Rook Community, 23 Oct 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2017-10-23 Rook Community Meeting

Description

Rook Community Meeting

A

Zombie 1, hey.

B

Guys, can you hear me yeah.

B

You're annoncement.

A

B

I was hoping Alexander returned. He was quite active.

C

B

You guys see my screen the page.

B

How about this some.

C

Call you will see the vertical screen he make the window wider yeah. Is this better yeah.

C

B

Maybe we can start with where we are with zero point. Six I have a list of things on Mike that tracking but I'm curious to hear where we are so the Steve in the room. No.

C

He just got here.

B

Okay, so can we start with bill? Shoes looks like builds, are still failing, so the parallel stuff is not. Has it helped or not?.

D

B

So we need to look into this I'll. Take an action item to go figure out. How to this is the can find import issue? Yes,.

C

E

B

I was suspecting that's the case today, all right, okay, so I'll. Take that and we're seeing it frequently right. So it's not it's not like it's one. Ever you ten anymore. It's.

B

E

B

Yeah I still have a feeling that it's glide but okay.

B

Because if it is a shared directory, footlight knots, that's why I'm suspicious of it? Okay! So that's that one can we talk about removing the API because it sounds like we went down a path, freshly and not all the way.

B

It's the first stand-alone is gone right.

C

B

What are the remaining uses of the API.

C

And so there's.

B

C

B

We were having this discussion on the issue that there is still a need for it beyond legacy and that's what I'm trying to understand.

D

Not now, but maybe it is subsequent releases.

D

B

I guess the part I'm curious about is what is I understand that, where we haven't we're using it I guess what I understand is. Is there what? Where are the gap? If we were to remove it and assume it doesn't bring anything we're? Where are the gaps.

C

B

Okay, as the Sega API team, is turning this whole effort around declarative management and trying to improve CR, DS and understanding these gaps will help us kind of take that feedback back to them, so that that's why I'm I'm interested in kind of trying to figure out where the gaps are.

B

So you know the history of the root project is that we actually started out to be a to try to do an abstraction on top of Seth and now we're down a path of just orchestrate XF right.

B

There really is no need for two API. If there is a REST API for self, there is a. There are SF tools. If you're gonna drop below the sea, Rd abstraction, then Uranus you're in safe land.

E

B

Right and if these need to bubble up the CR DS or we have, we have to improve CR used to have monitoring or not much better status of progress. Maybe that's something we should take back to sing API.

A

Something I understand here still is that for things that don't really have, where does it make sense, like just metrics in general, you know any sorts of stats and stuff you're not going to have you know customer resource definitions and objects, those they don't pull out of the system. You know it's like like this does so, if you get rid of a REST API, how do you expose things like that? Would.

B

You bomb, why wouldn't you just use permit Ness.

B

Using exponents to get all detailed, perf, perfect, urns and metrics and analytics so.

A

Deploy other components.

B

No I mean use the exporters that are available in the you know. Your Prometheus exporter could be used to retrieve that kind of information. I think.

C

What you're saying is that with the C or D, so we get an API that basically has a bunch of operations line, yeah and you're, asking where the corresponding get operations and the right on this sort of thing that well you get some of that. By going around the corner of Prometheus, there isn't a orthogonal get operator to those foot operators to the set operator, and maybe that's what this API group this hands at I.

B

Think I think you're right and then you don't you see our needs to get. You know status or detailed status of curve stuff, but the path for that should be monitoring and logging, and you know they're. Those are the that's where I would go for that kind of stuff see are these are truly for declarative, management's. So.

C

Where would you go to find out how much free space there is in the cluster Prometheus.

B

Seems like a funny place to go before.

C

Yes, but is for me, here's what is that that what kubernetes wants to put forward as the preferred path for or yes, yeah.

B

I mean it's a good question: how would I, where would I go to kubernetes today to get say the cluster size? Where is that I'm curious? Is that an API call.

E

B

How many nodes in the cluster.

C

B

I mean yeah. This is this. Is again this feels like it's in the bucket of Ohio. Do we best use CR DS like? Should there be a status? Cr me, a rope status, TR t we populated bars that are really a misuse of series. I.

