solo.io Meetups, 12 Dec 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Online Meetup #4 - Dec 12, 2019

Description

In the last meetup of 2019, we have two talks from the team on recent announcements and new projects.

Autopilot -- an SDK to build opinionated operators for service mes
https://docs.solo.io/autopilot/latest

WebAssembly Hub -- a hub to build, publish, discover and deploy customizations and extensions for Envoy Proxy and Envoy based tools like Gloo and Istio
https://webassemblyhub.io/

Join the community https://slack.solo.io
Slack channels #web-assembly-hub and #autopilot are available for questions and discussion. DM to request a future meetup topic

See you in 2020 - have a happy new year!

A

All right, it looks like we're going live.

A

And we are live great all right, so thanks everybody for joining today. This is our last meet up of the year. We're very excited: we've actually been quite busy these last couple of months from coop con until now, until this week.

A

So about that feedback, but this is our last meet of the year since coop con we've actually been shipping. A lot of announcement, we've been launching a lot of new things. We've had a bunch of new projects as well, so we want to take the time to actually get a meet-up in at the end of the year and also spend a little more time going through these things and then have demos and opportunity for questions from from everybody online, so I'm Betty nice to meet you.

A

If you want to hop on the hop on video you're, more than welcome we've got Scott Weis one of our engineers, our principal engineer here at solo and then Christian posta, our field CTO online. We've got folks Twitter handles here, so that if you want to hit people up with questions later you're more than welcome to, but today I will be also watching the chat. So if you wanted to type a question in the chat, that's fine! If you want to just unmute and ask that's, also fine, we're we're gonna, keep this very casual.

A

So, let's go so today. We're gonna have two things. So we're gonna have autopilot, which is a new open-source project around building operators for service mesh.

A

So Scott's going to talk about that and then we we announce that the week of cute con and then we have web assembly hub, which is basically a new hub for building customizations custom extensions for envoy and envoy based products using web, and so there's a there's, a new hub for that that we announced yesterday and so Christians gonna go through show you the hub, show you the things you can do in a demo. So with that I'm going to pass the ball to these gentlemen, I think Scott. Can you just take over and share?

A

Could that shot.

B

Okay, mute can I share for sure.

C

Yes, should I take over with the screen share or go.

A

Right ahead, and then we can just go straight to your screen to talk about autopilot. Okay,.

C

Very cool Thank, You Betty.

B

Sure they're everything good.

C

Looking good okay, so thank you very much I'm going to show you guys a little bit about autopilot just to give some context. Solo IO has been in the envoy service myth space for basically since its inception, which is over two years, we started with kubernetes operators. From the outset. Everything has been done in an operator style, so our core product glue is driven by CR DS different events and kubernetes and has a declarative system for ensuring a desired State.

C

So if you're familiar with those concepts, you've already kind of got in the nitty-gritty of kubernetes a little bit like using CR DS you'll understand the approach from that point of view, and then that drives all of our automation and it hooks into our our get workflows. You know so.

C

We've been using git outs for a long time and basically, we've developed a few times from the past different frameworks for working with for building these operators, particularly, we have a lot of redundancy around service mesh, so configuring SDU virtual services, monitoring, outputs from Prometheus and metrics, and things like that are built into our existing operators and what we felt like doing is basically a next step for us is, with this framework, to take it to the open source and really try to get some community around it in a similar way that Red Hat's done with their operator framework and yes, I'm gonna go through autopilot itself, it's an SDK, it's a command line for building and deploying as a command line and a set of libraries SDK, essentially for building and deploying an operator, and particularly with attention to service mesh.

C

So if you think about a regular kubernetes operator has primitives for kubernetes built-in, but doing all the rest of the work connecting to metric stores, databases get outs, backends different web hooks cloud providers, that's all work, that's left to the user, implement manually using various libraries etc. So part of the idea was to take the operator concept that kubernetes has given us and extend it to the domain of in particular service mesh, because we were already finding such a strong investment in that space.

C

For us a strong need- and we we anticipate the market is going there as well. We've had a lot of feedback from customers and vendors.

C

Everybody in this ecosystem is really starting to build operators at this point, and the next stage is, is building operators for service mesh.

C

So let me show you what that actually means in the context of real technology, so that this is going to be done using go today and I'm going to show you how we can automate a simple service, mesh automation, use case using autopilot, so we're going to build a canary operator just an operator, that's going to do rolling deployments for us with the traffic splitting, if you guys are familiar with the canary use case, I'm, not going to think too deep into how it works.

C

This is more to see how you can auto automate your service mess with autopilot. So if I show you my my cluster right now, I have some GTE cluster running and one just.

B

D

Me, let me suppose from you, your slack.

E

B

I'm sorry dude: oh there we go.

C

So just clean up a little bit.

