Cloud Foundry CF for Kubernetes Monthly SIG, 23 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: cf-for-k8s SIG [June 2020]

Description

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

A

Well, here we go so efficient, Thanks, so official welcome to this week's foundry, who, when it is especially interested, call I think this one has a bit of a history attached to it. You might have realized that we had to cancel actually the last two calls- and this was I would say partially due to question round like what did people actually vote for, and they said they wanted to get an ireenie update. That was one part of the question and then I think that kind of went in a couple of directions.

A

So I think we ultimately settled on people want to hear about ireenie internals, but they also might want to hear about how arena is being used in both keeps EF as well as CF for K. So I think like for this call, as well as DD upcoming, two calls to be verified.

A

The DD upcoming two calls, but but like the plan, is to essentially start with ireenie itself, which thankfully house has a reach it to do, and then for the next I guess we'll look into getting an update for CF on on Kate's, on contacting in contact with SCI on that one and then, from from the ireenie sorry form from the cube CF team.

A

We have two colleagues from from suzer who are going to talk about specific topics in the area of cube CF, which probably also makes sense in like the brother CF phone on kubernetes space, so that being the intro I would actually hand over to couch. Okay, okay, as I said talk about like I, really specific, so I think people in in the poll basically says. They also want to hear like what what the ireenie team has has learned during by building ireenie itself, cetera cetera, but I understood.

A

You know again is also going to do a demo so over to you, Gary, okay, yeah.

B

Thank you. First of all, can anyone hear me good, because I always have trouble with this at Cisco Thanks burnt, yeah, yes, I, don't have any slides, prepared, I'm, just gonna share my screen do a quick demo and guess and I guess who just have a like you're in a session afterwards. I'll also talk about our current work like interesting things that we're doing at the moment and things that we'll like, we hope, come soon.

B

So I hope it's interesting for everyone. Let me try and share my console.

B

Cheering yes, it.

B

Alright, ok, so I, don't know like how familiar with irony Harini is the next generation runtime for call foundry to like. Basically, our goal is to be a feature parity with diego, but to run on two varieties. So basically, it's fairly similar. The gwelf controller is our main consumer in the CF world. It's going to like whenever someone does a CF push, the gold controller will go out to arena via HTTP and then arena will make the app run on Cuban ranges and I'm going to demo that right now, I have targeted a cluster here.

B

It's okay, now one of our def clusters with a fairly like latest and greatest arena version, and this is deployed a little deployment with CF or gates like we, we've been working with SEF for a long time. We've recently moved to share four gates in our pipelines and we've also tried cube CF. This specific one is running on cfirkus gates. As you can see from the like layout of the cluster, we have the CF system, namespace we're like Iranian the whole of cloud foundry lives. We have the CF work vaults in the staging, so split.

B

This come over here a little bit and do some watch commands I'm, going to be watching specific namespaces like most importantly, CF workloads, where Irina is going to be shading the ports and, like stateful sense, equivalent subjects backing the application and also I'm going to watch the staging pane space like on a CF. We were also responsible. The arena team was also responsible for spinning up the staging job, but on several gates were using capac. But anyway, let's watch the thing in progress.

B

See a for Claud staging watch for all resources, I'm going to push an app, that's called catnip. This is one of the few that a tree is working with several gates at this stage, because I think Ruby is still not done in the gate back like they don't have a proper who built back so most of our apps, not really working, but this one will do and.

B

Let's push catnip and.

B

This should, and it first creates a service. This is not something that every need as a but I think that working does that. But here in the lower right you can see that get back started a job I bought actually I, don't know if it's keeper and his job. This is not responsibility of fiorini, so we can see it's running it's in it containers it's staging the app and after after it finishes will we should see some objects being created in the upper right, which is pretty much the safe workloads namespace.

B

That Irene is responsible for so currently Irina can shade you workloads primarily in one namespace, which is the safe for closed namespace. They renamed space in SEF. The namespace were apps, run all apps from all spaces running a single kubernetes namespace, but we plan these two changes.

B

As you see in a minute, so yeah staging is now complete and you can see that it really created a stateful set, which is the kubernetes object, backing the app and this state was it only in its turn, created the board, which is the one up there over here.