C

Think that is the question and that's why they would ask me: I grew. That's.

B

C

Can make something up and put that forward as the remote, but you know and.

B

C

Pick our head out like that.

C

C

B

We talked about that.

A

We're going to.

C

Duplicate state: that's not necessarily.

A

In sync- and that gets even worse when you deal with things like our DW users, the.

F

More state we tried and re-export outsider set as a as a mirror of it. The more more work we're doing, seeking that back and forth.

B

Yeah, like a really soft spot ground right now for me, is that there are CR, ds4 or declarative match and declarative state, and if you get a real-time self information or sub state, then use the Ceph api or set tools. That seems like very dry ground right now.

D

We always had I would I was in HP, will is how hughster doing metrics and monitoring, because if there is kind of like the coordinated way to do that,.

D

D

He has an API and it included like a slash MIT to slash memory, size or whatever, so I'm, not sure what animated wonder it's both with Seth and I'm, not sure if it's half feet how we fit it with hipster. But if there's there you know app to use queries. It's.

C

D

B

I guess what I'm trying to figure out is: can we arrive at a place where we're using the CR DS and the Ceph API and set tools? Initially it and nothing else, I mean maybe Prometheus too, but if we can get to a place where we're doing that, then I, don't we can remove the API moves through of CTL and then, from that point onwards we can figure out how whether we need to add more CR to use or improve the monitoring or whatever else.

C

For some period to transition for.

B

Sure yeah, no there's no I'm just trying to figure out what the path forward is.

B

So we are not adding more api's in the Brooke API right now that correct we're, not improving the road.

C

B

Okay, okay, so it sounds like going forward. You know legacy aside or existing support aside, that we would head down a path normalizing on CR DS and the Ceph tools and API.

B

A

Think it's probably wise to capture that yeah.

B

I'll open in this shoe boy outside well.

A

They can barely, they already have an issue open. The became his attic comments over that sounds good.

B

So flex volume: let's talk about the flex volume.

B

We should definitely add documentation in our dog, that should say rook, CTL and the rook API is deprecated sure. So nobody builds on I, don't uses it.

C

B

Good, so so the rook, the Flex volume, if I'm, if I looked at the issues correctly, sounds like there's still the issue of where to drop the Flex volume and support for things like rancher, core, OS and atomic.

B

That's still one issue we just try to summarize in, and you can tell me if I'm missing anything formatting and mounting from the pod and the issue of mount propagation. Yes,.

D

B

The inconsistency is a potential leak of the volume attached. Crd I still seems to be open.

C

B

Leaving a trailing Sierra following attachment see are these oh.

D

B

And then there's the.

B

Volume like I guess that's in the first gutter, either the whole volume plug-in dirt issue.

B

So so you know if I step back a little bit, our goal was: how can we get rook to run on any kubernetes deployments without having to touch the host period? Yeah right.

B

With 1.8 doesn't sound like we're there yet with the kubernetes well I'm pointing doesn't sound like we're there yet.

B

So and it's not just the read-only back there's actually, uh we have to our agent has to figure out where to drop the flex volume driver in a place that it could be picked up and then the mound propagation issue you know the sheriff on propagation is another one that gets in the way as well. But that seems like it's, that is, that solved in 1.8, Steve.

D

B

There was emerged to farlap, cubelet is now yeah.

D

That was part of the part of the face of a bigger face, so.

B

The bigger face.

D

Is I should prefer a ship where you can specify I want this part to be able to you, know mouth teeth and be able to sweep it by other posted part. That's not there. Yet. Okay.

B

So I guess where, where I want to what I want to ask, is how we get to a point where Brooke runs on any kubernetes deployment where ever without any changes. It's.

A

A propagation issue: that's that's only preventing us from moving the code that does mounting from the Flex one driver to the rogue agent pod right that doesn't have anything to do with the ability to deploy anywhere yeah. If.