C

Okay, so we'll see we have sto installed is really what we want to look at here, so I have a basically all you need, for this is a cluster setup. Just with this deal installed, that's just installed using like the basic Castillo 135 that you can install on your own, and now we're going to do is, is basically build an operator from scratch using autopilot. So if you see I have an empty directory here, empty directory and the first thing I'm going to do is I'm gonna run a peon it.

C

An APN eight is going to create the initialization files that I need to build the rest of my project and we'll take a quick look at what that is. So, if I loaded my IDE here, we can see what that did. Was it created, just a go mod file, and these two config files and the main one I'm going to look at, is the autopilot that llamó- and this specifies this- this structure of the project so like I said before autopilot is very opinionated.

C

It makes the trade-off of making everything easy by being more opinionated about how it gets built. So it's got an opinionated structure about how the operator is structured.

C

Basically, the idea is that you have a top-level CRD and then you can define phases for the CRD and autopilot, treats the CRD like a state machine that allows us to be more opinionated about what your operator essentially does, and so in the vast majority of use cases it may be watching a large number of things, but in the end the output is going to be some kind of urban areas resource very often now. This is also something that's extendable in the system and I'll get into it.

C

But right now we can think of autopilot as a system that takes some inputs, depending on the given phase of your CR D and produces some outputs so from that, let's actually sorry one second I'm going to update this autopilot animal with what what we need for our canary essentially so what you saw was the default generated now. Here's one that actually describes the the state machine that our canary operator is going to execute, and so you see I've got a few phases here. There's an initializing, a waiting phase, evaluating promoting and rollback such a.

C

That means that in the initializing space, once we receive the canary CRD we're going to read some deployments from the cluster and then we're going to output some deployment services and virtual services first year, then we're going to have a waiting space where we basically pull and ensure that the or we watch we watch the deployments of virtual services. We make sure that if, if one of those things gets or the deployment gets changed, we initialize the evaluation process.

C

This is a little bit about implementation details, but essentially, where we're waiting for some user feedback at this point to kick off. So this might be the result of get workflow. It might push a change like update the image on my deployment and that's going to kick off the canary process that our operator is executed. So then it goes to evaluating. Then it's going to watch metrics, so I'll actually show you how that's working.

C

It's gonna watch metrics that come from the proxies through Prometheus and use that to evaluate the status of the Canaries is the the failure threshold. The percentage of requests that are succeeding. Is it above the threshold we've defined. If so, eventually we will promote, if not, we will roll back the deployment very standard canary workflow. The purpose of this is not really to teach about Canaries, but just to understand how we can automate this using autopilot. So I have the auto pilot llamo created.

C

Now, let's see the actual code, so I'm going to run the ap generate command. This is designed to be run and rerun as you iterate on your project and build it out so that you can. You can create the skeleton of the project without and only have to fill in the business logic yourself there like a service application. If you're familiar with that concept, so we'll see we got a bunch of stuff generated here in the interest of time. I don't want to dig too much, but essentially you'll see.

C

We have a docker file for our app, we have a main for it and then a bunch of deployment manifests that were generated as well and we can dig into those. But let's look at the main function.

C

This is a very simple binding and this file can be edited once it's been generated to add your own bindings in, but essentially it injects into the library method from autopilot called run. That's going to run the opinionated manager, the opinionated operator structure that we call a scheduler internally and what the scheduler does is. It calls the user-defined work functions and those work functions can be found in this workers and you'll see if you go. This is where you actually have to go and implement in your autopilot code.

C

So it's your responsibility only in autopilot to edit these worker files, so this worker, this worker etc, as well as the the spec. So this is where you actually define the spec for your. You know the desired state and then the status for your CRD. Everything else is generated for you.

C

If you're familiar with these concepts from other operator frameworks or you work with the go kubernetes before you'll- be familiar with, these structs you're only responsible for filling in the actual implementation of respect and the status, let's just see just for kicks what happens right now, so we can already build and deploy our operator. So just going to do that right now you can use the auto pilot command line to automate this stuff, but it's all docker and coops ETL commands. So you can run this locally as well, but it's nice to have this.

C

This can be made a part of a CI CD pipeline, so you can actually deploy and update your operator in production using gate ops, which is a really nice flow.

C

Okay. So right now it's deploying all my components and I should actually be able to see. Let's get our namespaces, we should see this canary operator. Namespace got created.

C

We'll see our operators running, she actually run ap logs to tail the logs from our operator and see. You know some some bootstrap logs that have came up, and this is all stuff out of the box right. I didn't I, didn't added a single thing on my project. This was all just setting up the autopilot yeah MO and running ap generate so, let's, let's go and for fun.

C

Let's actually create one of these CR decent cluster space, I'm going to create a CR D in the cluster and what you'll see is that my controller panicked panicked?

