wasmCloud Community, 27 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: wasmCloud: CI/CD Process Update, k8s Operator + Lattice Controller Demo - 10/27/2021 Community Call

Description

wasmCloud is a platform for writing portable business logic that can run anywhere from the edge to the cloud, that boasts a secure-by-default, boilerplate-free developer experience with rapid feedback loop.

A

Welcome to azcloud wednesday for october the 27th as usual, we're gonna get started with the demo, brooks we'll turn it over to you. Excuse me.

A

So the cloud team is all together today or some of the watsonville team is all together today in washington dc working on road.

B

Map issues.

A

So we're just uh hanging out yeah, um I'm gonna share. My screen should be able to see a nice little uh nice little yaml here uh I'll go to the nicest demo or the the diagram first. Actually so it's kind of a half demo half just showing off what um some of the work that I've put in on our build and release pipelines. Lately. um What we previously had was like.

A

We had three different um three different github actions with various degrees of building artifacts for testing and for releasing- and you know we were um not only like copying and pasting things in places, but we were also you know doing it kind of inefficiently, so I've refactored the wasp cloud host action to use matrix matrices more efficiently. So we have our build and test matrix, which tests a variety of elixir and otp version combinations.

A

uh And then, uh after after that passes, you know our building test pass. um We can uh we're now using a matrix to build um our uh our release, tarballs for windows, mac and uh ubuntu, and for linux we're also releasing the docker image as a separate step, and then once all that is successful, we create a github release.

A

What you're looking at here is just the um it's the the manual release for this, so we can manually kick off building the application tar balls at any time that doesn't release a new version of the docker image. It's not going to create a new github release, or anything like that.

A

So um to show you what this looks like previously, we had a workflow to build a host core, a workflow to build watson cloud host and we had a manual build action that manual action for creating tar balls for testing we've kind of consolidated into excuse me into the watsonville host action here. If we take a look at that, you know you can see the job build strategy that we um we've got the. Let me do this real, quick, that'll that'll. Give you a better view.

A

um We have the the build strategy for different elixir and otp versions. That kind of just runs our.

A

Great nice, that just kind of runs through our build and test process. If we run down here a little bit, we have the release, docker action. That only runs when we run a tag. Push it's not when we run manually, but notably something that's added. Here is the uh platforms section here so before we were just building and releasing for a docker file for amd64, architectures or like x86, but with the apple silicone like m1 um chips,.

A

um The apple silicon chips and other various arm devices, we have the uh arm docker image now, so the the next uh release that we make should be more accessible to them and then, lastly, just to give you an idea, what this looks like to build the artifacts, the the matrix, is actually fairly simple.

A

A lot of this previously was just copy paste between the different targets, but we have uh our three different os's that we target with with pen versions on on github action, runners, so ubuntu windows and mac os, and then this actually just runs on the the build matrix or it runs on whatever operating system we are uh using. For that.

A

um Most of this is pretty generalized. You don't have to change much per operating system. There are a couple places that, like setting up erlang and elixir is different for uh ubuntu than it is for mac and that's usually just a limitation of whatever, um like this rlf setup beam, only works on ubuntu and on windows. It doesn't work on mac, so we use the brew version here for mac.

A

um But what this ends up making is a much more concise flow for for releasing our application and um really enables us to do things like releasing. Only host core should be fairly simple, to follow the same process that we're using here so yeah, that's pretty much what I uh had going on anybody had any questions or want to see a little bit more about this. Well um yeah brooks: do you have any? Do you have like the next steps or improvements kind of laid out anywhere for this?

A

You know in a repost somewhere I have a couple of different there's a couple different issues in the otp um repo, so one uh that the taylor um taylor put up is is building host core only for the for the docker image. I think we should release tar balls with just the host core too. It's gonna, be you know.

A

Without the ui um it'll be more like minimal, something you'd run on an edge device because you don't need the dashboard and all of its dependencies included there, um and then this npm dependencies for wazo cloud host is more of a um it's another issue for just kind of trimming down what we use in this phoenix dashboard, anyways.