B

We want that the jerod that's important, because if we are going to dynamically, make the decision of how to mount and what to use and how to format on even on a per per deployment basis or per pod basis, we'd have to move that logic, or else we're gonna keep maintaining the driver we have to drop by every time. We change a strategy. We have to up go upgrade drivers.

A

Agree with make sure so.

B

You know so roughly if we were gonna say there, there really are.

B

Our vision is rook runs everywhere, kubernetes runs and without any changes- and there is probably a second part to that which is- and you don't make an assumption that the entire kubernetes cluster is is homogeneous like every node is configured exactly the same way and right now, I, don't believe we're doing both of those like you have to make a lot of assumptions about what's on the host or pass on the host or whether it cubelet runs in a container or not their version of kubernetes.

B

You know mount propagation, the volume plug-in derp and if it varies across nodes, all those things are issues, but.

D

I'm hoping that all these issues you're right, those are issues, but it will be that CSI. You know that will resolve then.

B

There's nothing si si. Why.

D

Not because si si we're actually doing the thing holding two cannolis. What is that every please write.

B

The are our ability to run rook in an agent and things like mouth propagation, all that stuff don't or where did where to drop all the stuff. This doesn't go away, I mean what what I thought CSI is a team. They said twelve to eighteen months from when it started. Has that changed from.

D

Then that's a few meaning that I've seen I think they're targeting an alpha release on one point: nine, but yeah. But basically this CSI. It is what gonna be replacing the agent and all that work. But.

B

They're, the path for is I understand is that there are always been flex and CSI. That's not know.

A

This goes that flexes.

A

B

Well I mean so: let's go back to what we're trying to accomplish. Brooke runs everywhere and heterogeneous kubernetes cluster with no changes are we saying now that CSI is the first time we'll get to do that or are we saying that that's gonna happen with flex, because we invested in flex yeah.

D

Not because most of the a vertical mmunity are being focused on on CSI, right and developed, but.

B

Why can't we go make the changes necessary or help go work with six storage to make the changes necessary to fix our flex scenario.

D

I mean CSI is coming and CSI. He has all the things that all did the same thing that we have only agents with this attachment. You know math amount, all those words I mean yes, we can. We can actually implement it, I ourselves right, I'm, sorry this morning, but I'm not sure. If that every is very studious.

B

We just went down a path of flights, I'm a little confused by this. It sounds like we're saying flick. This is no longer sufficient and given.

D

B

Given all these issues that.

D

We need no samples. Only one thing was not to be able to be star Hewlett when you have a new driver that was like the most. Those are very painful, so that's why they agreed to do no.

B

I think no, that's not how I understood it. I mister it as we're gonna support the rogue agent and we're gonna use flex to essentially be are to be are its are essentially our architecture for being able to run Brooke on any coronaries deployment.

B

Yeah a CSI instead of flux going forward, but this whole architecture stands yeah.

D

There is a wide, a yes, the English only because they were kind of reluctant of introducing a new entry route plug-in because that, if that happened, that mean that they must support an API, because that would be coverage. They were lacking about that and also the relaxed about it. Oh API, layering on and so forth. Almost an attorney that.

A

We do out until.

D

We can see site OCS episode. His data we can use in commerce are going to discuss in this, so flex Monique was kind of in the intermediate. You know, you know short term solution for this and then we'll say: okay, we can know we can know, usually solely because of you know, things like we have to be when you.

B

Same will all be able to get to Brooke running everywhere without any changes. If and when CSI is right, it seems like a pretty weak story right now to me.

D

For me, from the beginning of this I I build it we'll say in a way that I came back with a code easily, once the SI was ready. But in my mind, I was on to me most well understood as I thought. It was gonna, go away.

A

But so it seems like this has not.

D

Here right now, if there's a good idea time period until.

A

We get CSI or in its tables. Well, I, don't see why if we can identify a concrete set of like fixed, targeted fixes that will help us realize our overall goal, I, don't see! Why do we do just say down with that right away like if you have a concrete set of things to fix? Let's take it to the six words or that's. You know, I think we could make forward progress, yeah and not just our hands I'm just.