C

Would this implement any message so we're actually right up to the point: we're already ready to start implementing business logic, and it was this initializing worker that we hit, because that's the first phase, it's declared in the autopilot right here where we say initial phase true, so what I'm going to do in the interest of time is I'm going to just fill these in I have these already written, but this is an example of what it essentially a work function.

C

That's going to perform this the sync operation and synchronize our state, our our state, when it's in the initializing phase, I'm actually going to run a command here, because I've pre-written these and then we can just take a quick look at them.

B

You could be a second.

B

C

I'm just going to place this command here. This is gonna copy. Just so it doesn't look weird. This is going to copy some pre-written files, I have for each worker and regenerate the deep copy types. So we can see we filled in the spec here. Here's a little bit of context on how our Canaries going to work.

C

You know this is again. This is all user implementation details. This is up to you, which you want to define. You can use these status as a place to store state. So if you treat your system like a state machine and the workers are stateless, you can have them pass state information to each other by modifying and updating the status on the CR D.

C

So that will finish up and we can see we have a phase for each a worker for each phase and we should be able to build and deploy again. So will we build.

B

C

Now, if we watch our logs for our operator, we should be able to see. I was a canary going.

C

Okay, so are now our canary actually got processed and I didn't push yeah I didn't down. Show you a good example because it went straight to waiting, but essentially we can see here what got created from our canary. So you see it got put in the state waiting, observe generation. All of those boiler click the details will be handled for you by autopilot, so you can do things like ensure that the generation serve generation matches the real generation that your operator is in sync.

C

It's also really easy to run this locally from your machine, so I can literally just hit run here. Hopefully I. Do you like it, and I can actually run this from the IDE step through and depending on how your operator is set up. If all it's doing is you know, pushing and pulling yamo from kubernetes, then you can actually run this locally right. So we're able to step right into our location here we can maybe jump all the way into our worker. Let's say we have a waiting worker.

C

We can jump right to our work function here and step through and see how locally how our operator is going to behave. It's it's hugely helpful for us is so to be able to do this kind of workflow, so we put it in the waiting phase and just to watch it work. We can try to.

B

B

C

It's injected so it was not injected, so we have to label it for injection so that it's actually going to be in the mesh.

B

I'm selling it delayed everything.

B

C

We can just one of the things that you'll see is if you delete the route CRD, all the stuff that gets created from it gets automatically garbage collected.

C

Okay, so let's reapply that we should see.

B

C

Okay, it's back in the waiting State. We can now apply a change so I'm going to bump the image version on that deployment and if I show you the pods now we'll see that the.

C

Okay, this is embarrassing I'm. Sorry, this is this thumb example is still development right now, so I'm undergoing a little bit of a change. I'm, sorry I, don't have a better.

B

Working version and.

C

I'm sure that there's something I don't want to dig too deep right now in the press of this demo.

C

But it's it's a it's a problem with the example with autopilot, so I hope this has demonstrated essentially the simplicity of building operators with autopilot how it's it's designed to take away so much of that boilerplate and redundancy with between projects that are not only leveraging kubernetes but service measures well and simplify the process so that we can focus on the business business logic that lives inside the operator, rather than all the piping that goes around it.

C

B

Thank you so much I'm gonna pass it over back to the sky.

A

Because this is how can people we're due if people.

C

F

An operator and.

A

If they want to cook, you know, can they contribute about.

C

You want to show you the installation tax right here. It's all done with the CLI, so there's nothing that needs to run server-side. In order for this to work, we have a tutorial that takes you through the steps really to illustrate like a few of the different features in autopilot, as well as how to get started with your first project.

C

We have a git repo, it's all open-source, we're encouraging developers to help us build it and to build on top of it we're building internally on soullow. So you know this is this is a real framework, that's being used in production today that yeah I mean it's. It's been informed by our processes. We want to build it with the community. We want to.

C

You know, get good feedback on this and use it to improve our own as well as contribute. What we've learned back so.

B

A

Scott, if someone built an operator, one, that's they're able to kind of just share with the community. Where would they put that so.

C

We don't have a hub for it, yet, unfortunately, but this is going to be something that, if you're familiar with our system is our service, miss fob service. Miss fob, IO has an area for managing service measures that includes extensions applications that you can build on top and we're going to have hooks for, like you can see an auto pilot, or already we create these deployment manifests, which is really a first step towards building a home chart, which is eventually part of our roadmap that we get.

C

We can get hump chart and a release process, because these are all the things that we need to do again for our own software. This is something that's typically going to happen for most users who are building operators, so we're going to have extensions that allow you to easily plug it into whether it's a home tour repository or something higher level, whether you I, like the service Michelle.

A

If folks on the folks online have any questions, we can probably take like two questions before we hand it back handed over to a Christian for webassembly.

A

Anyone you can mute yourself or type it in the chat.

