Cloud Native Computing Foundation Cloud Native Wasm Day EU 2022, 19 May 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Running JavaScript, Python, and Ruby in WebAssembly - Michael Yuan, Second State/WasmEdge

Description

Running JavaScript, Python, and Ruby in WebAssembly - Michael Yuan, Second State/WasmEdge

Wasm was designed to run applications written in compiled languages such as C/C++, Rust, Swift, etc. However, as Wasm gains popularity, there are increasing demands to run Wasm applications in scripting languages such as JavaScript, Python and Ruby. Compared with native interpreters (or dynamic compilers), Wasm offers benefits to both devs and ops. Dev: Wasm is a polyglot environment that supports mixing high-performance compiled languages and easy-to-use scripting languages. For example, with Wasm, devs can safely wrap Rust functions in a JS API. Op: Wasm is a sandbox with OS access. It can be managed as a standalone container or be embedded in a host. Native scripting language VMs need to be wrapped in other runtimes (eg node) and Docker containers. Wasm can achieve significant savings in computing resources. In this talk, Michael will discuss the approaches and latest progress of Wasm support of scripting languages, like JS, Python, and Ruby. He will cover language interoperability, ecosystem (eg packages and modules) support, and performance characteristics. Finally, Michael will also briefly discuss Wasm support status for popular managed languages such as Java and .Net.

A

Hello, everyone hi, my name is michael and um I'm sorry I can't be here. You know I really want to be a business meeting person, but um it's not meant to be anyway. So um the the title of my talk is running javascript python and ruby in web assembly, so essentially running scripting language in web assembly. Right, you know. So, since we had a technical audience, so I would just cut to the checks right. You know a heart started. You know. Web assembly was originally designed not to be a scripting language.

A

Runtime right, you know it's uh because we were unsatisfied with javascript performance. We want to run cc plus plus code rust code and the compiled language code. You know those um with close native performance right. You know those applications alongside um javascript. You know so that javascript applications can take advantage of that. So it's meant as a supplement to javascript.

A

So as such inside the browser, the web assembly runtime route side by side of this javascript right, you know so you have a you: have a javascript runtime and webassembly runtime. There is a bridge between them and if you are writing rust applications and comparing web assembly, you would have something called the watson bike and if you are running compiling c applications, it would be something else.

A

So essentially there would be a web assembly runtime with a bridge that goes to javascript the javascript runtime acts as a host for the website, meaning the web is something wrong time just to draw the computation, but the the networking data file system and all apps were provided by the javascript runtime and made available to the web assembly application through the bridge right. You know, so that's the original design. You know, so our original design web assembly really is to run compiler applications not to run a java screen language application like javascript, you know.

A

So what has changed so how it's going? You know. Why are we talking about web assembly running? You know javascript running inside web assignment now well, before we get to why you know. Let's see, you know how it starts. You know. That's because again we are technical audience here right, you know. So, strictly speaking, it can be done right.

A

You know because, as um we just talked about, whether somebody runs applications compiled from c superspace and rust, and then you know, um although in the browser, is typically around side by side with javascript right, but you know what is written in c, surprise, plus or last practical, practically everything right, including all the javascript, a lot of those javascript runtimes interpreters, python, runtime interpreters, ruby around having interpreters, those are all written in cc, plus, plus or rust, so they can be compiled into webassembly.

A

Once you compile those into webassembly, you would have a webassembly application that then you can feed the javascript source code, python source code or ruby source code to run them inside webassembly, so to run web to run javascript inside webassembly. You do a nested architecture, you know, so you have a web assembly wrong time on the outer shell and it has a compiled javascript, runtime or python runtime, whatever. What have you inside it and as a web assembly application? And then you have, um you know different api modules to provide access.

A

You know like networking- or you know we're going to talk about those later, but then the scripting language application itself is feed into the web assembly application and get executed there right. You know so that's the overall activity, so we know it can be done a lot of the content in this talk is um you know we have that on our our website, so the example applications and all that, so um you know.

A

So if you missed anything you you know, don't worry, that's you know all the source codes available, and so you know we know it can be done, but why you know that's, uh why do you need to run um screaming language applications such as javascript python, ruby side web assembly? So the reason really is because web assembly is growing in popularity.