D

Anxiety, I would just say that was not me, because even that don't possibly as I saw me, I think one good night. That's why her that's one point name like with the violence against them: they're exploding, giving that case, I, stumble or well.

A

Here's here's. My question is: do we know what we need? What does that concrete set of issues remove? Do we need or improvements that we need from FlexPod? Do we know that we.

D

Don't need you, but if you're, like those issue that you mentioned especially like they the whole the whole doubt, you know read only there. If things wonderful figure, I do not want an easier and employer no.

B

I disagree! It's if, if we require any changes on the host, I I want to make that are tissues so.

C

C

Time was great mood to see us up. I.

B

C

I mean not here will be here at some point.

C

Their response seems a little bit weak I mean we might know what some issues are. We have.

B

I'm strongly of the opinion that we need to get this story fixed on flex and when CSI is ready, we can and and the community decides that they are no longer going to support flex. Then we should. We should switch over, but I don't see why we don't complete our initial goal and given our existing investment.

A

B

So and so I'd like to know what these issues are, I think I know some of them: I'm not sure I understand all of them, but even if we said in kubernetes 1.9.

C

B

We've made all the changes we tried for one pointing that doesn't look like we're. There, I I thought we're gonna get there with 1.8 and we missed a few things. So now in 1.9, we have an opportunity to get to making the fixes necessary so that rook could run on any kubernetes to plant.

B

So well, okay, what? What? What are those issues sounds like the handoff of the driver is an issue still mound propagation is an issue maybe that one's already fixed or about to be fixed the path and then supporting cubelet in a container that was identified as an issue. Even when we started this session, the safe storage looks like it's still an issue.

B

Are there others.

B

We don't know, I would put him as it is. I also.

D

Like you know, also the delete happiness, or maybe he gives you the volume source instead of ammount dear I, that way, if it doesn't, if you don't have to do some hacking to get money right, but.

B

It sounds like a great thing to come up on in 1.9.

A

But that issue came up in San Francisco and what I understood it? What I took away from that brief touching upon that topic was that that's actually support it's not asking support it like you can count on that path. You know gonna format with data and then.

B

That's fine I, whoever stood.

A

There correctly, but that's shock the hell out so may w hack. Look it's like a backdoor bit.

B

Yeah and, for example, supporting some things in downstream API would make it if the volume plug-in Durer or even the cubelet baster was supported in the downward api. That means that you could start up the pod and not have a guess where to drop the flex volume and where to mountings. I I really would like us to think about how to make the fixes in kubernetes such that this scenario just works, like literally just works and works. Well, every single every single time.

C

That you know this problem that problem so they're trying out immediately hitting hitting issues. Those are the early adopters yeah.

C

B

I think we basically took a step backwards with everybody in awe Tomic, core OS and rancher. We took a step backwards, so we now have an it different ask model for them to make changes to support rook that.

A

B

Case before this, so we basically you know- I mean I, think we're heading over all we're heading in the right direction that there are corner cases that we need to go fix, but that's do all work.

B

We should we should have a good understanding if it's something that we need to fix now or document.

D

B

C

Look at it as a core.

B

Os issue, or you can look at it as a kubernetes issue, I'd like to think of it, as how can we make changes in kubernetes such that it works more ubiquitously.

D

Yes, basically yeah.

B

D

B

What my concern is we started with 1.8 as possibly the the solution for our flex and deployments of pro/con everywhere now, I'm, not sure we even on a path for 1/9 to be to solve this problem, I'd like to get to a point where we get to rook running everywhere. Kubernetes runs in some version of kubernetes. That's not in the distant future. Is that 1:9? If.

B

So then we need to get on fixing these issues. We need to work with 6 storage. There should be features and pull requests that follow to fix these issues right now. There is some issues that are in the Rope repo and not that's the extent of it.

D

I'll collect those issues, regional information, one where we dropping with dropping there. There will be any measure in watering container. You know somebody in the neck, meaning with a stick: storage.

B

Okay, and as far as dates and deadlines, how, where are we with the process? We should at a minimum, have features open or bugs open in the one-line repo, or else it's going to be too late.