B

This is our app, so I can now go ahead and scale the app so, let's say three instances which will again the Gaul controller will send an update request to Rainey HTTP, and it really is simply going to update it. Stateful set.

B

And it should something didn't age.

C

Didn't change so yeah.

B

I had like I was in a scroll view, so I had paused the watch and but yeah you can see like we, we basically modify the replica count and the stage was set and it creates three ports. What's interesting, like I can do this again like back to one and it's going to set the replica down back to two one, and it's going to terminate two of those stateful sets what's interesting about our objects is that they have some annotations with, like application, data can quickly show you.

B

So, let's, let's look at the stateful set.

B

Can tip something.

B

Less so you can see it has some called foundry data like an application, ID application name, that's how we know it's catnip and it has replicas counts.

B

Biggest one: that's what we update when we scale so this is. This is pretty much like the basics of. What really does something we're currently doing is also like. This was the recreation of eller piece. You might ask what about tasks? That's something we've been implementing for a while. It's almost done, but I think we're waiting on a copy release like copy shoot, releasing xkp release and it will be usable for now. It's only implementing ireenie I have watched this cluster and I'll.

B

Try to show you what a how we run a task, because it's you're going to see here on the right. It will result in not you know, stateful set, but in a cube orientation job, which is the cube object for a one-off. So if we try to run a task for the oddball, catnip I'll just run our sleep so that it lingers around. So we can see it because there's no like so far. This is the only way that we can see.

B

Then you can see we have a job here and it results in another pod. So it's pretty similar to the airflow, but we're just creating another another object. There is also an updated with all that like it, you can correlate between these pods and stateful sets and jobs because they have the CF specific annotations.

B

Let's see were quotes. This is a job.

B

The name mission think.

B

It's pretty similar, it has the cloud foundry upgrade version name all these things, and here you can see that it's is defined the container with a sleep in it. That's what we requested. So this is. This is actually pretty much. What I really like is is currently an abstraction layer that more or less more or less Heights the kubernetes world from coffee, although that's also about change. So just take a look at my have some notes around here.

B

So yeah we're currently working on on v3 tasks, which are finished in the range light almost completely, but we have to wait for for recovery release. I said earlier that it really is operating in a single namespace. The rename space with chef workloads namespace on the cluster. That's about to change. We are about to introduce.

B

The capability of, like CC should be able to specify a namespace where, where we will spin up the jobs and stateful sets we're doing this, because this will let CC and the system like the higher-level system of called foundry, to define what apps should run where, because currently we might have different orgs and spaces, but in the end all the apps are going to run in the same kubernetes namespace and by letting people run their workloads in their main space is provided with them.

B

We can use, we can better, isolate I guess we can better isolate the apps brink on gates and another another nice thing that we're working on is a C or D API, because currently arena is exposing a REST API. Just simple HTTP server, much like Diego do I think it was RPC, but it's similar and then Gulf controller has point-to-point communication with with irony sending case, T requests. We want expose our own CDs, which is pretty cool because then we'll have.

B

This, like we'll, have more, not point to point but message like architecture where we define two major lr2 major series that will be the LLP and the task and by like simply creating a young object on cube, you can tell you need to materialize that, in this respect is main space and I can actually show this to you as well. I have prepared some like.

B

We have already partially implemented the one of our series, the lrpc Rd, which you can tell. If you go on the Gloucester and do I get sear these, there should be an LLP door there a need or something. This is our eller. This our CID, we can can try to show its schema.

B

Right one, so this object is pretty much one to one mapping roughly want one mapping of our current JSON payload that the call control is sending to us. It has the cop quit the app name resource specifics, like disks routes and so on. We we have the two life cycles, which is like built back in docker life cycles, and the most interesting part of the resource is its status, where we will be keeping information about all its instances and something interesting like when moving to sear.

B

These will have to rethink some some parts of the communication, because the with the golf controller, because currently when we are creating a task it while controller, sends a put request. We create the task and then, when the task is done, we call back to the quad controller. This is not going to be possible with with you like that, rather than that, this 12 controller will have to create that CLG and watch its status.