A

You know people are, you know, because, although it's start from the use case where the native compiled native applications running inside this browser is now on the server side on the edge on mobile devices on other places and the people wants to you know the the traditional c surprise, plus developers or rus developers are using web assembly, but new developers also want to use webassembly and a lot of those new developers want to use javascript and python.

A

You know so there's has been increasing need, I think in the last um maybe two years, that's uh we have heard a lot from our customers from our community. You know how you know how they want to use javascript and python and ruby to let to a lesser extent in webassembly, because simply because those are very popular languages, a lot of people know that a lot of people knows. You know um to write, especially to write simple embedded functions. That's uh that's that that are ideally suited for web assembly.

A

So next let let's look at some use cases of web assembly to understand why the scripting languages are so important. So here are some use cases. You know we we touched on them casually. Just um you know just now. So the first, of course you know um it's. What web summary has been been very successful is to extend existing applications with user defined functions. You know, so um you have applications that written in um any number of languages.

A

You know just to see java, you know whatever, and then you want to turn your software into a platform into a pass right. You know by allowing your users or other people to customize it by adding their own functions to your software right. You know essentially turning your um you know. This is what we call application. Serverless right, you know, that's, uh you know adding some for the user to add serverless functions or user defined functions into your application and that customize it to do the jobs that they want right.

A

You know so um in that case the user, the application itself may be written in, say, rust or go. You know whatever, but but the extension functions or the serverless functions is oftentimes written in the in a language that is low code. So you know javascript and python are, I would say, you know, a fairly popular languages that people want to use for this particular use case right and then there's um called h-cloud use cases I mean you know, that's.

A

uh You know, um for instance the the the compute node on cdn networks, or um you know, five gm ecs. You know those are data centers um on the edge that outside of the cloud big cloud data centers that are closer to the to the end users- and you know so there are a number of application use cases there. So, for instance, there's uh isomorphic rendering you know so meaning the server-side rendering of react or server-side rendering of popular javascript frameworks.

A

You know today, you know those are um those are rendered inside, say node.js or the node.js or dnode running inside docker right. You know all inside of vm and easier way to run those javascripts in in a lightweight environment such as web assembly and there's ai inference right. You know, so you have.

A

um You know pictures or other data captured by edge devices and it's we are economical to send it all all the way back to the to the cloud data center to process them, and we can process them right on the edge in edge cloud and uh um you know and have the results stored there or have you know, generate actions from there and you know things like that and then there's edgewise, there's age devices use cases.

A

You know webassembly, because websimply is a lightweight runtime, so it can run on mobile devices as a cross-platform runtime, uh essentially replacing the jvm in a lot of cases. So you can, you can embed a website runtime inside a mobile application like uh like amazon, has done with amazon prime video. You know that so they have a though the video player they have um needs to run.

A

um You know c c based um you know, applications and in order to um simplify the the work to put the to run the c based applications across um you know, according to their blog over 8 000 different devices types, you know uh all kinds of android devices and and ios- and you know things like that, and so they need uh in the middle layer. That's that provide cross-platform compatibility and web assembly. The uh with ability to um you know wherever somebody would be.

A

You know one looks ideal you um you know runtimes for this purpose right so but yeah, but in this case whether somebody not should not only be able to run, say c or us um compile native code, but can should also be able to run um say you know javascript and python things like that and then um last you know, there's a um you know, um there's um a very big set of use.

A

Cases that we have seen uh we're, probably seeing today from from this um from the cognitive washing day, is that um you know a very interesting use case of web assembly is to use it as a standalone um container. You know directly managed by kubernetes right, you know, so you could have a you, could have a service mesh that has you know, docker based or linux, based services and but also have web summary-based services, and- and in this case you know, there's uh um you know for a lot of computing tasks.

A

We especially embedded computing tasks. You know dark linux containers could be too heavyweight and I'm so you know we. We want to write the service, we want to write the microservices in webassembly and- uh and in this case you know, java javascript and the python both provide. Very you know compelling um you know, um service frameworks, for you know for running those micro services. So those are so. Those are the reasons why you know the use cases. Why do you want to run javascript or python inside a web assembly?

A

So um for today's talk you know we're gonna spend. Perhaps um you know um most of the time talk about javascript. You know because that's like we said this is one. This is. This is the most important use case and we're going to talk about python and ruby at the end of this talk.

