Kubernetes SIG Service Catalog, 29 Nov 2016

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Service Catalog 20161129

Description

SIG service-catalog continues discussion of next steps after code drop, as well as design discussion around unbind.

Agenda - https://docs.google.com/document/d/10VsJjstYfnqeQKCgXGgI43kQWnWFSx8JTH7wFh8CmPA/edit

A

All right, hello, everybody welcome to the October 29 2016 special edition of the brunette e-cig Service Catalog, where we left off our heroes were discussing the discussing directions forward and next steps. After the the code drop I. Think any of you that have been paying attention in the in the github repo have seen. There's been a lot of movement today. We've already got some folks that are acting on those next steps, and so let me let me kick this conversation off by by drugging everybody's memory.

A

The last thing that we discussed in yesterday's meeting were some some code moves of the brokers to contributors and examples. Two examples there have been, as I said, some other subsequent pull requests today that have been along the lines of making the project structure more similar to coop Nettie's, which I think is a good. You know um now I wonder from the discussion yesterday.

A

Does anybody have next steps that they would like to discuss today um and I? If so, let's, let's all discuss them now,.

B

What see I on the list side was one of the issues that you Paul opened and I. Think if we go too long without things won't break. Yes,.

A

I agree: I. I will put that on the agenda up for today um and I I think that is a good thing, but not especially in trinsic to yesterday's conversation. So any any other hanging hanging chads from yesterday that we need to to dress.

C

Not entirely sure if the supply, so you tell me, we had discussion and slack about the lgtm labels and the veto labels. That is an issue number 50.

A

Okay, so that I think that's fine to discuss Erin. Would you like to summarize that for the group in maybe we can just perhaps see what how everybody feels while we're all convened here together.

C

Sure we I created issue 50, we had discussion in slack, there were a bunch of alternative possibilities, brought up I, think poll approve, I or calm, or something was brought up. I had never heard of it. So I did some research on that. We had Garrett for github brought up, uh and then we also still have the current situation where we put LG tiemz and I'm assuming vetoes into comments which is still on the table. I believe as a an option I put at the bottom of that issue.

C

A second comment to the bottom on that issue: a proposal for how to use labels to do this and then I just saw via they put a response to that. So I'll put a link to the issue into the chat and your dream. That's the summary all.

A

A

Okay, viele, I see that you already have registered some comments on this. Your comments were: it's insufficient to apply italy feet a label with that justification.

A

And then there was a question about lgtm, but there's a few tiny things that need to be fixed. So let me just say I when it comes to lgtm labels. If there are things that need to be fixed, it is non lgtm.

A

It can be lgtm, modular knits in a comment and if we choose to go with this labels thing don't apply the label, but if something needs to get fixed, it is not lgtm. In my opinion, unless there is a explicit agreement that it can be fixed in a follow-up which is something that happens with some regularity, any thoughts about that.

D

A

E

D

Aren't there possibly circumstances where something might look good to everybody and you know, might have received the required amount of approval from from three people, but for whatever reason we don't want to merge it. Yet.

A

If we don't want to merge it, it shouldn't be lgtm in the first place. In my opinion,.

D

Okay, that's fair well,.

B

The co-owner case that comes to mind is the one: where are three LG teams, but within less than 24 hours, and that I still one still has to wait for the timer to expired. Wait for vetoes technically, that's.

D

A

Yeah and I actually don't think that we were enforcing the timeout rigidly today, um I I'll be honest with you. Folks, I think that the moves and stuff that we're doing out are probably not ones where veto is likely. I think that it is sufficient to me to have a gentle person's agreement amongst us all that will act in good faith and we'll use our best judgment about when something may be controversial and I think you know so to make expedient progress on things.

A

Everybody agrees on that are below impact um that it's probably not especially necessary, but I think that the you know being a good citizen is part of like if there's a dimension of doubt I, and you think there is some ambiguity- that it is polite to let the time lapse. I.

B

Totally agree to a favorite expediency, um in absence of doubt,.