B

For example, it might be bending running failed and this way we want, we want to reach back to the CC anymore say at this task: filter this lrp crash, but the 12 controller will have a convergence loop or something like that. You'll be watching for our resource, and we think this will be a way more kubernetes native way of doing things and way less astonishing fool to the community and I think this was enough words I'm sure you have questions afterwards.

B

So without further ado, let's try and do a very basic demo of creating. How do we push it up with CDs? First of all, when, when we're creating, when we're grading can up with the surgery, we were grading the CLD, the LRP in some specific name space, and this name space- it can be the CF or close name space, and this is going to work, but it can be any other name space, and this is the other thing that I mentioned.

B

We are trying to give our users a way of specifying what's their name space, but this comes with a bit of cost. Currently, the safe worth of the space is generated by arena. C on chart it is, there is a service account in there. There is a role binding. Both security policies created. All these things are going to have to be done by D by the user.

B

Now, because this is their namespace, so I have prepared a I have prepared a llamo with some boilerplate like I'm, creating a demo namespace I'm, creating a service account and binding it to both security policies, pretty standard stuff, it's letting me run as any user use, some new IDs and so on. So this is going to have to be provided by a user. So, let's apply.

B

Let's keep going to apply that.

B

B

And there we have it our demo namespace with all the things in there that we're going to need, and then we have a lllp llamo, which this is a simple, tiny little lrp. Most importantly, you can see there's going to use a docker lifecycle to run a docker image. This should also be able to work with with a standard like built back a life cycle that we that we have like. We are setting all the good versions. These are going to be set by the ground controller in reality and yeah. Pretty basic.

B

Most importantly, number of instances is to know not a lot of cool stuff in there yet, but let's try to apply it and see. What's going on, let me I'd like to need the workloads anymore, because we don't have we don't stage docker images more, basically, so yeah that's to apply, but it's important. We should now apply in the demo namespace, so keep crypto namespace demo apply file, LFE.

C

You don't watch the demonym place on the right, yeah.

B

You're right now, just.

B

B

And we have quite similarly like quite similar to the CF push. We have a stateful circle its respective to two ports. Unfortunately, does the only thing I can show you I mean if we edit the thing I, don't think I, don't think we've implemented scaling down. Yet if I did there appear, nothing will happen basically, but this is. This is how we envisioned to to be able to run apps in the future. I have also prepared a service, because I have now bypass the cloud controller completely, which means that no one knows about my app.

B

There is no service, no root created for my app, so I can again in the demo. Namespace apply a small service, that's going to make it accessible on the internal DNS. It's.

B

Exposing the 8080 port on 8080 and it's basically bounce to my app via these specific quit labels. Unfortunately, this is only internal DNA, so if I want to hit my my my app I need to work on to like some port, because this is not it's not visible. Outside I, don't have an ingress on it or anything, but you can. Try like in the CF system, should be able to go on to the support like doesn't really matter which one I think this one has rush.

B

Put the name of day in a solar cell: this is the command I need to exec. This is the you know that I've exposed anything I'm, not Dora. This is one of our desktops and yeah. This is this is prema I can show you our series, it might be interesting for you, I might send you those links. There is a Google Doc talking about some design. Aspects of our series. I might need to copy this.

B

So this doc contains more information about series in irony and how we, how we intend to like to implement all the things that we're doing right now, like matrix, was ever going to do about callbacks and so on.

D

I've got some burning questions to ask.

B

B

Let's, just let's just go to the cue and.

D

I'll go with a simple one: how come you all picked underscores versus dashes and the labels and annotations is there plan to change that.

B

D

The app and the score gooood versus like app good.

B

Probably historical I haven't heard of plans to change that, but I guess, if that's a I guess open issue, that's something probably easily do.

B

D

The I guess probably the I, already forgot two other questions. I was going to ask, but the last question I had, on my mind, was kind of it was a bigger picture of maybe in future Cappy being driven by some of this year, Dee's exposed to the user. Does it actually make sense to keep that separation between Cappy and Iranians sounds like if Irina is planning to go into the CR deal? And then you know, there's a factory day, the APF boundaries kubernetes, and so it's not really pluggable anymore right.