B

That's right so.

A

It's something you know, I've been something I've seen work best you want to get into coming is to not just come with a complaint, but the proposal. That's.

B

A

A

Come with a full PR, but in having a proposal, this is a way to do. This seems to be very well received in that community. In other.

C

Words: here's what I want to do about it. Yeah yep.

B

Alright, but let's I think we should we should try to do that, maybe for the next six torch meeting, even for the ones that we know you know or know a little but and is that that's Thursday.

A

B

Okay, so I mean Jared and Steve. Can you guys yeah.

D

B

D

B

Then the issue of local storage and volume V's so Travis the whole local storage support is in external. It's an in sending external storage pod repo right.

E

B

Future, so it's not in kubernetes core.

B

So could we not make changes there? Even if we were to just include those packet, we already take a dependency.

E

B

I'm I'm there's something I don't understand. So it external storage is.

B

Independent repo I would assume that that feature would work for 1.8 1.7 as well. It's all the logic is going inside that repo as an external provision.

E

B

So could, if we were to just like we have our external provision or today, that we're using for for Brooke. If we were to use local storage, the data, you know an updated version of the local storage. Would we not be able to fix some of these issues or or use them to support our scenarios.

E

B

Along with I, mean.

C

E

B

Question is if there are a couple of things that we can do to enable the use of local storage to remove data hoster. You know which is not enabled by default and that's the my biggest concern is that people start using rooked and it loses a state and if we can help drive that forward, for you know in a quicker time frame and have it apply to older versions of kubernetes, that would be.

B

And then the other thing is I. Also don't see how like, if we could just wait for these features to show up in 1/9, and then we assume that they're gonna work for us I feel like they're, not.

B

So how do we get involved so that we can get? You know normalize on our pods, OS, DS and others just use persistent volumes for storage, not have to make any assumptions about the host directory or really anything. Just we use a persistent volume and as it presses the volume, if you wanted local storage, then you just use the kind of a persistent volume that you know says. Little storage.

E

B

And the cool thing about it is if it's true, if that, if the hypothesis is true, you know you make a couple of fixes in the external storage repo and they apply to all the versions kubernetes, then you could actually, in the 0.6 time frame a new and updated external storage commit hash and you have the features.

E

B

Right I mean we, we take a dependent argue idea. Mo takes a dependency on external storage.

B

What I'm trying to say is that the logic to do local storage does not depend on the version of kubernetes.

C

I'd actually already be implemented in the theorem one night, repo or an extra in the external auditory pop n 0.9, and maybe we could take them today and and all of a sudden get functionality all the way back to zero points. X.

C

B

That's exactly right and if we don't get involved with these changes, I feel like they're not going to be adequate for our snacks. So.

B

So that that's another bucket of things that and we could decide that you know if, if it doesn't fit in the 0.6 time frame, which we should throw it out, but it still seems like we should somehow get involved early with this feature, because we are going to be using that going forward.

B

B

Those were on top of my list. What's what else do we need to look at that so far, 0 6.

B

It seems like the biggest issue right now. Is it really doesn't feel like master is more stable than the 0 5 master right now to me on all platforms, is that a my my, the only one that feels that way right.

B

Yeah I mean it's hard to tell, because the failure you know shows that can't mount the persistent volume or cannon can't bind it and it's usually the root cause. Is that but it's a really common thing. It's like right now, people show up and say, is it's not working I guess even Jared asked you know is this? Are you on core OS or something, and that typically answers yes, but I can't I'm, judging mostly by slacking people on slack, saying that they can't get it working, but.

F

If you have to flex that it will work, I'm personally running it right now, okay, if the flag sets and yeah overall I'm happy with it, are.

B

You running an arc or less yeah. I said later.

F

Deflect to the tectonic installer so but I.

B

Did yeah unplug endure? Yes, okay, so so maybe maybe it's basically. We have to very clearly in a documentation, say hey if you're running in core OS, Rancher or atomic rook will not work until you change the volume plug under I.