A

Okay, great now, apologies to everyone. uh We had uh some pretty insanely hot chicken for lunch, so there's uh still some lingering after effects on our spicy eating contest. Here, um a couple of us here in the weeks. uh Well, thank you brooks very much. uh uh Jennifer you've got a great demo or you have a demo some questions you wanted to pull up and um take us through. You want to go and share your screen.

C

A

D

C

You guys should be able to see a couple of vs codes windows open as well as a terminal down. The bottom is that correct.

A

Yes, that's correct.

C

A

So, on the right hand, side here I have the.

C

A

A little small we so it's like we're like a screen for ants.

C

Yeah, I'm on dual monitors. Well, I'm on a large monitor, which is kind of why I wanted to have windows open. Is that okay? Do you want one minute switch over? I think you're gonna be sure again. uh Are you on a oh you're on linux, I'm on the next computer, but I wanted to use a large monitor because I wanted the window side by side.

A

If you could, maybe just like command plus or like control plus it just make the text a little bit bigger that would probably that would help them yeah. They should be able to stay side by side yeah after you change the font size is that cool, maybe just a little more we've got to project it on a wall.

C

A

Quoting quoting taylor you zoom in until it feels ridiculous and then go two more steps.

A

C

Lost the effect of the larger monitor now I'm having.

A

Really precise.

C

um So, on the right hand, side here I'll just blow this up now. Actually um this is a simple test that is written in. Go that pretty much does what the operator does, which is to load the applied yaml um in onto the wire. So we had an opinion of doing this through json and when looking at the uh the lattice controller, the ammo seems to be a lot better, so I just converted that and the ammo marshaled it um and that will go on the wire also.

C

I did choose zigzag case because that serializes a lot better than this guy here, because it I mean, I guess, the recursive design processor- doesn't actually work well for the data structure that we've chosen here have an ad server running in the background here: it's not using jet stream, so I will actually end up starting the test here to listen um so right now. This is how the link started. It goes through.

C

There gets and listens on the same topic that we're publishing to which is just uh well somewhere down here, but actually down here um has a small timeout, so I don't wonder what's going on, then I get that and I serialize it into a map because that's the way it gets read um and then that gets called into an aspect.

C

um This is pretty much copy paste code from what um kevin had before and I've kind of hacked it, because I don't necessarily understand all the parts and also the fact that I haven't completely figured out how to configure vs code to work nicely with elixir. Yet uh things are a bit rough on me, so bear with me there. So this is just running the reconciler working.

C

I've changed the weights here to pretty much use the same um that we had before or that actually triggers this and as normal, um so jesus 50, 50 and us less than east here and that maps down here to the zones.

C

um So what I will do is start the test.

D

Which will actually no and I'll? Probably.

C

um So I'm starting using just a tag, um so this is me learning index from the fly here actually, but that tag actually might as much as must exact here. So I'm going to do that. That will actually be listening. You'll see it actually listen there down the bottom.

B

C

Then I'll actually run the test here within 10 seconds and so yeah. The message actually came through and it got well. This is for, I guess, I'm eating just to dump it out as an inspect, which is just coming out of this part here. So it is getting. It is coming through the wire, it's listening on the right topic and then it reconciles so to speak.

C

um I so the questions are, I'm not necessarily sure how the apply event actually works, and I see that the reconciler actually has opinions on things like if there is not enough um enough actors or there aren't enough factors on a certain zone, black key, throw up, for example- and I don't I guess I just don't know how the whole flow is supposed to work. So, for example, if I change this to 80.

C

Probably nothing with this one here, so let me put it into this and again and run it through, and this test will actually fail um with the error based on the records at it.

C

Yeah, so this is well, I thought it actually fell, but I guess in this case it doesn't actually fail because it did properly um yeah, that's kind of the extent of what I've actually looked up. I haven't hooked it up all the way to otp um and I'm also not sending messages back yet, and I really wondering whether that's kind of the intention of what we want in terms of the flow of what that, what needs to happen between the operator and the lattice controller and then going from there into.

C

I guess I haven't looked at the observability code yet and I guess that needs to eventually talk out to otp.