D

It's like it's not as pluggable as like the G I go back and right because Diego isn't talking via kubernetes yeah right. So what is the you know, given that you know? Maybe this is uneducated perspective, but from what I understand adeney is just spinning up stateful sets based on you know the record that cappy provides it is there.

D

Is there actually need to have this separation between Kathy and Union, since Kathy will be watching kubernetes api? We might as well just have kappa credo, stateful cells directly and you know, save the hop yeah.

B

We've had some discussions like this. You know like you're right. We have this point that the world is going flat a little bit with kubernetes, because from this multi tier architecture, everything pretty much reaches down to the to the cube level. Right.

B

It's like a message, bus where you put seed a series again and of like messages, so it might as well end up being that, but I guess it's probably too early to say and I guess, we'll probably see, because Kathy has other things to do like I, don't like keeping the model and creating the initiating staging and orchestrating all that things. I'm, not sure. If combining everything like if if copy should be like just one component, that does almost everything but.

D

If you have to end up creating those theories anywhere right and I'm guessing there isn't really. You know unless unless we have some future plans to make the app make up the more complex beyond like a stateful Sarason or maybe a deployment right, it seems like a pretty heavy. You know it was a necessary abstraction when we're trying to let's say make it work for kubernetes and non kubernetes right. But if we are going into the kubernetes api land then feels like a pretty heavy abstraction. To have I mean.

B

E

So, even if you implement CDs at the cafe level, you'd still wouldn't implement them in Ruby right. So you might as well just go every kneecap. If that's what you need right.

D

Well, Cappy holds a lot of other stops that is doing right, I know, I, don't I, don't imagine it going away anytime too soon right, and so it's probably unreasonable to expect that you know a lot of the logic that it contains will be shifted into it in your right. So I. Imagine that whatever it is much easier to pile into cabby, I I.

C

Could could imagine that, like I think what you're referring to the dmitriy? Is that like copy letting the user create the CRD and then just create another CRD that somebody else picks up and that creates a community resource or another CRD?

C

If that is the point that I could imagine, a world word like copy, creates an app C or D, and there is networking picking picking it up to configure the network for that app and report as a condition when networking is set up that ireenie uses that same CID picks it up starts the stateful sad reports. The status and copy does other things too. I could have managed something like that. Instead of having like a whole bunch of C or DS like at least one / CF component yeah.

D

Because if Kathy, eventually, you know enables like, if it adds the first Class C Rd user experience, that will be like, let's say, called app or something like that right, yeah, and at that point you know how many you know how many CR IDs that that actually need to need to involve right. It depends I could I.

C

Could imagine that it's just one that it's just one for that one and that other components just pick it up report conditions if they are ready, but that still the copy team doesn't want to own a decision like? Are we using stateful sets or deployments or is? There is some justification for even a middle ground, because we don't want stateful sets but still need something from that. So all these decisions- I'm, not the copy team, actually wants to own them. So.

B

There are things like setting can virus like there is a contract in ViCAP. You want to own that, like I, mean I'm pretty in line with what I just said, like that's what I've been by saying message like architecture, it might be just one searly, although I think it might be more than one, because it would be a hell of a synergy too many details. Why not like?

B

Why imagine a world where CFC like race acidity, then copy watches that CD does some other see all these chain reaction, things that we're used to seeing in kubernetes like? Why do you need a stateful set when you can just create replica set I mean it's adding a tiny bit of functionality and that's I think that's similar.

B

What we're doing here with arena copy and so on, I, don't know what the final division of components will be, but it's an organic system I mean something might get merged into something something might get written and get deleted. But I guess it's like evolutionary kind of thing that.

C

There's by the way, one thing I really wondered about why the two different life cycles, like ireenie, that the lrpc rd would not trigger actually Capek to stage a bill tag application to an image right that that's at least something word I'm, not sure what the direction would be.

B

Fortunately, I'm not too familiar with kpac, not sure if I can fake it now.

C

What I meant is like the irony CRD, should that really be concerned about two different life cycle types and one of them would actually require to build a docker image. I, don't.

B

Think so I think the docker think will become the only yeah. Okay.

