Kubernetes SIG Apps, 24 Jun 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Apps 20190624

Description

Led by Janet, Co-hosts(s): Adnan

Announcements:
* Kubernetes 1.15 is out. Supported versions of Kubernetes are 1.13 - 1.15
* New Subproject: https://github.com/kubernetes-sigs/execution-hook

Discussion Topic:
* [Stateful Application Data Management API KEP] https://github.com/kubernetes/enhancements/pull/1051 (@liyinan926)
* Sidecars KEP implementable https://github.com/kubernetes/enhancements/pull/1109

A

Okay, now it's recorded so welcome everyone to June 24th, see cast meeting. My name is Jenna Kuo I'm going to be hosting us today. We also have, and today on the call, who's co-hosting, so I'll share the agenda in the chat so that we can follow along.

A

Today we have two announcements: the first one is that kubernetes 115 is out, so that means supported versions of kubernetes and now 113 to 115, and then we have the new sub projects under C caps.

A

The sub-project is the execution hook and we have Iman today to talk to us more about what it is and let me any.

B

All right so well, so, basically, this repo is for a sub project for running arbitrary hooks in your containers like, for example, you can actually use this for running quiescing on Christine cooks, for you know supporting snapshots. So basically you take. You run the freeze commands doctor freeze application before you take a snapshot, and then you run the UM freeze commence, who owned? Freeze an application post? You know the snapshot, so this is the primary go off this sub project, but you can also use this to run outer.

B

You know under kinds of arbitrary comments if you want so the designs, basically I like having some CRTs for modeling execution cloak and also hook, actions, that's or or the commands you want to run, then the controller will be responsible for running the hooks for you as service the status of the hook execution through. You know the status section of the xq2 hook, CR so yeah. Let us know if you have any question regarding this new sub project and new ripple yeah.

A

So does anyone have any questions about it.

A

If not I'm going to start our discussion topic, the first one is the stateful application data management. Api cap also proposed by Iman.

A

Let me hand it off to me. None.

B

Okay, so um Jenny, maybe you can also on opened the cap.

A

B

Yeah I can also do.

B

Okay, can everyone see my screen.

B

Okay cool, so basically, this cap is for proposing a new API. We call stateful application, so the idea is actually to use this new API to support data management semantics for stay for application, so basic application. That meant in you know some persistent states through persistent volumes. So the idea is that we can actually use this API to support snapshot, backup cloning, restoration and rollback operations declaratively well, mostly. Actually so this API structure, let me see if I, can naturally pull this out.

B

Oops. Okay, it's not for two files in this I'm gonna, try to talk to them to the structure diagram, see if I cannot show that okay cool. So this is this is the basic API structure. So it's extra centered around something cost.

B

If application, which is sort of like an extension to the application CRD that you can not only you- can actually describe because your metadata, but also you can actually use this as a bridge to do you know, data management tasks, and so we have C RDS, called application, snapshot and application bag, backup for supporting a snapshot and backup operations and there's also a dedicated Ciardi called a picture rollback for supporting in place rollback of an application. So the the dotted lines here means basically like the direction that you can actually take.

B

But, for example, you can actually take a snapshot of an application or you can also to a import of a backup into a snapshot on the other direction. You can do applicator restoration by creating a stateful application that has a source that points to a snapshot, for example, or you can also restore an application from a backup directly by pointing your state of application to a backup resource and rollback works.

B

Similarly, so you can do a rollback to us to a state capturing a snapshot for application or you can actually restore to the state capturing a backup. So we try to make this API kind of like having a symmetric structure so that you can do different kinds of operations depending on the source of your operation, and so any questions so far.

B

Okay, so we have a lot of details, inner inner caps. So if you you know, if you're interested definitely check out the cap and see all the details about this new API, so I won't jump. You know into the details of, however, every C or D, but yeah. Definitely, if you're interested, take a look. The the cap so I guess now, I'm gonna to we're gonna show a demo of this nuke some implantation of this new API.

B

So so the example I'm gonna use is a MongoDB. So basically assuming like a new, um a MongoDB application running somewhere, you want to take a snapshot of this MongoDB application. So me show you first I'm, gonna use some gke for this example and also I already have two GCE CSI driver, pcpd's, etc. Installs I'm, gonna use a GC EPDs that persistent storage for so or to have an application running here. So the is already running on its run by a stateful set was two replicas.

B

So the first thing I'm gonna show is to take a snapshot of this MongoDB instance. So.

B

This is how the application snapshot CR looks like so you can see it has a reference to the class or application snapshot class. So an application snapshot class basically defines common parameters used about a snapshot or for a cause of applications, and it also defines who's the snapshot or for this particular application. So only a snapshot or with a matching name can really handle this application snapshot request so basic outer, you know, controllers or snapchatters.

B

They will basic ignore this if they are not um the one that should be responsible for this, so we also have a stateful application CR here, which represents the mungo stateful application. So you can see it's actually pretty simple, so it basically has a a reference to an application CR which describes the application metadata about this MongoDB instance through this application name field and another interesting thing here is the sourcing. The application snapshot object. So you can see this the type of the source in the application.

B

That means I want to take a snapshot of an application, and here the application name actually points to the name of the state for application, which you know represents the stateful application that I wrote. I want to take a snapshot for so let me show you the class, so the class basically defines are fusing like what I said. Like you know, first of all, it defines who's the snapshot or for this particular class of application, and also it defines the quiescent um quiescent hooks here.