F

Hi everybody I have a question that I heard that it's irrelevant now ahead to the discussion. I want to know that about about the multi mesh solo multi may support. If you have time in this meeting today, please, let's start more about the multi mission, single mesh support that you have on on the plan. So.

C

What what type of support- because you know it all depends I mean multi mesh- is something that we support and various ways like autopilot has its own definition of what that means. Glue has its own definition. So it's a question of what are you trying to accomplish? Xavier you, you want to I assume, manage multiple meshes.

C

F

Sorry sorry I interrupted the first thing that I wanted to to clarify the definition of the single mesh and multi mesh in your terms and then then what what is the roadmap of supporting multi mesh in the new future from solo? So.

E

That is an excellent, excellent question and is probably one that either it's gonna be more involved. Okay, and we definitely want to get the answer for you either.

E

We have time at the end, or we can set up a separate discussion for you, because that's that is that is core to what we're doing at at the company and I don't want to shortchange it. Okay,.

F

A

Let's spend some time talking about kind of like to mesh at the end. I think that's good yeah.

E

We can do that at the end. Yeah.

A

Because we're talking about some service mesh, some tools for service mesh here and then we can yeah. You know we.

E

Can talk about how it all kind of wraps up actually because autopilot and what I'm about to talk about with webassembly is all very related to the product that we are working on right now called service mesh hub and that's where our service, our multi mesh and management capabilities, start to come come together. So, yes, I'll make sure to leave time at the end, so that we can. We can dig into that cool.

A

Thanks Nelson, anyone have a question on autopilot, specifically right now or even random trivia for Scott moving tips. He is moving this weekend. What's.

C

Appreciated yeah yeah I'm just come. If you got any questions, autopilot is still in development. We plan a lot more features coming forward soon. Multi cluster support is currently in the works. So if you've ever needed to write an operator, that's going to manage more than just this local cluster autopilot is going to help with that.

C

I didn't even showcase the features of connecting to Prometheus and using that to monitor metrics from your cluster, using that as a source of data for your operator to act on, but we can say that for slack for future tutorials releases, webinars yeah, so please come and check it out. Oh.

A

So then, let's pass it over to Christiana. So if you want to get in on web assembly, hell yeah.

E

E

Right, you should be able to see my screen right. Yeah.

A

And if there's specific topics that we talk about here, folks, that you want more information on and something like an auto pilot, if you want like to have a session on, because you have an idea for an operator and you want to talk through it- a little more just hit us up in slack or you know ping us in the chat here and then we can. We can. We can do a deeper dive, yep.

E

Thank You, Betty and I'd also want to point out the beginning here that this topic will probably spend 15 20 minutes on it, but it is actually there's a lot more to it and we are going to do a full length webinar in January or February, we'll announce it on webassembly.

E

Why it's important, how it fits in with envoy on boy proxy and what we're doing at solo in more detail, but right now, we'll just probably go a lot quicker and and get to get to the demos.

E

Scott showed autopilot, which is a way of building workflows on top of service mesh. Now what we're gonna look at is so, if you look at that is kind of a up here above above the mesh, but we're gonna look at is slightly below the mesh in the data plane what's running in the request path and and we'll bring it all together at the end when we discuss the multi cluster stuff it for most. So what web assembly is is a a binary language that can be injected into edge processes.

E

I'm gonna leave that vague for a moment that can customize the behavior of that edge process and the initial use case for it was in speeding up the execution of very, very CPU and resource intensive computations in the browser wherein people would write their applications in JavaScript and but they needed a way to run things a little faster, but they needed to do it in a in a safe way right because you don't want to inject code into a browser, and you know, go through all of the vulnerability issues that Java applets had and that JavaScript had and so forth.

E

So we needed a safe way to do that now. That also applies to things outside of the browser and in an edge application architecture. For example, you might be running a a proxy at your edge for an ingress or API gateway scenario, or you might be running those proxies within your applications themselves in a service mesh scenario, and you need more fine-grained, customization and control over the routing over, let's say, transformation over security.

E

You know we we work with a lot of customers that have very unique security requirements that need to stay backward-compatible and sometimes out-of-the-box. Solutions need some level of customization to fit nicely into these environments, and that's where webassembly comes in so with webassembly we can. We can build that customization and inject it into envoy proxy. So that's that's. What we're gonna be talking about today, inject it into Envoy proxy in the anon voice, request processing pipeline so that we can customize it now.

E

It's important to note that webassembly is a compiled target, so you're, probably not going to be writing webassembly directly. What you would pry do is use a higher level programming language like a C++ or a rust or tiny go and as more languages come on board and the garbage collection support gets, gets better solidified in the web assembly community, you know more, more languages will be supported, so you're not forced to.