C

So yeah I've put some pieces together. I don't necessarily know that they're in the right order. It just uses the fact that it looks like this is the layout that is kind of going to work.

C

But that's just me guessing at what is what I'm trying to make sense of. I suppose.

B

So can you can you hear me through the mic? Okay yeah, I can hear you okay, um this is a great demo. I think um part of the the stuff that might be confusing right now is the reconciler just emits recommendations to be taken and there's no code written yet that actually takes those actions. So the reconciliation is designed as a pure functional model where it looks at the state of the lattice and the uh application plan or the application spec, and then gives you a list of things of gaps that need to be closed.

B

So it might tell you that you need to start one more actor or terminate an actor or you know, do something like put a link definition or something else, but there's no code written yet that actually acts on the recommendations that the reconciler makes.

B

The other thing is that we've moved the lattice observation code, so the observed directory in the lattice controller. That's actually now part of a repository called lattice observer, and so the goal is to make it so that the observation of a lattice is a reusable elixir library that can be used by both the lattice controller and the dashboard for the weizencloud otp runtime.

B

So what you've got here is a pretty close to like a hard-coded version of what we're actually going to end up build building. So this is great, especially, you know like the the piece that we haven't coded yet is how do you get the app spec yaml file into the lattice controller? And so that's that looks like that's what you um kind of hard coded in your your test there, and so we'll. We'll eventually have all that stuff fleshed out, but yeah. My apologies for uh us not having the code there. It's just you know.

B

We haven't. We've been uh working on other priorities, so the latter's controller stuff hasn't been moved as forward as I'd like it to be.

D

Yeah, that's fine.

B

D

Impressive, given.

B

How little of it we've actually coded.

C

Yeah, thank you, um so is the reconciler. So I guess how does apply even actually come into play here is the intent that you get the recommendation from the reconciler and then you call apply event to so apply event produces a new.

B

Produces a new version of the state which is the uh observed lattice and so the workflow once we code it is we receive events from the wasmbus dot, ebt dot, whatever uh that's topic and apply all of those as they come in to the observed, lattice state and then at some time dinner will we will then ask the reconciler to look at the uh observed state of the lattice with regard to a given app spec, and the reconciler then gives you a list of actions that need to take place in order to bring the state of the lattice to a state to its desired state versus its current state.

B

So uh if the reconciler has no recommendations, then you can assume that the lattice has or is in your desired state in order to satisfy that particular aspect.

B

So what's not coded yet is the control loop built into the lattice controller that applies. Events uh puts the state through the reconciler and then converts the recon reconciler recommendations into commands issued to the lattice control, interface and and so on. So uh I'm I'm hoping that we'll have some more uh some more scaffolding in place, so that you'll have some more to work on.

C

So then, how do we so I guess the with the operator with the kubernetes operator. If you're I mean it kind of sound. Well I mean the controller body right so like the operator itself is designed as a controller um kubernetes operator. Sorry, so then there are kind of two things running controllers running. At the same time, you.

A

B

C

Grop that all the way, how that's supposed to work.

B

Yeah, so most of the time when people build a kubernetes operator, it's that operator that does all the work.

B

um In this specific scenario, what we wanted it was to have the ability to monitor, deployed applications in a lattice and manage those either from a kubernetes operator or without a kubernetes operator, and so what the situation we end up in is the kubernetes operator is really more like a proxy for the lattice controller, and so when you push the yaml app spec to the kubernetes operator, it should then just push that on whatever api endpoint, the lattice controller has exposed, and then the lattice controller is then responsible for, uh for you know, running the control loop on it.

B

So the control of the kubernetes operator uh should, at a high level, be pushing the app spec to the lattice controller and then receiving uh second order or higher level. Events from the lattice controller.

C

So I guess then, for example, we had a situation where the lattice controller makes recommendations which are realistically different from what the aspect asks for, like the error that I was actually trying to trigger is when you don't have enough actors in one zone, and I guess I did not figure it- probably because I didn't drop the numbers properly. But when that happens, what is the status that we want to return back to kubernetes, because now the abstech is going to be different from?

C

What's actually on, I guess, what's declared is going to be different from what is reality? How do we want to respect that back.

