Kubernetes SIG CLI, 9 Aug 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG CLI 20230809

Description

Kubernetes SIG CLI Bi-Weekly Meeting on August 9th, 2023.

Moderator: Sean Sullivan

Agenda and Notes: https://docs.google.com/document/d/1r0YElcXt6G5mOWxwZiXgGu_X6he3F--wKwg-9UBc29I/edit#bookmark=kix.cpcia4lljiyg

A

B

Hello, I'm, Sean, Sullivan I'm. Your moderator today uh welcome to the August 9th version of the six CLI bi-weekly meeting. uh Why don't we get directly into the announcements? uh The only announcement that we have is at least from from my end, is that we have within about a week slightly less than a week we're going to be releasing 128, and so um the code will be unfrozen shortly.

B

A

You have PR's that are out. You may want to ping some reviewers to take advantage of the.

B

Unfreeze, that's that's scheduled to happen next Tuesday August 15th. Does anybody else have any other announcements? Oh sorry, the steering election is uh I saw that that's starting up um and there have been I've seen already today. Several people have been nominated for the upcoming steering committee election. uh Don't forget to vote. uh You will be on a big list if you are eligible to vote.

A

C

Have four seats that are going to be opening up, uh Kristoff, Steven, Augustus, Carlos and Tim pepper? uh Are all their terms up, uh gonna be nominating mache, so fingers crossed he accepts and does well we'll see and uh yeah not to sway anyone's votes there.

D

Nice, sorry, what this! What a steering election, it's like a board or something yeah.

C

You could think of it as like. The governing board for kubernetes.

C

There are seven seats available for a two-year term.

B

And- and it sounds like four of those are going to be uh contested um or elected, this.

A

Coming up election is that correct, Eddie.

C

D

A

Excellent, so do we have um any other announcements from anyone before we continue.

B

So are there any uh sub-project or kep updates.

A

If not, we could move on to the first item which I put on the agenda.

B

Which is, and it requires some context, it is a proposal to add another command.

A

And we would like to especially get the tech leads feedback on this.

B

um Let me first but start by saying we I believe with server side apply. We are kind of at a dead end in the fact that we cannot safely move server side apply to being the default with because of Legacy concerns, and it turned we. We've invested a significant amount of effort into server-side, apply and yet I think that the statistics I've seen is that there's, a very, very small percentage of Coupe cuddle use, which is server-side apply.

B

It's almost even though we've had server side apply around for years and years, client-side apply still dominates and in order to get the uh the most out of our investment into server side, apply or brainstorming ideas. In order to move the ball forward on how we can uh make server side apply easier within Coupe control, so just as a quick update for those who are need a little bit more context on this server-side apply was created in order to allow multiple agents to apply the same object relatively safely and so client-side apply.

B

We um we calculate the patch for applying on the client, whereas and server-side apply. We calculate the patch, we send the entire object and let the server calculate the patch and we recognize this as being or we believe this is a far superior method of actuation and we've put significant amount of effort, as I've mentioned into the server. In order to be able to do this, we've we've integrated it into Coupe control, apply as a flag, and it is not being used not very much so.

B

The proposal is what, if we create a new command, coupon will actuate that is Coupe cuddle plot server-side apply as default with uh integrated migration from client-side apply the metadata being stored. The last applied annotation would be moved into the managed fields, which is the way that server-side apply.

B

Does it as well as we have brand new pruning that uh that Justin and Katrina have uh have spearheaded, and so we could, if we agree, create a new command which is a best-in-class, apply or actuation, incorporating all that we've learned over the years and being and allowing it to be used basically as opt-in, um with without being Shackled by the Legacy concerns.

B

So that's my my initial pitch I'd love to hear. um Actually, what I? First, uh if, uh if Antoine doesn't mind, maybe we can have you uh describe since Antoine had a lot to to do in the work for service side apply, describing um kind of. Why were roadblocked.

B

um With with the current uh apply, yeah.

E

So two things I, don't think we're a roadblocks specifically and second of all. Well at least maybe in Cube curl, but I, don't think Cube curl is necessarily the most relevant out of suicidal player. We have controllers and we have a player El zargo CDG Etc that are already using solicited player. Okay, so.

F

E

Not an Excel set of play. It's a wasted effort. It's yeah, I.

B

Think it's called.