E

B

Mean we're currently switching the defaults, so let's say that dr. think will become the primary thing, for example, cube CF, guys, I, think they're still using classic staging, but I think they want to get rid of like bits and all that so I think this is the direction funny.

D

Things I got so more question, it's no one else jumps in.

D

Good, so that a question that I remember it was regarding tasks. This may be a little bit down into implementation details, but how you're planning to deal with the sidecar running for like anyway, for example? That's something that you.

B

D

There is there's going to be some lifecycle management to do rate to, for example, let the anyway know that it's time to shut down so that the job can exit.

B

Currently, we're currently have a spike in flight I think we're going to pretty much run another like we're going to read all the side, cars. The C's are going to pass a sidecar I'm just going to spin containers in the same put.

D

Prayer right, so how do you control the lifecycle of those containers?

D

So how do you how what what tells the end way that it's time to shut down or in other cases like what tells the the main task that the user is running, to wait to not to start before the envoi is ready and you know, is able to I, don't know. Do the processing of the connections.

B

What what does it, what does these things currently Diego.

D

Well, I think Diego is you know, because the model in Diego is very how you call it specialized right now, I'm imagining that they're doing all kinds of my orchestration themselves right, so they spin up and boy wait for it to start over ever then run the thing then, when the thing is done, then it probably kills off my rate, which is not necessarily something that you could be doing with the pod without trying to put like some some shims around user-provided tests.

D

Alright, if this is not something that you'll have I guess.

B

It might be a bit too early, I think we might get back with that when we wrap up this bike.

D

Yeah I'm definitely curious, they're, more discussion about it. Since various projects out there.

B

It might be useful to have some discussion because we're not too we're not too sure about all the different. You know different use cases we have with sidecars it's a bit fuzzy for for us, so it might we might reach reach out for, like you have some questions. I mean definitely ask in the channel because it might give us some ideas.

B

D

I was like I was mostly interested in the invoice. Id card said that you know the task can be doing useful things and talking to the applications or whatever that can be secured. Somehow I didn't have anything else in mind for sidecars, but that add another separates question / note: I have as I notice that the lrp contained M, section and M section I imagine would hold some sensitive data. Are you all planning to make it referenceable to a secret.

B

Thought about that, but maybe maybe it's a good idea: it's fairly experimental the whole the whole CD thing, but I guess we might run into things like that. But no we haven't haven't considered yet, but yeah.

D

B

Thanks yeah for for the side cards, I guess you might want to ask in China, because other people's might have more knowledge or opinions, and if not, we might just have a discussion there.

B

Something else that we're working on currently is redefining the whole surface area for arena, because you know it we're currently releasing a film chart and we're pretty close to deprecating that, if not deprecated, that already we want to start shipping just a static set of GMOs with a config map. So, basically, in order to deploy Rini, it will be way easier. You'll be able to just keep go to apply.

B

Of course you have to provide some secrets like, for example, the cloud controller secrets, the key and the served in the CA, and things like that will need to be provided by convention rather than templating and in the config map. You'll be able to tweak all sorts of useful things, and this will make it. We think very easy for people to just like try the thing out and of course it doesn't. It doesn't make like using quite a teacup he'll and other things.

B

The question like I think cube CFR, currently using the help chart so yeah. This is something we're quite excited about as well and I'm interested in in what you'll think about this direction like is there someone who thinks that's wrong or who thinks that it will like break something.

F

There's one point from earlier that I wanted to go back to super quick. Sorry, it's not a direct response to your question about packaging, but it relates to Dimitri's question about the formats of labels coming through and like the pods and the stateful sets and whatnot.

F

This is something I think by the way product manager on the logging side of things uncie F for Kate's. For a little bit of context. We care about some of that meta data to make sure that information is correctly annotated, as it goes through our own logging system, and we would be very interested I can, oh, if the format of that changed, being able to update things on our end and be able to react to any of those changes coming in from ireenie.

F

Is there any good place for us to track these sort of changes like release, notes a set of architectural decision records? It's like what's the best way for us to be in the know, if these things change.

B

F

B

I, don't think we have ever changed.

C

D

I suppose like if.

C

You agree that this is the right one. It's for the very specific problem, then just open a github issue to change the annotations to contain dashes instead of underscores, then you will be informed by get yes.