E

If you want to extend envoy- which we do, it's all lonely and we've done for the last few years and that's you know we built that as a as a core competency of our company, but that's not everybody's core competency. If you want to extend it, you don't have to write just C++, which is what envoy is written in and then, if you do that, you have to compile it and link it into the Envoy binary. And now you have a bespoke Envoy binary that you not have to maintain so so web assembly is this.

E

You know you pick the language that you want to write your extension in that you're comfortable with you use a tool chain to compile that down into web assembly, and then you can safely inject into a running envoy proxy. Now I do want to make a disclaimer that the Envoy at upstream envoy is the the webassembly or wasm. Support is experimental right now and that it is in the process of being merged upstream and we'll get there.

E

I would like to say shortly, but depending on how quickly the on Boyd community is willing to work with it, but so, let's, let's pop over to what we're doing now, to support that at solo.

E

We think that these extensions that people build to the data plane to envoy remember autopilot was the extension for the service mesh now we're down and in in the request path. These extensions that people build for transformation for security for various auditing and logging purposes that these will either internally in an organization wish to be shared and reused or even outside, in a more of a community effort, shared and reused, and so Tuesday.

E

This week we announced a offering called the web assembly hub or a community offering called web assembly hub that allows you to build web assembly modules and share them with the rest of the community, allow you to discover them and search for them, and and we've also built some accompanying tooling, to simplify building the web assembly modules specifically for Envoy and I'll.

E

Show you that in a second is that's that's huge right having web assembly is awesome, having it working on boy is awesome, but now the workflow and the tool chains and all the stuff that you need to connect together. That can be pretty difficult and we've solved all of that for you, so that I'll show you that in in a second, but if we click on explore here on the web assembly hub, we see we have a seated a handful of extensions or modules.

E

So in this case we have a transformation module that you can use for transforming messages. Inside of envoy. We have one that exposes more detailed metrics. If I click on this, we can see total requests, count and number of rejected requests because of external off failures and so forth. So we've added a little bit more richer metrics. That would be useful for people and then the Amazon lambda routing to Amazon, lambda and taking care of the handshake. For that that is just you know, we're a day or so away from from that one coming.

E

So it's coming soon and if some of these other ones we actually having glue today and but we're working through exposing those into the into the hub, as well so on to the more interesting developer piece which is okay, yes, I have this use case. I have I need I need to customize envoy I, don't want to start to I, don't want to become an envoy, maintain maintenance company I, don't want to build it directly into the into the binder itself. I want to use web assembly. How do I get started?

E

How do I, how do I do that, and we announced in in concert with the web assembly hub is a new project called waz me well, as me is a tool that helps you automate the workflow for getting started with and building and then publishing and sharing your web assembly modules for Envoy. So if you come here, this open open source project, you know there's some instructions here. If you go to doc so Lodi, oh / web assembly hub whatwhat week, we can send this out in in the notes. Afterward.

E

You will you'll see some some information on installing it and for running through what I'm about to show you but I'll show you here live so the first thing that we want to do is: let's: let's take a look at what we have running in our kubernetes cluster. Already we have a couple serve.

E

We have an echo server that when you call it just returns your whatever you sent it to it, alright, so very simple: we're going to use glue, which is a which is a a gateway, open source, API gateway built on on top of envoy, and that supports the web assembly modules extension in in a beta form right now, and that will start to solidify as upstream also sort of solidifies, but I also do want to point out that you could do this directly with with upstream as well, and if you look at the that's a if you look at the docks here, we see one with with glue and one with vanilla envoy, but I'll show you with how it works with glue.

E

The API is much nicer right now, so the first thing we're going to do is we have this echo server? If we, if we take a look at our glue deployment, this is our API gate, but this is on glue right here, a gateway proxy.

E

What we want to do first is show that we can call this echo server through our Envoy proxy through glue, so to do that, let's just run a curl. We can see that yep. It makes it through. It should be two hundred. We just get the request back that we that we sent in into echo request. So let's build a web assembly module that we will inject into envoy to modify the behavior of envoy as it's processing a request and a response. We're going to keep it simple.

E

All we're gonna do is on a response, so request comes in gets process on a response back to the client. What we'll do is we'll add an additional header or two headers that that we can show that we can customize the request so to do that, we're going to use a tooth, the tool that I just showed on getting up called waz me and we're gonna, say waz me in it here and actually will put it into a folder called filter.

E

So let's do that if we do LS here, we can see our new our new folder here filter, so we'll go into filter, and if we do a listing here, we see that it generated a new project for us, in this case we're going to use C++, but you could also use rust and and, like I said, more languages will be supported soon, but it's set up here and we'll just open it up.

E

It's set up our project.

E

We have the the filter code, so this is the C++ code that we will we'll look at in a second which will then ultimately compile down into webassembly, but it also set up a bass or basil, build project so that we can use that to pull in all of the right dependencies, the all the right, the in scripting tool chain that that compiled C++ down into wasum and any of the other libraries that we would need for for this to be able to run successfully inside of on void. So that's all good.