E

Cube girl, it's not that critical I think we initially did it Focus like Whiskey Girl in mind, but eventually we realized that it's not necessarily where it's the easiest pass to entry, especially because so some things are different between client-side appliance outside of play. Does the conflicts, some things can't be applied that used to work for client-side apply, um like least associatively style a lot more and false than they always they're inside of life, and so there are cases where the AP, the the behavior, is going to be different.

D

E

I'm personally, at this point very uncomfortable, changing the behavior of the parallelogram for anyone. So.

B

So, that's why you um so that, so you do agree that, in order to move forward, we don't want to change. We can't change the default.

E

B

E

Coupe cuddle apply: does the millions of people using Cube curl apply in the world that do not talk to the community at all that don't know. What's going on, what's happening with the work ordering and what city saying and and because of that, I think if we change it, it's just going to break everyone for no reason and I. Don't think that's the good experience we wish for people using Google so.

B

If there were a new command, that would obviously be an opt-in. It would be somebody who actually is saying: I want to try server I want to use server side apply um with this new command. Yeah.

E

So yeah I feel like if there's a new command and we change the dog throughout to use that new command people are slowly going through migratory, because they'll be like oh I apply is not the new thing.

E

The new thing is whatever name we come up with, and then but at least people will make a conscious decision about that and they can have like a migration plan that suits their company's policies and whatnot, um and if they have a bug, they know what they're looking for they're, not just like Mario block or Mega Blade script doesn't work anymore, um so I think that's a better way for a while yeah it's gonna be a less less invasive for people for sure.

B

Okay, I I. If uh so, I think that I've talked uh enough. I'm curious, there's already one hand. Leonid has his his hand up uh Eddie does as well Lena. Would you like to to go first.

F

Yeah thanks uh I have a look question uh to me. It is still like unclear uh your concerns about switching the people Behavior like uh to the extent of my understanding of most of the people. It will be transparent.

F

So can you bring examples? What of the demonstrate exactly as concerning points the problematic points that if you can happen, if you're switching.

B

That's that's an excellent question and I'll I'll add to when, when I've uh done a little bit more research, I'm going to add to the notes on specific examples, but but as as Antoine was saying, there are the the behavior is different enough that if we switched it without, even if we told the community that we were going to switch the behavior, the the behavior would be different enough. That it would end up breaking existing applies and would be extremely disruptive.

B

um And and basically break people I'll, let I think Antoine actually.

E

No I think the the main thing is the conflicts right. We don't have conflicts, we Supply. If you change a value, it changes the value we saw us out of life. You have to acknowledge that you want to change the value if it's been changed by someone else and so scripts that people may have written in, like random places in the world are going to break and I. Don't think that's great.

F

Basically, if I understand you correctly, you say that uh yo, our concern is that, instead of override behavior, that happens mostly today when people use apply as it is to their client.

D

F

They will get a merge Behavior and it will break as is existing like a functionality or or Workforce visits right.

B

And if, if there's a conflict, now you actually have to put an extra flag to to override a a field.

B

um So so, if you have your client-side script, which is applying- and now you have instead of just overriding which is what's happening now for client side apply, it actually would not apply that and it would tell you that you have to add an extra flag of force forcing the conflicts minus minus Force conflicts.

B

So that sounds like that's the biggest behavioral change, which is, it would end up breaking a lot of people.

F

Okay, thank you and.

E

There are other changes like um I was mentioning the associate degrees, the environment variables. You know, for example, can be duplicated with client-side apply, but they can't be duplicated with south side of life.

E

So if someone has somewhere and I'm sure someone does has a deployment which duplicated environment Bibles they can play inside apply that, but if we switch the behavior, that's not going to work for them anymore, and so we're basically like breaking it for them yeah, at least if we change the command and one like this is now using server and the people are like okay, I, actively Target on command, I see it fail, they know exactly what happened if we change the command and they don't even know underneath underneath.

E

Why is this version of cube, Coral broken.

B

Right so so um Eddie, would you like to to go be really interested in hearing uh your feedback on this yeah.

C

I mean this is something we've been on for a while right. The honestly I would love for you to pitch Jordan on changing the default and just film his reaction.

C

He would absolutely never let allow this to happen right, like it. I agree like this would break a huge chunk of the internet and modern Cloud native world. As we know it.