B

Yeah, so I'm not really sure um of all the details, but generally what we're going to want to be able to do so.

B

The the lattice controller will also allow you to query the current state of the observed lattice, and so the lattice controller is responsible for mutating the state of the lattice, so that it makes the app spec uh happen, and so the kubernetes operator is basically just responsible for getting that aspect into the hands of the lattice controller and then sitting back and letting the lattice controller do all of the work and um the kubernetes operator. Then reports on the state changes.

B

So if there's a gap between the current state of the lattice and the desired state, then you should be able to you'll be able to query that through the lattice controller's api and you'll be able to get that information uh you'll be able to bubble that information out through the the kubernetes controller, but that api doesn't yet exist. So um there's like I said, that's there's not really that much! You have to go on, so I'm I'm kind of impressed that you've got a demo. That goes this far all right. Thank you, yeah.

C

C

So I guess what can I try next suppose, because I think I have more information. I think I would like to hook it up all the way to otp and see what actually happens, but that would just be me like grappling through a bunch of stuff. I.

B

Would probably not spend effort hooking things up to otp because, like I said we have, we have work happening in the lattice observer, repo, that is building an otp supervision hierarchy for observation and we're then, going to once that's up and running we're going to go back into the lattice controller and build another otp supervision hierarchy that incorporates both of those. So I don't well, it feels bad to tell you to not do anything. um I think at least for the the short term. It's probably best to uh to wait a bit because.

B

Okay, you have to you know, rig everything up to otp on your own and then uh you suddenly see a pull request come through that uh redid. All of the work that you did right.

C

Yeah, I understood.

A

Janitor, this is awesome, work um great job on this kevin. How do we coordinate um you know the next steps here between the lattice controller and this operator.

B

uh Basically, like I said, the suggestion is gonna have to be to uh for for people to hold off on contributions to the lattice controller and the lattice observer, at least for a very short period of time, until we can get the changes that we're making to them pushed out. And then, after that point when, when all the scaffolding is in place, we should have, we should be able to take contributions. But right now it's just kind of it's too much in limbo, and we just haven't had the time to move it forward.

B

The way I wanted to.

C

Okay, janitha is this blocking you for something, or did you just think of it as an interesting project, just something interesting to do? While I have some spare time and the spare time might be running short pretty soon, but um for me it was just well, I mean, since I kind of kicked off the controller or set up the direction. At least I wanted to see how far I can push it um to see the whole thing working together, but I don't have a business use case, at least at this time.

C

A

Super cool all right. Well, um we are doing uh you know roadmap planning now this week, so let us get through a little bit of a prioritization on our side, but I definitely feel that you know we're kind of you're kind of blocked on this at the moment. So that feels like this kind of escalates the priority a little bit, but let's see what we can do to try to get you moving again.

A

So this is great work, and I think that you know my personal belief is- is that this is pretty important to go and start where users are today in the kubernetes ecosystem, so that we can help be the bridge to the future. So this is to me. I think this is pretty valuable um effort here.

A

um What else open calls here? I don't think we have anything we want to present on zendesk, steve or anything roadmap wise from planning. uh Just given what we're doing this week is just you know, trying to plan out the next couple months of work here. Yeah we'll definitely make some uh posts on the slack next week. Okay, all right! So we'll we'll take the output of this week and we'll do some posts next week and pick up and go from there. Is there anything else that anyone would like to demo today.

A

Okay, I think we'll go ahead and stop here today. I really appreciate everybody's time and um uh jordan, a huge call out and thanks to you, the um you know, the watson club wednesday videos look great. I love all the new headers and stuff you put in there and it sure is great to get that stuff out and I noticed you've even started tweeting um out. You know contents of the meetings and calls to action, and things like that.

A

So thank you very much and I think we shared on on slack the um you know the output of the survey around what languages to focus on next. I think we do have uh some business use cases coming up around c plus plus, so uh the business users may kind of push um that to the top of this uh top of the pile if somebody is uh blocked on going into production there, but we'll give some feedback out over the next few uh days as we get the roadmap together here.

A

Okay, thank you very much everyone. I hope everyone has a great day bye.

A