E

What we're gonna do now is going to come to the code and we're gonna make a change here on the on response. Headers- and hopefully, y'all can see this in the in the room here. What we're going to do is we've got these new headers. What we're gonna say is, let's just let's just add a new one, just for just for this.

E

So let's say this is online: meet up and we'll call this glue blossom. So this these Hatter's should be injected these three headers new value online, meet up and location should be injected in the response when we when, when we add this to envoy, so that's good. Now, let's do a build. Let's do a build of Waze me build and what this is going to do. Is it's going to pull down a docker image that will then run this basel build?

E

So if you don't have basil on your computer- and you don't want to deal with all that stuff, then you know it. We could very easily run it inside of docker basil, pull in all of the dependencies and so on and build the C C++ code down into web assembly. Now, the only issue with that, when doing it in a live demo setting is that this could take a little while, on the first build on the first run, basil will pull down the whole internet and eventually it should succeed.

E

So we're not going to wait for that, because we want to get to questions and so forth. At the end, so I did I did put together this existing filter project. If we look in, if we look inside of it, it should have the same changes that we made to the filter we come down here. We should see yep on my meetup clue possible exactly in this thing right and up just still, all bees are still running.

E

Kill the stalker: let's kill this thing, don't.

B

E

All right, we shouldn't see that stuff anymore, but sorry coming back to our code. We have we have this. This project that had already been built. So all the Basel stuff should be pulled down.

E

If we do a build, not allows me be able to do a Basel, build and we'll do we'll build it locally here and we'll give that it shouldn't take as long as it otherwise would have if we were downloading the whole internet, and so now what this is going to do is using the inscript M script in toolchain, compile the C++ and then so. You see lang to Kampala C++ and then take that and compiled down into a web assembly module which we can see right now.

E

It is linking, and hopefully does it take too long to do that and then once we do that we have ongoing running and we will hopefully have a web assembly module. Now we need a way to inject that into our Envoy proxy and the first thing we need to do with that.

E

If we need to publish it, when you publish this module to some registry, where we can, where we can then share it and and allow people to to pull it down and use it, and then this there we go okay, so we can see that this thing was built. We should find it here in basil bin.

E

We see it filtered Azam. Now this filter got wasn't like I said well, what assembly is it binary format, but it also has a textual format. So if you do or if you are interested, we can very quickly take a look at the the textual format.

E

Filter not allows them look at it here and what we can see is it's, you know the web. The way web assembly is structured. Is you define all of the functions that you've created in your module? You might need to import certain functionality, so webassembly is interesting in that it cannot by itself reach out of its VM out of its sandbox. If you want to call functions outside of it, they need to be imported and they need to be manually put into the web assembly sandbox and here's an example of that.

E

So we're importing the we're expecting these to be available in the web assembly VM so that we can import them and so on now you can see the function bodies down here so up above was the declarations. Now the function bodies here, but so you can, you know webassembly, is really interesting. I highly recommend looking at it or are coming to our follow up webinars now. What we want to do is when you use the waz me tool to to push our new web assembly here. Let's, let's look at this first.

E

If it allows me list, we should see some of the existing modules that have been pushed to the registry. This is ultimately kind of a docker registry or OCI compliant registry running. In the background, what we're gonna do is we're. Gonna do waz me push and we'll say we're gonna push it to webassembly dot io under my repo and we'll call this online test easier about one and then what we're gonna push is basil and filter, dot blasted so we'll give that a try.

E

Fingers live demo and it pushed, and we call this online test. And now, if we do list list again, we should see our online test and we do and that's excellent. So now we push it into a registry. Now. What we want to do is configure our Envoy proxy to pull that down to pull it down from the registry and load it into the proxy. So if we do.

E

This gateway wasn't, if we take a look at that, what we're going to do is we're going to configure Glu's gateway to load the correct image, which is it's not correct here, so we're going to have to look at that.

E

So this is version of online tests and then we'll cube apply this.

E

It may be too wasm that will then prompt envoy to up or to inject this new woz encode that we created and if we go here and watch should first of all make sure that proxy doesn't crash. We don't want that, especially in a live demo. We see that looks to be stable. So that's good. Now the the web assembly module should be loaded.

E

Let's try to curl that the echo service again and see whether the headers were correctly added, and if we do that, we should see indeed that in the response headers, that we have our three new headers new head, new header and online meetup header and the location header. So now everything works. We like it. Let's share this. Let's share this with the community and that's where Waze me, catalog ad comes into play. Alvie catalog ad is going to say alright. Now you want to publish this to everyone else and make people aware of it.

E

Let's so what assembly down I/o and that was online test v1. Let's do that, what it's gonna, it's going to prompt us for some of the details that you want to include in the catalog when people come and find it what's the name of it online test, yeah! That's fine! There's a short description! We'll go a little quicker because we running out of time the long description, yeah, that's fine, yeah, dot, IO, yep, yep, you're, all dots again,.