E

um Yeah I mean we could also have fun and look at the reaction of the internet.

B

And yeah roll this out and see and and see what the reaction is yeah.

C

um The I have a couple thoughts, so the one first thing I think this would be. It would be really helpful for you all to have um some reproduction examples to help people understand exactly what we would break so just like set the stage with like a simple thing and just be like if we switch to default like you're, expecting this behavior and then this will happen uh so I think that'll be super useful in like making this decision and educating everyone.

C

um The other thing I was going to say was.

C

If we just if command I, it doesn't really matter whether we do a new command or we um change all of the behavior under the a new flag or something to apply I I think it probably makes sense to like be like hey. The new thing is actually just everyone moves to that. You don't have to remember flags and like this is the now best practice for doing an apply like I. Think that makes sense too, but I'd probably be in favor of a new command.

C

um Yeah I I totally see the the block there like there's no way we'd be able to just change the default, like so yeah I. Think I I appreciate all your work on this I know you all have been Trucking down the server side, applying to change that route for such a long time and yeah I'd love to get it into the hands of people. So we can stop giving our regular Public Service Announcement during our our Kubla Khan uh updates.

B

uh Yeah I think I mean I, think that this is. This is the best way to move the ball forward. So I think that you know we're going to have 99 of the people using Coupe control. Doing client-side apply five years from now. Unless we uh we try to to move the ball forward um and and I think that this may be a a useful method. That kind of uh side steps the the Legacy issues by making it a new command that people are opting into.

C

Yeah and it's not like it's not like you're jumping to oh, let's create a new command as the first step here. Right like there has been a long road to get here so.

E

I mean I suggested creating a new comment five years ago.

C

I'd, rather a five years like getting the exact thing we roll it out all together.

B

So so Justin mentions that we could create a whole new binary instead of Coupe control.

G

I think that should be our worst option. That is our but I like the idea that it's being proposed. But, yes, that was mostly making a joke about the correct pronunciation. Correct pronunciation of the the binary.

B

So so, as I mentioned, I think that we can. We can actually uh put several uh several bits of functionality that we know are best practice. You know we we've just recently as as I mentioned, uh with with Justin and Katrina moved into uh the the better pruning, and so we could make that as the default as well um and say hey. This is our after after years of working with Coop control, you know, especially if it's Greenfield you're just starting now.

B

This is the apply that you should be using, um and this is the reason why.

E

We could also look at things holistically, like I. Think applies definitely spreading across many comments and there's like the safe config in create yeah. We could first of all we could have a warning to say: that's a price deprecated eventually applies deprecated use, actuate right, um so that people realize they have this option, and then we can clean up the flags in places like create the actuator.

E

Like the new comment, whatever name it has, um can have different flags all together, like there are a lot of flags in apply that only apply to client-side and some to sell side. So it's kind of a mess. So that's a good opportunity to look back and I clean up a lot of things in Cube, curl, I.

E

Think throughout and I was mentioning these I don't know if we can right now, but the the cube curl Thief right now is using the same flags as cube Coral apply, so we would have to decide what we do. Whiskey, curl diff, or we could make a new comment for this or whatever, but uh as.

B

A sub command like.

E

Yeah, as you mentioned, we have to we have to think about this holistic area a bit and see how the flights make sense between each commands.

B

So obviously, there's a lot of work to do, but we wanted to at least kick off this idea and get feedback before we, you know, started doing too much work here, but it sounds like from what I'm hearing the uh most of the feedback is pretty favorable.

B

Is there anybody else who has any different feedback or even the same feedback um or different concerns? uh We'd love to hear that now, as well.

B

Is that Justin? Do you have your hand up yeah.

G

I I don't have a concern per se, but I did we talk about um in tree versus plug-in like what is the right execution strategy? It sounds like we aren't entirely clear on exactly what is in and what is out of this right, so I felt like that was sort of maybe something we should talk about.

G

B

I uh on purpose, I I uh I described the proposal as a new command. In the notes, I mentioned that it could be a plug-in first. I think that a new command would be significantly more uh would would give us significantly more adoption. I think that a plug-in we would, if I mean we could actually it's possible to create a plug-in and and release it with the coup cuddle binary so that we're releasing two binaries. That sounds like a lot of uh coordination effort with the release team.