A

Okay, so I I think that it would be great if we could a like right this protocol in a document that lives in the repo and have that to point newcomers to because you know, assuming that we are successful in our mission, we will eventually have many many more people coming and being interested in contributing to this project and I.

A

One experience that I've learned with Cooper Nettie's is that it can happen without warning that suddenly everybody will want to jump into the pool. So if, if we could get a volunteer to produce a document I with this information that goes into the repo and a pull request, I think that would be great. It sounds like we're we're very close to agreement, and probably we can work through any remaining nits in a pull request. Anybody disagree.

A

I'm not hearing disagreement, okay, who's, gonna who's going to raise their hand first I.

C

Will do it all right.

A

Old, faithful Aaron, it's on you.

C

I'm, your there's still the question of difference between lgtm three and merge I.

E

Think that was answer but Cohen. Okay, so, let's just say the three people getting the say, the first hour and they're like yeah. This looks the bomb diggity is, but then you want to go and wait for 23 hours for somebody to redo. It I think.

A

Okay, my own my own, take on this is that lgtm three as a label is meant as an attestation that a specific person said that it was lgtm uh I, think it can have lgtm three and uh the under the best judgment rationale. If it's not controversial, if no one expects it to be controversial, it's probably okay to merge. If there's any dimension of doubt what the timeout elapsed be a good citizen and- and you know given give it a day with that said- I the.

A

If we continue meeting more frequently than once a week, probably much of the work that is done will already have been discussed over some type of video chat or in person me being optimistic about everybody's travel budget there. So for things like that, it's probably it also probably lowers the need to have the the time out. But again, this is. This is an area where good judgment and being a good citizen are very important, and we can.

A

We can write down that people use their best judgment, but what matters the most is that people actually do. I.

C

A

C

We all cool with, let's say, I, contribute a core maintainer dozen lgtm three. Are we cool with? If it's been a day that person can label and merge immediately uh yeah.

A

I think so I think the clods the clock starts ticking when the pr starts kicking.

F

So rather focus on the LTM 3. I'd still like to understand the need for the lgtm labels at all. I don't understand what value they bring a guy. I can understand a label that says this thing's ready to get merged and we're just waiting for the time out, but I don't understand the one two and three I gotta be honest: okay,.

A

When you say you don't understand, can I ask, do you think it's unnecessary? Are you willing to live with it for now and if it's really cumbersome weekend, we.

F

Can I think it's unnecessary I can live with just about anything. I reckon I can grumble to myself all day long. That's fine! I do anyway, but.

G

F

I do think it's unnecessary, I mean I, don't see the difference between 12 and ready to merge, wedding, I'm. Sorry I, don't understand the difference between just to put in a number there, because it doesn't tell me anything: I can't query on it for any purpose. Other than that I can know someone has lgtm dip. It doesn't tell me whether I have or not so it doesn't reduce the list of things that I need to look at. If I do query, I just don't see the point. Ok,.

A

uh Then Aaron, perhaps you or someone else from Deus, can explain the rationale. Do you do you feel that there is a value? Add for labels, / comments, I'm completely agnostic on this mechanism for the record I lat.

C

Kent's, can you want to explain this one yeah.

D

So I do personally feel that there's a value to it, I know I'm. So at dais we've been working this way for a long time, I'm, certainly not advocating that we do this just because it's familiar with I do actually believe there's value to it. A lot of us have created queries that we've bookmarks that allow us to look at individual projects or across many projects, and look very quickly and easily at things that make our attention.

D

Doug has made some very good points about. You know it if we don't all have our own personalized labels, you don't really have a way of saying. Okay, find me of the things that I haven't already approved, and that's absolutely true, but you know this is, I would say I would qualify it as good enough. You can write a query where you can. You can look at openwork items have not received the required number of lgtm or have not been vetoed or so forth or so on.

D

You can query along those dimensions and you can see in the index you a list of things that might deserve your attention. So you know as a for instance, if you query- and you know, 50 different PRS come up, and one of them has three lgtm and many many others do not. You might not want to focus your attention on the one that has three LG TMS already. You might want to sit and spend your your efforts elsewhere.