B

C

B

Thought I thought it was more about how to find out if something changes right.

F

Because for us we don't have a strong opinion about the format of these like, like there's community conventions. I think that Dimitri is hinting at the like, so they contain underscores is anything we just care what they are right and if they change you know, in order to kind of match with those conventions, then just having some way to stay up to date. On that or other relevant changes like, for instance, we might have some dependency on the name of like container names coming from ireenie, and things like that. Well,.

B

Yeah I'm afraid this is like too low a level to put in release, notes and I. Think as of today, I don't think there is such a place. I guess you should just I, don't know like you keep in touch with the team or like.

B

If you discover I guess you should just discover those today I guess that's what you're probably doing in the future, though, when we're going to sear these I would imagine that they'll be carry or carried over from the CD, and the CD will be the API, so I guess it will be in the readme and it would be, it would be quite visible. I would imagine I mean that's just my personal opinion. I haven't had such discussions in the team for sure that's something I.

B

D

I would add to that. Is that a lot of times the the metadata passed in labels and annotations right is part of your API. So, even when you all move over to CDs, you know there are patterns with, indicates community right where, like it's expected, that the pods show up with particular, you know labels right and so, if I would actually argue that this kind of thing should be included in the release notes because you, the external contract, you know with the user. This.

C

D

Something that you all expect other people to use right.

B

And that's fair because even with serious, it's like you have an input and an output. It becomes like a function, so I need I. Think one.

F

Particular exalts sorry to interrupt you Georgie go.

B

F

I think one particular pattern here that there's a lot of similarity. These kind of questions is having to do with, say the format and labels on metrics coming out of these various components. That's something that you know there can gradually accrete a number of downstream dependencies on, and so it's often asked that, like hey, you're, gonna change, the name of this metric or the label of this other metric, like just letting folks know so they have time to adjust.

F

I think that's kind of the situation we're in consuming some of these annotations with a blogging system.

B

So what what would be useful for you then like do you want to find changes to these in the relief? Not is that enough, or do you want to find a section in the reader? How do you imagine I.

F

Think I just asked you if there are changes that are coming downstream to parts these annotations or labels that feel relevant to mention that in the release notes, don't help us out a lot just so we don't have to be like. Oh man, ireenie version this shipped and now things are broken and we don't know why. Yeah.

B

Or just in general,.

F

Mostly labels and annotations on the objects coming out of irony would be the ask here: okay,.

B

F

B

F

B

Both stay for sets jobs they're, like.

F

Pods are I mean the most important for us. It's I, don't think we would probably be the only people in these situations like making some decisions based on these, but for us annotations and labels on pods will be the thing we would ask for notifications on, but.

B

Did did we change these things? I, don't remember like us changing that a lot I.

F

Don't know if there have been some added recently due to requests. There might be some change in the future, depending on kind of sort of things. Dimitri mentioned just wanting to make sure there's a way for us to stay.

B

I'm sure I'm just trying to get a better idea for the problem, so basically promise that you didn't find the ones that we added as part of the release, notes well.

F

It's it's that we have a dependency on the name and form of those annotations like for kind of functionality, rather than just user visibility. So wanting to make sure that we would know if they did change, because.

B

I think shouldn't be a bunch of a problem right because it's backwards compatible generally right.

F

B

B

All right, we can keep this in mind, I'm, just trying to figure out what exactly we did. No.

F

No, it nothing the ground, just something I'm asking you to keep in mind for the future sure.

B

D

The rephrase it I think, what's being asked, is to the API contract for annotations and labels more explicit with certain backwards comparability guarantees and discoverability of that contract.

B

Maybe maybe even in the readme, maybe we should like examples like I, don't know, should kind of quote what annotations and labels we put on these things.

B

Compatibility I think this was definitely on our our minds. We've been mainly adding but yeah. Maybe we can post them in the readme and change like give them up-to-date or.

B

I'll tell the PM I guess about this request: excellent.

F

Thank you very much think.

B

Something sure any comments about the helm, less deployment think like you've got to apply these yeah most with a config map. I.

