's where that's why we have the entire 127 to figure the transition steps.

C

Transition to what like, we can't warn them. What they're gonna have to do without knowing what they're gonna have to do.

I

Something's gonna happen guys be careful, look.

B

Out we will feel worms and remember that whenever you're touching, this you'll have to do something, but we still don't know what you have to do. But do you feel you should be worn and the adventure will unfold.

C

One: okay, one one one suggestion, maybe that we could get something into 127., um this auto value for the flag. um Maybe we're like pretty close to thinking that we have some idea of what that should look like that could be hidden behind an alpha flag. We could um have like yeah. If, if you have some Alpha flag, enabled that's our common pattern right, then there's an auto value for that flag and people could actually try it and then, if, during this, like bigger process, discussion we're like.

C

Actually we don't want that to be part of the plan, we're free to remove it 128.. What do you think.

I

Yeah we'll have one John Doge right, that's gonna be good.

C

um I can do the blog post thing as well like much I was saying if we have that as an anchor, it can be like we're trying to figure out how this works. We made this auto mode thing put this flag on. Try it out, tell us that why it's terrible idea and here's our big plan, please comment on our camp. It's like kind of an anchor.

I

Yeah to have solo set a play enabled in the first place like if people are um crying Alpha things they've probably already tried to also to play. I would hope. We don't know if I'll show.

I

Yeah we can try it's uh it's starting for you to experiment with that.

C

Yeah, if I'm not set on it or anything, it just occurred to me as something that we could potentially get done. You know it's a tiny timeline that we have left to propose something yeah.

I

Okay, that's good I think we should continue this discussion online offline. Sorry.

A

So to the notes, I've added a little bit more context for those who are deep in the Weeds on this uh Daniel Smith has a Blog on the kubernetes uh kubernetes blog post uh about why server, side's, important and and as more context, just so everybody knows, we've, the kubernetes community has put a significant amount of investment into server-side apply and until we actually make it default, it's you know we're not going to realize that investment.

A

um So so that's why we think this is important and if you'd like to know more, please go ahead and read that blog post uh that Daniel Daniel's blog post thank.

I

A

Foreign okay looks like we've: we've filled up our entire hour, I didn't think we were going to at the beginning, but we we have. um Is there anybody else before we uh stop recording before we log off? Is there anything else that anybody would like to address in in one minute.

A

Okay, well I appreciate all of your time and I look forward to seeing you all in two weeks. Thank you.