D

So I I do think that it's kind of useful as a progress bar of sorts to show how far something is in the process and where you in particular might want to focus your efforts. Ok,.

A

I I think that makes sense to me Doug d. Are you convinced by that.

F

um No because to be honest, um if there's one out there that has three lgtm zacks, that's actually one that I would focus on it, try to look and see why we couldn't push it over the line. um It seems to me that you get the exact same effect by doing a query that says, find me a PR. That's you know over a day old and has not been vetoed. It has a comment with Dell DTM in it somewhere. Absolutely if you don't want the exact same result, so Doug.

D

I think you may accidentally have just made an argument in favor of the labels, because you said that if something had three LG TMS, you might be giving an additional focus to it. So without those labels you wouldn't be able to easily identify those items, isn't that correct I totally.

F

Agree, yes, but my point in saying that wasn't because I actually care about that as much as you would say: you'd want to avoid them. I don't think you should avoid any PRS because of the neurology teams. Well,.

D

Wet weather, okay, so whether whether the labels are there for the sake of avoiding particular PRS or for targeting particular PRS is really kind of inconsequential. The important thing is that the metadata exists so that you can query along whatever dimensions matter to you. It's not it's not for me to say what what what matters to you or for you to say what matters to me, but as long as metadata is there we we can both work as we please yeah.

F

I know what you're saying I: if that's what you'll want to do? That's fine, like I said I can live with just that anything just I. Just it's just extra work for people to do it because I typically forget to do labels I'm constantly being reminded by the doctor guys to update the labels appropriately and I. Can just tell you it's just another thing that I forget so liking.

H

Them with it so I'm, sorry, if I, if I, missed something in an earlier conversation or in the slack channel or something. But it seems to me that we're arguing about or discussing like what mechanism we should do to get code fruit, so there can be merged. Yes,.

E

H

Is but we're talking about mechanics and I, haven't really heard or seen a concise description of what behaviors or what? What are we trying to accomplish with said mechanics like what is the? What is the thing we're trying to ensure does or does not happen, and if you can describe that in English, then maybe we can stay hey. We could solve this in 15 different ways.

H

Maybe one of them is with labels and three variants of lgtm and V tonal stuff, but it seems like we're talking about mechanics without having agreed on what the purpose is. Yeah.

A

So I, my own impression of the purpose, is to na it's very rough. It's not detailed now, but in my mind the purpose is to ensure that we have sufficient group by in without describing what the decision plane for is something lgtm.

A

But and if we we find that to be something that is useful to do, then we can explore it.

A

um But basically, I think the the point of this mechanism is to ensure that everybody has time to review in that uh the the reviewers or that that each change is reviewed in attested to its fitness, to go in by at least three people, but now I I will interject also into my own sentence uh that I I think that there are um that there are other things that have a higher utility to address as a group, and so what I propose is that we take about 30 seconds if anybody's strongly disagrees that this is movement in the right direction.

A

On the subject of governance, I I would really like to to make some progress on governance, and then there are other things that I think are very important to discuss. That I'd like to get to today, and we leave now we're 20 minutes into the set of an hour.

A

um Does anybody have any reservations that are strong enough, that they think we should fundamentally reassess what has been discussed so far I and continue to spend time on it here with the knowledge that, if a need arises um that we can easily bring this back up for discussion further elaboration for their formal outlining of what we think lgtm is.

H

Well, you know uh I'm wondering if it's been explicitly stated that the three LG TMS can't be from the same organization right like if three of the Google guys all lgtm, a google PR that doesn't seem like the right thing right or it you know substitute any other organization. I mean. Is it should they be?

H

Some sort of the lgtm should be coming from different entities or you know you get one vote per company or like I, don't know, I mean obviously nobody's gonna, try to use anything but I'm I'm just trying to get at what is the? What is the end goal? We want to make sure that you know we're. All in agreement is community right, uh so.