E

And now it will say: are you sure you want to publish this, and now this will kick off? We say yes, this will kick off a get up style, workflow for creating a PR into the web assembly hubs catalog. Now, if we come over here and look at this pull request.

E

We can see now the the owners of the catalogue then can go through the details and and decide whether to to approve it or not, and then that would bring it into the catalog. So that's that's what we're working on with respect to webassembly?

E

We think it's a an important part to our story, which is how do we make service mesh simple for people to use and to operate to to make it realistic in an enterprise environment, because in in many ways what's what's out, there already is lacking in that area, and we want to build the tools to to do that and webassembly hub represents and whether simply stuff in general represents how we can get the customization level of customization that we need within a API gateway within a service mesh to get it to run nicely and safely in your environment.

E

Service mesh hub, then, is our management platform for directing all of these changes and coordinating configuration, managing multiple clusters, managing security across the clusters and so forth. And then, when you have a service mesh in place nicely running, you can you can, you know, get the value out of it, which is how do we do canary automation? How do you chaos experimentation? How do we do service debugging?

E

All of these things that you need to build on top of the service mesh and that's where autopilot comes into play, to be able to write these extensions to the the mesh? So that's all I have I know we were going to get back to the multi cluster question, but were there any other questions in between that yeah.

A

So if there's a any other questions, mute yourself or just pop them in the chat.

A

Don't be shy, nope, okay, vomits, there's, no questions all right. If no one has a question, then Christian do you want to just address the multi cluster or the multi mesh kind of question? Yeah.

E

And so we are working on a project call or prod called service mesh hub service mesh hub yeah is our management, centralized management plane for running and simplifying a service mesh. Irrespective of what mesh implementation you decide to use, so it could be ISTE. Oh, it could be linker D. It could be at mesh from AWS console from from hasha coke and so on. Now, simplifying a mesh is a huge demand right now, because they're all fairly complicated and that's for a reason.

E

The API is that they expose are intended to be lower level, api's they're not intended to be driven by developers and by humans, actually in general, and so with with service less hub. We can simplify that by giving a more natural user experience depending on who's operating the mesh. Is it the sres? The developers interacting with it is the security folks, and so on now operating a single mesh? Is you know you could put?

E

You could probably get away with doing that and you know enough duct tape and and wire and so forth, but when you get into a non-trivial deployment of a mesh, that's going to involve multiple clusters, multiple control planes. So when you deploy the mesh you're gonna have multiple isolation zones, but you give up the most important property of the mesh, which is service, identity and observability and traffic routing. If you start to fraction fracture the deployments so what service Mashhad does is bring them together.

E

So if you have multiple deployments of an sto service mesh, we can we can unify those into a single pane of glass. If you have an sto and then you're also going to go to AWS and you you know, AWS app mesh is free, so you might use that as well and you need to bridge those two.

E

Then you know, service mesh hub can be used to do that and you know I think it might be a great idea to do an online meetup at some point where we have, but we show what we have so far with with this and illicit community feedback, because, right now we can do a multi mesh.

E

You know flattening of of the various meshes so I think that would be a useful topic at some point. So what well there any others were there specific questions that you had listened? Yes,.

F

And the first question is that so you define by by the service mesh hub we can in estill we can install a single mesh on on top of the to more than one cluster. Am I right? That's.

E

F

Okay, so what what is your definition in your term as a single mesh single mesh means that all of the clusters has there has one one control plane, separate control dock to each other or just one master? Pilots and everybody is their slave No.

E

So in, in our view, you will deploy a the the control plane, components to logical isolation zones right so think, availability zones right you'll have a you- would have a control plane for your your mesh in various, let's say three availability zones, because if one goes down you don't want to, you don't want to lose the capabilities and the other available. But that's that's the whole point for the isolation right, and so what servers much hub then does? Is it drives those those three control planes and it configures them?

E

So, for example, we're talking about sto. This do to achieve a unified.

E

The flattening of the mesh that services in one side of the mesh can talk to the other and maybe they're in different availability zones. You need you need some configuration or a lot of configuration that goes into place to make that happen, and so what service mesh hug does.

E

Is it simplifies that by automatically configuring, the service entries the virtual services any of the route CAS and intermediate CAS, and so forth, to be able to establish identity, single domain of identity across those three different availability zones, so that when the services are running and they're interacting with each other they're talking to each other, as though it's just one big single mesh, but in reality the deployment is, you might have a cube, a cube, a nice cluster in availability zone, one with an sto control plane there, as well as in the other availability zones, so admit it.

F

Means every cluster has its own pilots. Yes,.

E

F

And you define ease as a single mesh, not not multiple mesh, that connected to each other. They.