F

Mean that's one thing: I'm! Definitely thinking that it's covered rook and the first thing you know I go through the docks and if we put like a big red mark there and say hey if you're on continued Linux, you need to do something extra I think we should be good off, because if a user Danny creates the ticket we can be like hey. We have a docs for that. Have you read the dogs? Please you have to oxide look at it. It.

B

Does it does still seem like a step backwards because I think in 0.5 things roughly just worked on core.

B

C

Set of issues, of course, but it's.

B

Better, but it doesn't this one feels like a step backwards from I'm a 0.5 release.

C

Even if it's someone I believe that just sort of powers through the surgery, it makes it go away.

C

This is how this this.

D

Is because of our Korres is: is configuring tuberculin, so they are? Naturally, there knock off your cubic to work with fake volume et al Sony access man. This is wouldn't.

C

Be my preferred at, but what we check to see if it's core OS and if its core was then we do the right, then then we make it work. It's.

F

Maybe impossible, are you Kurtz at least the docks of Korres, say something about mounting your oval AFS over the user partition over the user? Lip partition our directory, and if we could do something like this from the agent container, we could simply mount an overlay overlay use on the exact directory and then simply put our binary on it and.

B

We have at the Hewlett now is that how the key will find it? We would still need to find the Flex driver, yeah.

F

But it's the default directory, so you wouldn't have the need to set change the directory set the flag and we just create an overlay on the normally read, only use olympic sec.

D

F

D

F

I try to search the documentation for this I.

F

Think it's depending on which you saw on the system.

B

F

B

Let's look at and see if there are ways to fix this I yeah I mean if we have to document it and move on move forward. Let's do that, but it does. It bothers me a little bit that it's a step backwards for some users and and they were seeing it I mean I- talked about slack channels full of it.

C

C

We're recaps, but this might be a time when using one you know makes, makes the product much more usable for all the users. So probably this is crazy, but if it's the waited for OS or container linux is deploying.

C

B

Yes, that is also an option, so I think there is only an issue, but we can we can sir I mean I'm sure it's a one-liner change in some system be unit for the cubelets and chorus.

C

B

I mean honestly in general, I feel like the kubernetes core kind of dozen assumes that it's cubelet runs on hosts in most scenarios, and you know using using user live except it's a great example.

A

B

But so maybe maybe there's a different default or maybe we can do to defaults as well, so just as I made so that it looks like that.

C

B

Right yeah again, our goal is to get to work runs everywhere. Kubernetes runs with no changes. That's our goal.

D

Mount overlay that also because one thing that cube leader you have to do is you have to bind mount they plug-in directory, so I, don't think they're by mountings use, slash users lovely his thing.

F

F

Don't know right now, I, don't know I'm not funny, finding the docs right now, but I'm still searching.

B

Line on this issue do.

A

We have that specific issue opened up the net already, you know the ability to deploy reflects driver on across all environments. I, don't think so.

A

B

This will be in the bucket of things that you know around flex and they are agent.

F

Do you mean an issue on, for example, a call us or for the Flex deal.

B

More and more of a general around okay, yeah counselor.

F

Cool cool con, I hope I got this right. Yeah already opened an issue: it's where's the chip, that's chorus as the 2201 up all flakes, william blog and dearest read-only yeah.

B

So that's encourage the holy will be a similar one in atomic and a solar one and wherever yeah I really. I my gut says that kubernetes, one nine should look and etsy kubernetes plugins as well as VAR live exec, but that should be an alternative path just to support scenarios where cubelet runs in green I.

B

Don't know why they wouldn't take that pull request. It seems like it's very scoped, it'll help. Everyone certainly help us.

D

So that helps core OS, because Glory's already by mounting that it's equally nice directory yeah.

B

It's worse, plugins! No, sorry! It's you know: everybody's gonna put stuff there.

D

B

Make an alternative path for flex volume: I think it helps the whole community. That's why it's you know.

B

The whole thing.

A

C

E

C

F

Quickly, I found a link to it; they use it too, in this case, mounts kernel modules. On top of the lip modules directory. Okay, maybe we can work something out to do something like that.