A

That, yes, that is, that is correct. We haven't discussed the need for different companies, I'm willing to give this another five minutes. If we would like to discuss that now, um but I will know so far, I don't think, there's been any situation where in lgtm has come from more than one person at a particular company on a particular issue. I think we've kind of we've been informally doing that so far, I would support making it formal. If that's what the group has appetite for.

B

Another another point is: if we are intuitively doing the right thing, do we need a set in stone process codified? That actually brings up a lot of controversy because we're getting into the legal speak and hands the which is written down. The intent gets gets hidden in the corner cases that arise so.

C

Could I bring up a possible the lack of better term easy out in its discussion, claude, and that would just be, let's add a veto label and then put off the lgtm discussion. Stick with our informal procedure now and see how far that will take us I think.

A

That is a sensible thing to do. I would like to see a show of agreement or dissent in the chat.

A

F

Think the word is Vito.

A

Okay, well, this is.

E

This is a vote. You need a label, that's.

F

Right I can't see without a label.

A

Okay, uh I'm, sorry, but you appear to be outnumbered, so Aaron I think if you will, if you will proceed as we've discussed here, if you would like to just make a note in the issue as to what you're going to do, that would be good and I would I'm totally fine continuing to have the governance like discussions, but we do need to be conscientious about how we allocate time.

A

So at this point, I'd like to move on to a design discussion unless there are, there are any additional things that we need to discuss around governance or next steps from the code draw that have not already been discussed.

A

So everybody, okay with that.

A

Alright I hear no dissent new descent, all right. So um let me let me share a link in the chat to an issue that I've created.

A

Now this issue is on unbind. I am not sure how many of you have read it. It is not necessary to have read it to participate in this discussion, because I want to kind of walk through it. So this is the video chat version of issue 39.

A

So we in the face to face meetings that we have, we had really good productive consensus around the generals. Shape of things usage of the API up through the following someone adds a broker. The the controller has a watch on the broker resource goes and can like it's. The B to catalog endpoint gets the service classes from the broker / system into the API server. A service consumer says: hey I'm, on an instance of the service class. They create an instance.

A

Controller goes and calls the provision endpoint and controller updates the the status of this resource later service consumer says I want to do a binding against this instance and we've gotten up through the constructive portion of this right, like we've gotten to the point from empty catalog to now, I've got an application with a binding and I. Think a logical next step to discuss is what happens when you want to do an unbind, um so the the precedent that we have for this in Coober Nettie's is in storage for persistent volumes and persistent volume claims.

A

When you want to release a claim on a persistent volume, you delete the claim.

A

This doesn't analogize perfectly to the Service Catalog world, because once you declare your intent to unbind, there's work that has to be done. So it's not as simple as you delete a resource and then it's gone and the controller can figure that out right, because the controller has to ensure that if it misses the delete event that it can still have a way to understand that it has to go and do something without keeping a local cache on that persists across controller restarts.

A

So the pattern that we have in Cooper Nettie's to do to accomplish these types of things is called finalization. And basically, let me draw a picture for you all in cooking, show style: I have prepared a whiteboard. Can everybody see this.

G

A

C

That's better: okay,.

A

All right can, can you see the stick figure guy over here just.

F

Out of range, no okay,.

A

Let me let me back this thing up see if I can get more in the shot.

A

Okay, how about now we'll.

F

Move your shoulder yep there, yes got it all.

A

Right, so um this is my just it's not all the top of my head impression of how we could use this pattern in a Service Catalog, but it's it's a possibility. I've spent a little time mulling over um and so this this diagram should look familiar to everybody on this call.

A

At this point, it's a simplified version, because I think we've probably all got this in our heads- I'm happy that to add any more detail if its desired, but basically I, just I I've, written down the material posts and yet some cuts um that are essential to understanding the API flow that I've described so far. Everybody cool with what's on this board,.

A

I'm, not here, no okay, so here's my here's, my I would call it a pitch. Let's just call it a thing: here's my thing so going to use green for new stuff and we've already done a post on this binding, so the binding exists. The protocol that we've all discussed so far has been carried out where I posted I'm.