A

So the way to run javascript inside webassembly, like I've just said, is to take a c c plus plus space to javascript um interpreter or javascript runtime engine and compile it into web assembly around. You can do every single byte code and run as a web assembly application right. You know. So um this is work. That's that people have done. You know, that's. We have done this as well, but you know um the community, you know um you know have have been supported. One of the you know most lightweight and quickest.

A

um You know javascript um interpreter called quickjs um into web assembly. So, as you can see the example here, it's about leveraging the web assembly was it standard, meaning that um you know to have this quick, js runtime running inside the webassembly runtime. In this particular case it was managed runtime, it would have access to the operating system. It would access to the um to the command line argument, for instance, and it will have access to the local directory, so it will be able to read files and read from the command line.

A

You know- and you know basically around a very standard um javascript, so it can. Essentially, you can put any javascript content here and be able to run it right. You know. So that's um you know, that's the basic stuff. You know that's uh to have um to have a quick, js, runtime, compiling webassembly, to run any pure javascript content and with this with quickjs um it's um you know um it's very easy to have. You know es6 module, you know so meaning that's um now we can.

A

um You know with um with the quickjs runtime inside webassembly, you know inside, say, boss match it can have access to the operating system. So it can read multiple js files right, you know, so you can now have um you know es6 syntax to um you know to export and import functions across different files. You know, so here is example. You know so at the top panel you have a you, have an exported function called hello, and then you have uh on the top panel.

A

On the on the other side, you have a synchronously exported function and also a variable, a string right and then you know uh in the middle panel you have the um you know the the the javascript application that imports the um you know the es6 modules, because the imports was imported from the file system. So it's a it could be a synchronous operation. It could also be a a synchronous operation right, you know, so it depends on the the the the syntax flavor that you want to use.

A

You could do it asynchronously or synchronously, and so both are fine and we want to demonstrate both use cases, and then you know using the quickjs runtime you can. You can load this and run inside web assembly and you can see the results here. You know that's, uh you know it would be able to execute. It would be able to import those functions from those modules and execute them right.

A

You know so that's a very simple example of you know um how to run uh yes, six modules inside the web inside webassembly and uh um the um so es6 module. Is uh it's a it's a relatively new standard for javascript um for a lot of older javascripts.

A

um You know, especially the ones that in the in the node.js ecosystem, um we are still using the the common.js module cjs module. You know to um there's a lot of javascript libraries are written in that way. So in order to use cjs, we have uh we use another tool called um uh called raw app. So you know that's essentially you know. There's uh um you use rob.js to to package a javascript application that imports those npm modules into a single file. You know that's, essentially it takes down. Also it's.

A

It's pulls down all those um module files from from npm repository and then combine them into a big file. And then you know you can use website. You can use the uh the quickjs environment inside web or something to run the entire file right. You know so that's um you know. So we also have a demonstration here. You know. That's basically it takes. uh um You know the um digital signature module and then computes md5, and it's also that's one.

A

You know- takes one of the math libraries and computes that so you know so those are so you know uh through rawa.js. We can also support, um you know commonjs modules and a large part of the npm ecosystem. A lot of npm modules can be supported. This way.

A

But if that's you know, if we just support, es6 and and cgs they'll, be you know, it'll be interesting, but um perhaps not yet competing, and what I find most compelling that adds new features to um you know to running javascript inside webassembly is really to uh is to allow um rust-based, implementations and webassembly and javascript to interact with each other, meaning that so you know you could have. You could write a function in rust and then expose this function as a javascript api.

A

So as a so as a javascript developer, they would not need to know the rust implementation behind it. They just need to call the api. So so um a lot of this api is provided by um by quickjs itself and our roster wrap is around quickjs right. You know so um on on the left side, you can see, there's a there's, a there's, a rust function.

A

That is, um you know, designed to interact with javascript, so it has constructs like the js function and js value, and things like that and then on the on the right side is that we we, when we compile the the quickjs runtime we now have hooks to import. This function, this function is a struct called slow, fn right and give it name. You know in the context, given name called hi so so then we run a javascript.

A

um You know in the javascript. We can simply refer this function as high right. You know so the the the javascript code is, uh um you know it's is on the right panel. It's code equals to high one two three right. You know that's uh so meaning in javascript. We just call hi one. Two, three and the javascript interpreter sees the symbol. Knows it's a building function that it can find its implementation in rust. So it goes to rust to find this implementation and execute this function and then give you the results right.