B

um I can understand why you know we would want to uh ensure that this uh worked in a plug-in. I could see that you know the the advantages of that. I think that the disadvantage is that it it may not be used like it may not actually be if it were especially a plug-in that that somebody would have to individually download.

F

B

We're not going to move the ball forward, because anybody who knows how to download and install a plugin also probably knows how to use server-side apply with the with the flags um and we're trying to make this as easy as possible. It's.

E

Very well in Sheriff Corolla, prior duplication. That is, please use this plugin instead, I think yeah.

B

So the deprecation notice again would also be uh difficult um and and disconnected so so what? What are your feelings? Justin? uh Do you on command versus plug-in.

G

I totally share your uh your evaluation. I think the one option that might be possible would be to do a plug-in first with an intent to move it into a command and whether that that is just like more not worth the pain but like I. Think if we like it actually wasn't too bad working in in tree as it were so I feel like we can iterate in tree as long as we like put suitable guards on it. Like you know, Alpha whatever, like.

E

Yeah I was that was going to be my next question. What is the timeline followed.

B

um So I I hadn't thought of any timeline. I thought I would just like drop this bomb on the meeting and see what happened um um and got some. You know got you and Justin involved slightly ahead of time and and uh so I I, don't know what the timeline is.

E

I'd be happy if we had say an alpha version index release and we can play with that, and even the command doesn't attribute I. Think for our comments. They need to be hidden somehow right, so I think we could totally have that as a alpha command. We decide what's going to be in there. We try, we start playing with the flags and see what flags make sense and which don't and then because it's Alpha we have the ability to change them anywhere. So I would just go ahead.

E

It's like it's, not a big change to create a new command and put outside of playing. Now it's like yeah, it's a day's work right, so yeah and you can totally make an alpha version for the next release. If you want it.

B

Yeah I don't think that that would be uh too difficult at all. Yeah.

E

And then from there we can be like okay, let's see if we can add more things, and how can we be consistent with the flags and make something that is going to work for the long term.

C

Yeah, putting on my Katrina hat for a second like obviously, we'd need a cap for this, um which.

D

I should be fairly straightforward.

E

I, don't know who's going to be doing it um here in this room.

E

Looking at you Sean, but uh yeah I, don't know if we have a lot of capacity right um so I'm happy. If we have some help to be honest, so I came to the people not in this room.

B

Okay, so so I'm going to put in the notes that the uh the feedback is generally favorable and uh there's still a lot of questions, but that we could move forward with the effort. um Is there anybody who does not think that we should move forward with this effort before I? Go on to the next item, foreign.

B

So I will go on to the next issue, which is um a notice of a particular problem with the new aggregated discovery. There's a particular set of versions of the uh client go as well as uh the uh the version of the API server, as well as whether or not there's a metric server in the cluster. That can cause very bad things. We don't have to worry about this for Coupe control. Coupe control um has had this fixed for months in the latest versions of 126 Coupe control, but there are some clients.

B

In fact, uh if you, if you look at grep.app, you'll, see that there are certain clients like istio um capped, kubert I actually contacted the kubert team and they've actually got a PR out to fix this right now. I just wanted to as widely as possible distribute this particular notice so that any other kubernetes client that is importing a client go should move to uh the the version that's mentioned in the the issue moved to 126 uh sorry, 0.26.4.

B

um Or higher, in fact, it's completely fixed in uh 0.27.

B

um So, if you're in charge of or know of a particular kubernetes client that is importing client go, um please uh update the imported version of 26.

B

Does anybody have any questions about this issue?

B

I think that istio already has a a PR out within the last day as well we're trying to reach out to as many of these clients teams as possible to get this fixed? Yes, Justin.

G

uh If we had a security, vulnerability and client go, uh presumably people would get like Dependable type of things right, I, don't know exactly how that magic Works. uh Could we do something similar here where we'd be like where we say like these versions are not security issues but, like hey, you really want to update to 264, because in practice, once you get to Coupe, 127 you're gonna have a bad time.

B

Okay, let me put this in the notes. Let me um make this is an AI for me.

E

B

Into dependable.

G

Or anyone that knows how to depend about magic works, I mean if it's a security.

E

Issue shouldn't it be a cve or whatnot.

B

It's it's not a security issue. It's a no pointer exception.

E

B

It's client-side, okay, so the.

E

Even in controllers, in what now I mean it's only like if.

F

E

Kind of like denial of service or anything.