H

The service consumer in this.

A

Case the service consumers posted the binding, it's gone into, the API server, persisted and EDD controller has watch on API server and it's gonna put on the e2 service instance by names. Endpoint on the program. Alright, and so life is good. Life is good and I say: hey now, I want to unbind from this thing, so analogizing imperfectly to the current precedent that we have for this, um save it to declare my intent to do to unbind I. Do a delete.

A

On this binding- and at that point, what happens is that the the rest api says? Okay, the service consumer wants to delete this resource and um instead of going out and carrying out the delete, we we know that there is some work that has to be done.

A

So we need to keep this this resource around and put it into a record somehow on its status, that it's being deleted or being unbound or being finalized name TBD, but that we're entering the endgame for this resource so say, for example, that the the rest api records a condition in this binding status. That says this. This has a deleting condition. That's true, and so what happens after this is that II, the API server gives back at 203 accepted 202.

F

A

Sorry, you will find that I have very imperfect knowledge in HD p, so I I I accept the suggestion to and 2accepted odd e.

A

I'm trying to figure out how to draw this, so the one thing that happens in here is that the state or I'm- sorry the status as a let's call it. The leading condition.

A

A

The controller has a watch on this binding resource and says: haha I've got an update event for this binding and I see. The leading condition is true and here's where I'm going to call on Brendan, because I think he is most familiar with the ins and outs of the API. At that point, controller will invoke on the broker the unwind operation. Brendan. Do you happen to know the request and Method combination that is to be used with service broker API? To do that? I think it's just to delete.

G

On the binding yeah.

F

I agree: ok, so.

A

It does add elite.

A

V2 service instance blah binding.

A

A

Depending on what it gets back, the controller can continue doing this until it gets a 200 back um now say they say that this is happy land, and this happens. uh You get your 200 back the first time. Life is good, so what controllers going to do is and here's where we get into kind of low level stuff.

A

So like, let me let me push your frame onto the stack, um did ization pattern and Cooper Nettie's is applied in namespaces, and the way it's implemented is that there is a part of the spec which is the list of finalized errs that have to run, and so.

A

In this model, we'd have a part of the binding spec that was like list of finalized errs and right now, I can just be a list of lists of one which is the controller finalizer, so the controller will do an update then, and here's, where is where this like diagram is starting to break down, will have to decide on a new diagram convention. So so.

E

Paul, yes, we will um so this is sort of kind of what I was worried about or I wanted us to spend a little bit of time talking about in Seattle, when I said that we need to have state machines, because we are, we talked about defining an API, but we are not really defining an API. You are defining bunch of resources and then there's some behaviors that go with them, and we are going very deep into delete here.

E

But, for example, simple questions like when, can I call delete binding like like overnight, when can I called delete broker? It seems like there's like all these things that we are talking about here also apply to both instance and a binding, as well as the service progress. Well, and it's.

A

Like what pilot is.

E

A

Machine so the way I agree with you and I chose the binding as an example. That's the simplest because we don't have to talk about cascading deletes with bindings, since it's like a leaf node. Oh sure, so. I expect that we can probably use this to like illustrate some core concepts. And then we need to have discussions about every other resource and I, like this fair.

E

Enough, I guess all I'm saying, is it also liked it even applies to fix a create binding right.

A

Maybe I'm crossing the streams, what's the application to create binding what.

E

Creep I didn't see what's on so.

A

Is the the touch point with create? When can I do a create? No.

E

Well, it's even in the create case. It needs to do a couple of things. It needs to call a protocol to go ahead and create the binding and then, if probably immediately die after that right, then there's also a secret and service instance. What if we call it's the injector policy and things like that community crew, so it seems like we should have a similar mechanism that we use a PPS to describe the state that the the resource can be at any given point.

E

And what are they the possible ways it can go from there, and this one just seems to me like more of the same, but maybe I missed understanding why this one is special. So this one is special.