B

You can see the pre hooked or quiescent hooked, em finishes and also post or unclear seing folk definitions and there's also something called elation policy which it defines what happens when you delete an application start out? What are you want to keep the snapshot Iran and or you also want underline, snapshot to be removed as well as well. So any questions so far.

B

Okay, so here the for the protocol consists basically running a Mongol, lock command and for the um um quiescent codes, just you know basically unlock in the table a lock in the database. So the first thing we're gonna do is actually to create this.

B

So this will trigger the snapshot process. So if I'm I do.

B

So you can see here we have a bunch of events that shows what happened you know during the process. So, first of all, we retrieved and persisted the application metadata, which means like basically, the set of coronated resources that make up the application. In this case we have a MongoDB stay for set. We also have got a service. That's gonna capture, there's also a application CR that also captures as part of this snapshot, and then, after that we run the pre-hook.

B

So you can see we actually execute it, one pre-hook and then after that, so once the application is freezed, then we start the wall and stand trial process. So the one impressed. So here we got two replicas, so we actually have two PVCs to a snapshot. So that's why here you know we basically created two wall and snapshots for the two PVCs and then we have something called application snapshot content, which is a non namespace resource that basically capture information about the underlying snapshot, so yeah. So once both one and snapshots become ready.

B

That means they are, you know, they're done, then we run the post hook to unfreeze. The application- and then we call the po process done so you can see notice that the status of this that traffic becomes a pond the means it's usable, so you can actually use that to do application, cloning or restauration those kind of things.

B

Okay, any questions so far.

A

B

It's actually created by the user for an existing application, so I can show this I guess.

B

So this is the original definition of the application, so you can see that it has a got a application CR, which also specify the component kinds of this application. They got one service and once they Fassett, which is you know basically, so we use this label on the sap label, to you know, group all the components together so now what I'm gonna show is, after to to use this snapshot to do a restoration or clump whatever you want to call it. So this is what I'm gonna do as before.

B

Actually so, I do that I want to clean up their existing app so because otherwise, I might not even have the enough resources to run new callin apps. um What I'm gonna do delete this existing.

B

Instance, and after that, I'm gonna do a restoration of app into the names same namespace before I. Do that this is how to with see ours for the cloning operation look like, so what I'm doing here is basically having another stateful application, but this time you can see extra has a source field which points the snapshot I just created and there's also something we call ladder substitution rules which you allow you to specify things that you want to replace.

B

For example, you want to replace certain data, seeing the labels or environment variables wearing annotations or even names in this case, I just want to change one of the environment variable from one battle to the other because of the change of the after name so I'm, gonna name, the new application or the clone application clung to you know just make it different from the orange and application in terms of like no naming so because of this.

B

We also need to change some outer things like, for example, this label actually points to a basic specify: a label selector, which we also have to rename so.

B

This basically creates the State for application object and then triggers a restoration operation. So this will go back to so. You can see this new clone application was created because of this you know restoration operation and if we actually jump into the details of that, you can see all the labels are also renamed. So the old labels used and now the new labels all become Kong, and if we further go down into an integrative pod, you can also see that we actually also renamed like the wonnum yeah.

B

So you can see we also a curate renamed, the PVC that that's you know attached to the pod as part of this resolution process because of the name, the name change of the application yep. So now the new mount MongoDB instances have been running so any questions so far.

C

Hey yeah I have one question: where is the natural content actually stored, like the the.

B

Actual one thing, I think I've forgotten to Roy emphasized- is that here by a snapshot we mean a snapshot that has everything basically has your so has a basic a copy of each of the component or accumulated resources that make up your application and also it includes snapshots of all the persistent volumes of the application, as you know as a whole.

B

So then, by restoration we also restoring those application objects, but also all the you know were Quantum's. So basically, here a one interesting scene to show us is this.

B

So if you see this PVC, this is basically the PVC used by the first part of the cloned MongoDB instance.

B

So you can see this PVC ACTRA has the data source which points to have on in snapshot that was created as part of the application snapshot? I just showed, so this is basically the way we use for doing one restoration.

B

I think so otherwise I guess people will be.

B

B

Sorry I I didn't turn on my speakers. Oh I didn't hear what you just said. Oh so.

C

He might so. My question was where this snapshot content was actually stored, so you kind of shown how to the volume snapshots are stored, but what about the actual resources? Ok,.

B

So actually, by default we just persist. The application can fake information onto a persistent one and attach to the controller. So basically by default, you know just uh persisted the application config to disk I, see ok yeah, but API also supports outer kind of storage options that people connector just extend.

C

B

Okay, any other questions.

B

All right, I think that's about it. So yeah like what I said if you're interested in this proposal- and you know definitely check out the cap and comment on that we'd love to hear you know feedbacks on this cap from the community yeah thanks.

A

All right, thank you and go back to present the next one and our agenda is the sidecar cap do do we have anyone on the call to talk about it, yep.

D

Yeah so yeah, this PR is just to change the cap to be implementable. It's we've filled our rolled up sections now and decided on pretty much everything, so it was mainly just does anyone think there's anything missing.

D

I will also be going to sig note tomorrow evening to ask them if they have any final questions, remarks, etc.

A

Okay sounds good.

A

So that's r22 caps to be reviewed. So if anyone has are interested just take a look at the link, we post it in the agenda so now we've gone through all out of the items in our agenda. Is there anything that anyone wants to bring up for open discussion?

A

A

Okay, so um think we can end today's meeting earlier and thanks everyone for coming see you next week.

C

Thanks everyone.