C

B

That would show up where that we would have to put that in use.

F

The lip excerpt kubernetes and it.

B

Still, it still requires a modification of cubelet wrapper.

F

Depends on what what you know what permissions had cubelet runs with by default. I think it has a lot of permissions because it can mount stuff, so yeah should kind of have everything at least on chorus. I can only speak for covers yeah.

C

B

Had another question: maybe on a related thing, can the can the rook agent pod somehow restart itself.

C

B

Could it, for example, add another mount to a binder to the pots back and restart itself.

D

Sure so you see the box back to.

B

Add another mount directory and then restart itself. I.

D

Don't know that part of the spec and be upgradeable, because not everything on the pasta can be incredible after I am oh, so that's the density.

A

Right there, every.

D

How would you like you, can just kill the pot, no.

A

But if your updated apartment will it's part of a bigger Tanen set right right.

F

Yeah, this environment.

A

You can are doing something yeah.

F

A

F

I think we are limited by you can't update the volumes part in the pot and also. Secondly, it's from Adam said yeah. If the colour pot would for some reason, get deleted, you would lose your changes. Yeah yeah, so.

B

One of one of the things that we should capture in issues and imagine if you had a heterogeneous cluster, let's say I'm going to throw this out: half the nodes are running atomic half the nodes are running rancher and the other half are running rancher I made this up and they have different volume, plugin ders on some nodes than others. How do we handle those cases? I I right now, we're assuming a completely homogeneous cluster like everything about every node is identical and not.

D

Running that's.

B

Right, not we should figure I, don't think this is like a top priority, but I thought another one of those cases. That's you know. Rook runs on every kubernetes installation without changes. It's in that bucket.

B

So 40.6 it sounds like these issues are on reflects. We need some either documentation or some. We need to make a call on whether we should move forward with tagging. 0.6, local storage, I. Think Travis. You should figure out what the path for it is, but it doesn't look to me like that's a 0.6 thing. It's not fair.

E

B

And if so, then, we should clearly document the data host path that should be enabled, and unfortunately it also means that somebody has to you know, set it up correctly right now, it being empty is not a good case. That means right. Brooke is not durable by default. Right now,.

A

So remind me to open begin I know.

C

A

Some deficiencies here but remember we're open again. Why, if we, if we put that to false like Marlin Brooke, could we break and we'd.

E

Worse off for better off I was gonna suggest that one if we.

A

Didn't change default.

B

A

B

C

The cleanup documentation.

E

A

That is that the main concern is cluster tear down.

B

And that's exactly why using assistive volumes helps because, if we're building on top persistent volumes, there's an explicit action to delete them race.

A

B

Nothing's left behind.

A

Or soft state, if we just put it involved in the cluster yeah well, probably.

B

Not actually, probably in a better state.

A

Maybe just do that.

A

F

Maybe a small thing because it'll Oracle Sturridge from my side, it should definitely be able that we can move from the host year path right now to the local storage. Just a migration should be possible, it should be possible, it should be possible or are we saying yeah 0.6 is yeah, it works, but 0.7 is local storage and we break some stuff well,.

B

That's what I'm, trying to figure out I'd like to say that we, if we can make the changes I, ideally a for 0.6. We are local storage, we're only no more dirt, nor no more any host deaths. We don't have any dependency on those paths and if we need to carry it forward, we would carry that forward. If we don't you got, then you know we call 0.6 to beta and then we have to migrate one to the other.

B

That's my concern.

B

Well, I, don't know I mean if what depends on the credit, how we define beta. If we said beta is, we will carry it forward and we decide to deprecate data her late, a host path going forward. I would assume that the assumption is that we move the data forward. I'll.

F

Simply provide away that the kind of is in the cluster C or D or somewhere saved in the status of it that says: yeah I'm, a cluster we've currently really dis Worsham like 0.5 right hands. If we see Oh, we now upgrade to 0.6, we have to to migration X.