A

I think it's not uniquely special, but it's different from the the flows that we've discussed before, because this is where we're entering the territory of um we need to have like fairly a high degree of control over the actual REST API implementation to implement this. um Does that make sense like, like the the flows we've discussed so far, are like you declare your intent by creating a resource and now we're getting into you declare your intent by deleting a resource, but your resource is not initially actually deleted.

A

There's some things that have to happen and we're talking about the interplay between the the deletion at the API server and the controller, and- and this is where I'll I'll pop a frame off the the stack in this conversational stanza and say. I think this is this head. You've correctly pointed out. There's there are other concerns like this.

A

That I think this discussion kind of helps to ground and socializes the idea that, like we need some degree of control over the actual rest, api implementation that backs the api that we don't currently have and that we're gonna have to employ. To use this. To make an API that works, the way I works in a sensible way does that make sense. I.

E

Think so, but I will just shut up and let you roll I.

A

Mean I actually I think this is really good that you raised this because I like be late. Vla has raised a very important point because uh there there is this issue of like we still need to talk about unbind or I'm. Sorry, we still need to talk about any provision. We still need to talk about um what happens when you go and delete a broker right like and in fact, on the subject of deleting brokers. I think there's a fair amount to talk about these Ivy cascades, these a via leaving workmen things in what happens.

A

So my intent here was not to prescribe anything basically just to put push the design discussion a little bit further I and continue the design momentum that we had in a face-to-face.

E

Right so I guess.

E

So bleh I think about this for right or wrong. Is that when you create the resource it so when, whether it's a post instance or whether it's you know, keep control create, you know, whatever service are blinding whatever else a resource is created, and then it takes on a life of its own, that is controlled by the controller yeah and the way that the David resource is communicated and user. On the left there on the on the diagram you drew is via conditions, yeah, I.

B

E

Iii, like in the conditions to states in my small little mind and I, think that any one of those conditions, if it is a state, it has a transition where it can go from that condition to something else and and and and I think we are I, guess all I'm arguing for is that to me that state diagram is for the entire lifetime, not just the deletes, because I think we have to say ideally the same mechanism to continue to communicate the state of the resource.

E

The user I think we can agree on that, maybe so that if we deleted or created or provisioned or whatever, that the state might be it'd be great. To have a simple megan is a similar single mechanism, and I think that that is conditions yeah, because I.

A

E

With what you're saying and if those are states, then it seems like maybe we could take the binding, maybe step back a little bit more and say: let's go ahead, I. So first of all, I love this discussion, because I think we need to have this, and this is kind of what I was hoping we would have in Seattle. We didn't get around to it yet, but was the state machines like what what's they can resource be in? Let's say it's a binding and what are the possible states that it can?

E

You know, go and come from, but yeah you could stick rate fair statement, but I just want to make sure we understand that it doesn't start from just creating the resource. It's going to be in some condition. Maybe its provision or Bombay.

A

Yeah so um that'd, that is a really good point um and I actually have a pull request. That adds certain conditions to the API types that we I saw the ball, but.

E

A

E

No I was gonna say I saw that I was. I was curious, how you're going to drive that in with this one and the discussion and so forth, and I'm not sure, well, I guess the first question is: is how should we go about defining the state machine? What is the mechanism of that? Should it be something like just a state machine diagram with the things, or is that going to be a fancy-pants markdown document or how look?

E

What is the best way to go about that, because that seems like that's the crux of the from yes.

A

I I agree. That is the crux in like one thing that is missing to me, and perhaps if Brian were here, he could provide some counterexample to me is like uh I. You not think that there is any document other than code in Coober Nettie's that describes those types of transitions and so I.

A

My read of this group and my own personal predilection, is towards having some kind of document that describes this in some way, so that you can, you can look at it as someone who wants to understand how the system works and say say to yourself: oh okay. Well, this is what this API condition means. um I.

A

I don't have precedent that I can look to for like a super great way that that's been described anywhere. Yet, if anybody can point me to a place where it is described in user-facing, easy to understand language I think that's awesome. I couldn't.