E

Would be defined in our view as a group of meshes actually, but a single group, one group and a group means one logical mesh. Yes, okay,.

F

We I appreciate, if you plan for a meet-up for these topics, because that's hot topics and I think that we can. We can contribute a lot on these topics. Yeah.

E

We would love to have you back and we would love to show you we can. We can go into more detail about that. We could show you some demos we can dive in and break the demo, so we can do. We can do any of that. Okay,.

F

I appreciate thank you. Yeah.

A

Any other questions I think we have time for one more. If folks can.

F

If nobody, how question hey, listen.

A

I think we should do a whole separate whole top whole thing just around this whole, like multi mash, yeah,.

A

The idea for a whole other discussion topic just to dig into like how do you define it? What are the different permeate like ways it could be set up permutations and how would you, like you know, configure and like operate it pros.

E

A

E

You would favor one deployment architecture versus the other and yeah I mean I. Think that would be great yeah.

F

The things is that, honestly, it's my understanding, maybe I'm wrong, but there is not a. There is no. There is not as a single definition from for single mesh real versus multi mesh. Everybody define different things. The first thing that how we define a single mesh versus multiple mesh, okay and then another thing that is a mesh Federation, and then we can dig into the what? What is the best tools to to unify the.

B

C

Mean I I think that there's like there's a very clear technical definition of at least for now. The way envoys set up because envoy is always receiving. It's been fake from exactly one control planning today that might change in the future. But if you have multiple control planes, I think that that's that's definitely a logical boundary between.

C

You know a service mesh instance or a service mission segmentation. You may consider it organizationally speaking to be all one big mesh, but you still need some tooling on top of the control plane like on top of pilot, etc itself, in order to orchestrate that and join them as a single mesh. I think that's what solo is referring to. You know it's.

C

It doesn't really matter if you're talking about down it, whether they're in multiple clusters or the single cluster, but we've been seeing lots of use cases that call for you know sometimes kubernetes is a service, etc, where you have basically a different control plan for each tenant or you, you have multiple control points that need to be managed across environments and you need a control plane for that, essentially, because you don't want to be managing whether it's in the single cluster or multiple clusters, it becomes that the scale of managing all that configuration just becomes immense and really really difficult to do.

C

Unless you have some sort of controller built on top of it, I.

A

Wanted to give myself William raise his hand, so I wanted to give him an opportunity to ask his question. I, actually.

D

Have already answered my question: ok,.

A

D

Ok, so I kind of feel maybe stupid asking it because it's a little bit of more of an honest question, so I'm kind of new to you're. All are you guys, working on I'm gonna make this from sort of the at the nodejs world and I last couple of weeks really stumped on all the capabilities of webassembly I was mostly curious, I guess a I think I'll go ahead and ask anyway for sake of looking like a you know, new I'm interesting orchestrating orchestrating a cluster hub of webassembly containers by their salt I.

D

Guess that's something totally do right. So.

E

When you say webassembly containers by themselves, what does that mean? Yeah.

D

So I guess, let's say I say: hypothetically I had you know like I, don't know couple.

E

D

Running that are brought there, rust they're compiled into one something containers and.

D

Yeah, mostly I'm, mostly interested in like the process of one creating service mesh with those and then straightening I guess those.

E

Yeah, so let me, let me see, let me take a stab at it sure so with, and so this is. This is specifically web assembly running inside of on boycott, see yeah. So if it's webassembly running on its own, that's kind of outside of what we're what we're looking at, but so from from from a control plane.

E

Let's just we'll just be generic right now from a control plane, you might want to change the behavior of envoy when it runs at runtime and in the and from this control point what you should be able to do is say you know what I need. I need soap to rest transformation. You know, maybe that's an example for you, but I need soap to rest transformation on these calls over here you know: go! Do that control claim the control plane says: okay, well, I know, I, have a soap to rest.

E

Come assembly module here and I can go, deploy it to the Envoy running in this subset over here, and that I need to then configure those to behave a certain way all right. So we we need to be able to do that from a control plane independent of what service mesh is actually being run right, so maybe you're using this do, and maybe it doesn't have a really good way to do that. Maybe they do and then we just end up.

E

You know, building a nicer interface to be able to do that, but at some point envoy is the underlying capability and that's where we have our core competency and so, whether that's through the service mesh or around the service mash, to be able to do this. We you know that is a amazing goal with with the service mesh hub cool. Does that answer? Yes,.

A

Thanks everybody for joining were and actually staying on, because we kind of went over the hour so really appreciate it. This is recorded we'll, send it out along some links to learn more and if you have following questions just up into slack solo, do that's our community slack channel, so we can keep the discussion going on in there and then or and then there's also, you can kind of DM us respectively on Twitter or go to the so flow Channel until next time. Thanks everyone.

E

Yep. Thank you all for doing. I.

A

E