F

You know, like normal database migration, we have somewhere a state like o0 point five. Now we go to 0 6 1. We need to do a decent mayor, migrations and then run on every node, some stuff or enter the containers and moves yeah and.

B

That's that's what I'm trying to get up it's the work to get to using local storage.

B

You know how much is what's the scope of that work versus potentially migration we're coming forward.

E

B

Depends on how you want to define the data criteria if the beta criteria is I, deploy, 0.6 and if you upgrade me to 0.7 able it works things just work.

E

B

So then, then you're saying you're carrying the host path indefinitely.

B

But so again you could you could decide to do it. You could say that if you were using host path, you'll continue to use host path and definitely.

F

Maybe, if it's possible, if it's really possible, maybe half a script or something get words, I, don't know, call a function in the in operator. That tends replaces one OSD you move to data, call an x one, move it or something like that: Janet dad, yeah the backwards computer compatibility go for me, but there was 0.6 releases like yeah.

F

Let's go I can finally set up my cluster with Frug and yeah put some production data on it, a bit more than chest width right now me redeploying it close every day's to get the latest version to test all. If there are some box or something so yeah.

B

That's that's why I think it'll be interesting to you know, get a little more detail on if we can make some scope changes to the external storage project to help us in this scenario, and we just move forward on local storage so that we avoid all this migration stuff. Yeah.

E

B

E

B

Maybe Travis you can take a deeper look and give us an opinion just on whether it's feasible or not.

D

C

Guy for a federal provision.

D

Or volume project they're very good at wasn't murder, you think just yeah.

B

Like I said, we are we you're right, we already take a dependency on it, so you know, even if we have to expert, add some experimental stuff.

B

F

Change, it's outside of the car here, but Dawson to users still need to change something to get a local storage running on the systems, or am I.

B

It's it's essentially a volume, that's managed by by kubernetes itself. In this case, local storage path, I mean we'll look. This is look I, don't expect it to work out of right away. I! Think that's why we also have to invest in in making it work.

B

Okay, we'll probably find issues.

F

Guess I might comes on with the local storage is a bit like yeah, it's a provision in some kind and in the end you also need to set a flag or something to.

F

A

The idea being, but it just takes an external thing and get all the benefits of it is awesome. But if there is some internal.

A

F

You mean like when you want to enable something on the API server, an experimental API, that you need to also go to the API switcher at the flag, yeah.

B

Okay, so let's let's look at this and see what our path forward is. Ideally, we wouldn't have to do a lot of migration and take advantage of the Alpha State.

B

Yeah I mean, alternatively, we we don't call zero point six alpha beta.

B

No, that's another option.

F

Then, depending on how we go with the migration and my credibility, yeah I think this moral for comes on. If we're like yeah, you can migrate. If we will work on that, you can migrate it. Even if you have to put in some manual work, then I think we can call it beta, but foofie I'm, not if you're Mike yeah. We.

B

We still have to write all the scripts are very concrete steps for how people do that so yep.

F

Yes, a-plus on contra and pro on yeah on both sides right.

B

All right anything else, anybody wants to bring up.

D

Some of it that we wish later fix later.

D

An issue that will be coming Corizon would it be enough to document for stupid safety if using Korres make sure you, you know, send the body plug in there and I said, prepare that we're gonna be working to get it done. I do that? Is that enough, or do you you know, I think what.

B

I took away is that there are issues that need to be all open with about enough detail and some offline work sounds like looking at an overlay and others to understand where we stand with the Flex volume and then the local storage issue, at least some investigative work to see. If which path we should take 40.6 both of those feed into you, know declaring 0.6 ready or how we move forward. So I don't know if we can make the call right now. It sounds like there's some homework to be done.

B

So can we can we at least maybe on slack amendment issues, follow up this week and then we have six storage meeting on Thursday I hope they get there. We know there was a bunch of asks and potential solutions.

F

When is the meeting they're sitting.

B

There, it's usually I, think Thursdays at 8:00 a.m. Pacific, hey.

F

Pst right yeah.

B

I think so sorry, man, yeah.

F

D

B

Okay, great, thank you. Thank you guys thank.

A