E

Find any I spent looking iran and it's described in text and because English is my second language I like pic tures, burger, yeah,.

A

So, actually, you know what be late I also have this poor request. Number 40, where I have tried to.

A

Let's see I I made an attempt to begin describing conditions um and I would be happy to have. You know people's feedback on this, since it's basically around what has been discussed so far, um I made an attempt to incorporate the shins of the different API resources that were dealt with in this pull request into it. So.

F

So Paul, since we only have 13 minutes left, was the point that your discussion in the chart are the stuff on the whiteboard to get into the way we determined that something in the process of being deleted, or was it more about the elite process of rugs? I thought I was more about the elite process and not about conditions versus a state property.

A

F

A

Kind of touched, both of them and basically the the primary point, though it was to get folks thinking about um thinking about this because I, you know, I, don't I'm not interested in being dictatorial and the the consensus and socialization of of these concepts is really what matters rather than what any one person wants to do. So I don't expect to Korea agreement on this um I wanted to expose the concept if there were not any strong, like knee-jerk reactions against it.

A

What I'd want to do is try to write it down and write a picture drawn picture in some way that we could make an app or request to have this knowledge lived in the repo that we could then agree to, or veto descent from, etc, etc.

A

Make sense done yeah.

F

That's fine I, just think it seemed like. Maybe what you're trying to present was to the overall notion of I know: lazy, delete, isn't the right phrase for it, but best thing: I can think of right now right we sort of tag it for deletion and then we'll get some point later, you'll actually vanish.

F

So from the client's perspective, as you said, he'll get back a 202 if he keeps doing the sinking and over and over he'll, probably still get it to a 2 and at some point the tool to do was switch over to a 404 yeah.

B

F

The only question I would have is on subsequent deletes from the end-user perspective. I'm, assuming those are pretty much no ups, as opposed to he will redo the lazy delete mechanism, meaning let's say, for example, the lazy mechanism or the the indication that something is going to be deleted. Is you set the deleted timestamp right from Mel to current time, each subsequent delete from the end user does not re update that timestamp? It's a no up right, yes, yeah! That would be the only question I would have in my mind. Ok yeah.

A

So this is actually the we're getting now back to the point where um the then I was started, describing I and it's kind of a messy thing to talk about right, because it's like all right, so you gotta delete. You did a delete. You got, you know a good response and after that controller, the and I was going to explain the pattern for how this is implemented with namespaces, where what happens with namespace deletion.

A

Is that, like, as the finalizer is for a namespace run, they take themselves out of this list of remaining finalized errs and the API server accepts deletes when that list is empty. So in this case, what the controller would do is it would take itself out of this list of finalize errs on the binding and then once it was, that update was successful, it would do a delete and that would be the final delete to the API server. The rest.

H

Api implementation.

A

Says: ah ok, this thing has been finalized, I'm good! I can actually delete this thing now. Make sense.

F

um No, that one I mean be written down. Cuz, I, I didn't see the relationship between the final Wiser's but I think I, think I, see rundown they'll become clear. Yeah.

A

So it's it's written down pretty faithfully to how I've explained it in pull requests Oh. What was that number 39.

A

So since we are getting towards the end of the hour here, perhaps a takeaway that we can have is, if folks could look at issue 39 I'm. Sorry, it's not a poor best! That's an issue issue 39 that will probably bring some clarity and I will I will try to take some time at this point. It will probably be tomorrow because I have some stuff to get done tonight.

A

21 see if there's a write-up of the finalization pattern in goober Nettie's to see. If I can get a more concise description of it, um and perhaps we can. We can have our next discussion about about no pun intended finalizing our group understanding of finalization.

A

So since I had the bully pulpit, I, wonder I think this has been a really constructive. A meeting for this group um I am happy that we worked through like some of these governance things and that we're talking about design again and what I wonder is. What's the appetite of the group to have another meeting tomorrow,.

D

I'll be on board: okay,.

A

C

Yeah I think we do if we could do the same time. Same duration, okay,.