B

That's a good point. Let me look into it so.

B

Yeah I'm gonna, so so actually it wouldn't uh be that because the latest version of it has to be a 26 version of the client versus a 27 version of the server and only some parts of 26, and so there's the controller and the API server are going to be the same version. It's. That's actually not going to be an issue.

F

B

Controllers I believe.

G

All right not for built-in controllers, but for any, like external controllers, it will.

B

Be right right so actually some of the uh the teams that I'm reaching out to are in charge of particular operators.

E

So yeah, if we can frame it as a security issue, then it's going to be a lot easier to get people to update. Don't we ever fix that shot and he's pushing anyway, which is going to be.

B

So the the fix is a server-side fix that will not pass through the nil GVK and so basically.

E

Okay in trouble so.

B

Clients won't be in trouble once that particular fix that Jordan has out since yesterday, um but but that would require people updating their server, which is actually easier than updating a trillions, yeah clients, yeah foreign.

G

And to be clear, I wasn't suggesting that we we've magically discover a cve. I was saying there must be some mechanism, and can we like somehow like? Is there any way to use that same mechanism for a highly recommended, but non-cpe type updates.

B

Good point: I'll I'll, take that as an AI to dig into that thanks.

B

Okay, so can we we'll move on to the the next item? uh Jazwear, so was. Let me share this. This was the issue that you were.

D

Yeah I uh I wanted to screen share something. Let me know if it uh oh I am I allowed to do that or I.

B

Think I have to make you a co-host.

A

um Let me just make you a co-host and I should have made some other people co-hosts as well.

B

So so go ahead, go ahead and try now.

D

A

D

Can you see my screen.

B

It's it's yeah, I can now. Okay.

D

Well, wait, look: where is it.

D

D

Yeah so I've been working on this a bit and it's about Auto completion that it takes long when you use step and right now, I have uh is this plugin and I? Have this auto complete thing and yeah? What I wonder about is what it's like supposed to do like I? Think if I do this and I press tab, it's supposed to create another piece of text here and I was just wondering what.

D

um Yeah, so this thing what it has in its code is pretty much this. So what does? What is that supposed to do? Is it like supposed to add Port forwards to the end of this or I mean yeah, it's supposed to add Port forwards to the end of this or yeah. That's that's. uh I was wondering.

B

So that so so there's a an entire uh subject of of completion.

B

um We initially had completion as bash scripts that were being um uh sourced and I believe now they're in go as part of the um as part of the binary clinical binary I believe. So there was one of our members who was kind of the expert in who I shouldn't say was, but is the expert uh mark um Can someone remind me Mark's.

A

Last name, who is the completion expert.

F

B

Yeah, that's it um there.

D

Is I think I mean I think he he said something and uh comment base of tests. If you are I'm working on so.

A

B

So, if you're interested in working on uh the auto completion functionality, of course, he would be the one to to work with who's going to be the expert go ahead. Eddie yeah.

C

Sorry I just I just see you tagged me on this. My notifications are a storm. um I, don't have context right now to give you an answer. Are you there, but I will keep this tab open and take a look.

D

Okay, thanks: how do I put this off? Okay.

C

Thank you for jumping into this issue.

A

Yes, hi you're welcome, okay,.

C

ah Okay, we have one thing left before we can do other stuff or wrap up. uh We have the kubecon video update due tomorrow.

C

uh I was hoping, uh Marley or Natasha would be here to volunteer to record with me, so I will pester them directly uh and we have some Source stuff that we looked at last time. We started Gathering a list of updates to include anyone have any specific things they want to make sure are called out that we haven't talked about before context. This is we get a as a Sig.

C

We get a one minute, video update that gets played during the Keynotes of kubecon, so a great way to highlight our Sig, the work we're doing and what we need help with.

C

So if no one has thoughts, uh that's cool otherwise feel free to DM me and we're probably going to record this this afternoon. So thank you.

B

Thanks for uh taking the the lead on that Eddie.

B

So we'll wrap up the the meeting um unless uh there is another any other issues that the the uh 60 Li would like to discuss last call for uh for speaking out.

B

If not, we will see you again in two weeks. I will uh take it upon myself uh for Action items to to see uh to uh flesh out the proposal on the actuate command.

A

Okay, thank you very much.

E

Thank you, Sean yeah,.