C

Like the idea, like I like that, like that the the time Jules mentioned it in the industry call that this is how he imagines it, like some house like templating, similarly to using mocks in tests like when you start using mocks and tests, nobody using you can get out of the smoking hell and similarly with templating right once you start templating. All the upper layers need to use that templating and if you just provide that the upper layers can do whatever they want. Like.

B

It's an old idea. This convention over configuration they'll, definitely be some convention introduced because there's no way around that, but I guess, given that we have serious and a config map, is more likely something like a fluid CD like any surgery you want, and we can. We can watch that and reflect those changes, or this can be our main configuration point.

B

Quite I think it will make our lives easier. Definitely and if it's not burden for our users to be great.

B

There is also a document, not a document but yeah. Basically, a pre shaped version of what this static deponent might look like the poster coughing about just paste this in the chat you.

A

Could also send me, like.

A

hmm Thanks thanks.

B

I'll send it to links one about our current initial, like proposition for the CD API tasks and LR piece and one about the Elm glass deployment, because I think these are important topics and people might want to have a deeper look just to disclaimer. These are just initial things: nothing set in stone there. So definitely a comment and get feedback like Charlotte, Channel and so on.

E

One question: so if you don't provide ham shots anymore, how do people know what they have to put in the secrets? The that you need to create right because I mean ahem chart some, it's just a bunch of template's right. You can ignore it and you can write the objects on your own. It serves a little bit as documentation if you don't want to use it.

E

So now you don't even do the ham chart which I can understand it's kind of it's a user facing thing, and if you want to concentrate on your software it can be a nuisance to keep it right and to provide it to users. But now you don't have to hand shot anymore, so you have to document um what has to go into these manually, created conflicts and secrets and certificates that sound some like a lot more work. Then you know just putting some files on the template directory.

B

Well, it will be, it will be just some convention. We hope that it will not be that much like things like the smoke of got the cloud controller certificates. You'll, probably anyway, need to specify in the values files in the column. So we hope it won't be that bad for the end-user plus I mean every anyone can write their own template. It's not. You know it's not forbidden to do that. You can just like take our deployment. You also parameterize them if that's more convenient for you, but we think it will generally make it easier.

B

Actually, it will be definitely already documented. This will be kind of the contract. Now that we don't have templates, the input parameters will be documented. I think it will be mainly secrets if you're interested take a look at the dock when burn sends the links. If you have some specific worries, just just write a comment in the talk or like I think it's a github readme, this particular one, but you can always write in the channel, will be will be glad to like. We will have to have a discussion about this.

E

A

So, as we only have two minutes, maybe one last question for me: any plans still on getting rid of the stateful sets.

A

B

A tough one actually delighted to just just out doing just deployments, but I think there are people. We have had some issues about people doing some migrations who really want to have instance ordering which seems like it was a wrong idea because I don't know like how do you guarantee that instance, one didn't get recreated and stuff like that? I mean it's kind of dumb to be talking about instance, you want to when you're talking about a cloud, but people are depending on these. These have been documented.

B

The CF Docs there is a there are some in VARs people depend on, and not only people will also tools like cloud connectors for spring and stuff, like that. It will be kind of a pain to get rid of these. So for now, I think we're more on the side of keeping those for a backwards compatible depend.

C

Do people depend on these environment variables or do they really depend on in which order things are rolled out because that's in Reverse with stateful sense well,.

B

Honestly, I don't quite know, but there are, as I may, as I mentioned, these cloud connectors, which lookup those virus and it's it's quite annoying, because kubernetes injects, like the stateful set controller, injects the name of the of the boat like your ID number, but it doesn't give you the number separately, so it's kind of a pain to support those use cases. But given that there's been issues, we don't think it would be nice like drop that.

B

Yet maybe we can deprecated that and to like, let people you know, let people adopt stop using them and then we can drop but I. Don't think it's something we can drop right now.

A

Make sense, Thanks and I think we're at the top of the hour, so I would say. Thank you very much to to be okay for demo and answering all the questions that came up. I. Guess if there's more questions, you know how how to fight the folks, let's check ireenie slack channel, I I guess, is the main channel too to give additional feedback and with that I'd say. Thank you very much to everybody for attending and we'll talk again next time see.

B