A

I think that would go a lot I this this to me is probably the closest we can get to a face-to-face, we're all locked in the room together. So if people have an appetite for it, I can make this time frame work every day of the week. If I need you.

F

Just one question: I just look I'm okay with meeting tomorrow, but as I was quickly looking through issue 39. In that main paragraph, we talked about the finalizer. You specifically talked about finalize errs and namespaces. Are you using namespaces they're, just as an example, or are these finalized errs that operate on bindings part of the namespace? So.

A

I and I'll go back. That's a good question. Doug I'll go back and clarify. I think I probably rode that at like three in the morning, um so it's probably not the clearest, but I was using namespace as an example because namespaces where the precedent is implement a tease, okay.

F

That's what I figured I just want to make sure: okay.

C

Thanks yeah, so can I go back to a dumb question and I just want to verify. Are we considering any case where there might be two or more controllers running.

A

I think that we had agreed that that was not something that we were going to initially do. Okay, so I um yeah. Okay, so we have about five minutes left.

A

Is anybody? I am personally unsure of whether there is a unit of work we can complete in five minutes. So unless anybody has anything they want to raise to the group, we can adjourn for today and I'll see you tomorrow. I'd.

C

Say if someone can can at least start thinking about that state machine diagram for lack of a better term starting a discussion around that sort of deliverable I think would be super help, both in the online context, as well as, like really said, some of those creating instances, service classes etc. Context, let's.

A

Call it what's called a condition diagram, because many incur benetti's have a knee-jerk reaction against state machines, and since we want to make something that follows the the patterns, let's just call it a condition, diagram and.

A

Avoided future knee jerks.

C

Sounds good okay,.

A

C

Imagining than a PR with a condition diagram on it, where we all can comment, / ask questions not only about this particular finalizer logic, but also about which conditions are acted upon. When do we go into the failure, condition and several other kind of things.

E

C

E

And those I think are the important cases, because I think that I mean just just looking at the thing Thanks for writing all that stuff down Paul. But you know just just following along a happy case and then having branches all over the place. Where you know things take a turn for the worst, then you know then then called the section 7 B seems wonky to me, but does anybody so it does? Anybody have concerns with coming up with a condition diagram, in addition to words, I.

A

Think it should be totally doable and I think it is right and good to want to do. Okay.

E

It's a second question is: does anybody have a open source, cool or whatever mechanism they have for creating these? That is Cooper, Navy's approved or I? Unfortunately,.

A

Do not think there is any such thing. Ok, that's.

E

A

Was going to suggest, like I, think it will be helpful to think of these instead of states as a of conditions mmm again probably like, if, if we have to shove it into like a state machine like thing, maybe the state, that's in a circle is just like writing out a double yeah.

E

Yeah yeah I think that's different ways to do it. I'm just curious about like drawing. If there is some state approved, drawing.

E

A

The the people's condition diagram generator that came out of the Cultural Revolution, I think, is with drew yeah.

G

I've seen a couple different places in communities where do was used, DML dating let's submit.

H

G

And the DEA file, so somebody can check that out. Yeah.

A

G

Is pretty good.

A

If you're, if you're old-school, there's also an ex fig, oh, I think there's a number of web tools that will generate stuff from like animal. For you, too,.

E

Okay, I'll take a look at it, then I can take a first crack at putting together a what something like this could look like. It's gonna be very incomplete, but at least it's gonna have some place holders to go and start the discussion. Unless somebody absolutely minds this, I can take an action identified. Ok,.

A

Then that would be awesome feeling. Would you also look at pull request number 40 and see if it makes sense to you yep.

E

Then I'll try to go and use those things together and and and put together a skeleton. So we can say is this something they even want right? Is this one of those sounds great, but in practice it's total crap, I I.

A

Expect that it might be, it might take some effort to get to the point where people are happy with it, but I think it will be tremendously valuable to have to explain to users of this API. That's we're at time now. Thank you, everybody for for your time today and coming in and joining us and I will see all on the Internet. Thank you. Bye, bye,.