A

You know, so the um you know on the right. You can also see the the um the result from this. So you know um oh, why do we do that? The way the the reason to do?

A

That is that you know um so for the um for a lot of tasks that pure javascript is kind of slow or it's difficult to do you know, so the ability to have the ability to use um to write those tasks in rust and then have the whole thing compiled into webassembly allows us to mix the ease of use of javascript and the high performance of rust right. You know: that's really taking advantage of web assembly's capability to run both right.

A

You know to be able to run, you know rust, compiled applications and also javascript applications as interpreter right. You know, so I think this is one of the um you know compelling cases of why you want to you want to run javascript inside web assembly, because you could potentially run faster and safer right.

A

So with that in mind, you know so um we um you know, I want to introduce you to some some of the um you know rust apis that we have already built with our um you know, because I work at united internet at um at wasm edge. You know so at one mesh we have um some rust extensions, sdk extensions to web assembly. You know that's, uh so we build some of the functionalities um as javascript apis using the approach that I've just mentioned right. You know so, um for instance, here that's, uh you know.

A

As we know, um a networking socket has been a long amazing feature from wazi, but at watson edge. We added our own non-blocking, asynchronous networking. um You know socket support right and we have a rust sdk for that. So in our quick gs implementation we have on your quick js setup. We have. uh We have a rust module that implements http module inside javascript, so meaning that's when running quickjs inside was an edge.

A

You could write javascript, that's uh that does http request and does http and become http server as well as because http server could be asynchronous, meaning that it's not blocking right. You know it can handle um multiple concurrent. um You know, connections at the same time from the same port. You know using the same thread right you know, so so, there's uh so by incorporating a roster-based io library that was made has provided for for each web assembly implementation. We would be able to support. You know networking based java, the javascript applications invoice match.

A

Just using the uh the the techniques I have just mentioned right that, along the same line of sort you can, we can do the fetch api right. You know we can do http, because we can do the simple http request and the http server. So we can do the fetch.

A

um You know that's uh we're going to talk about the significance of this api in a couple of slides and you can also do the um you know um one of the interesting features of watson edge is it provides uh extension for tensorflow, so it also have was a wasting extension for openminder. You know, meaning that's um you know um you can use rust to write ai inference applications and have them compiled into webassembly and runs inside was made. You would know to use the gpu or to use other.

A

You know hardware or software components available on the device to do the ai inference and because this is a rust api for that extension. We can also incorporate that into javascript api. Using the technique we have just mentioned right so here is a is an example of javascript application that you can run inside inside was managed to take advantage of the um the tensorflow extension.

A

So now we have talked about you know, that's uh we have. We have networking, we can have. You know those um those those native functions in inside the web assembly. So one of the when we combine those together, one of the more interesting things that we can do on watson edge, is really to do server-side rendering for for for react applications.

A

So in react, 18, there's a there's, a a way to do server, side, running called streaming ssr and the way it works is to fetch the data from the from other servers. First build the model and render html and then send the html across the wire to the to the browser to do and then be rendered as uh as a ui in the browser right.

A

You know so this whole generating the html, dom components and the rendered content interest you know, can all be done on the on the edge server you know um inside the web is somebody around time now. So in the past you know you have to have node.js or dno and then have them wrapped inside a docker container. To do that, you know now we have you know, and now we have a web. Somebody run time. That is, that is much much faster. um That is much much smaller and almost as fast right.

A

So there's a um there's a uh example website and tutorials that you can. You can try on your own, you know, so you can take a react, 18 application and make little changes to um how the ssr server is wrong, because you are not no longer using node.js. You are using. um You know. The non-blocking socket involves a mesh, so there's um there's a tiny bit of code changes there and once you do that, you know you will be able to.

A

You would build, launch it and run it on the you know on your edge server.

A

So you know now you know one of the common questions I get is uh you know when you're choosing quickjs, you know how does that compare with v8? You know that's because we ate behind node.js and dino as the high-performance javascript engine can be as fast as v8 lan says. No quickjs is an interpreter. It's uh it doesn't have git, so it's much slower than v8. However, as we said you know, one of the things is that a lot of those bottlenecks, like you know, a huge time consuming. um You know, tasks like generating strings.

A

You know ai inference, and you know network concurrent, networking and all that stuff. It can be done in rust. You know, so you don't have to do them in in javascript. In that case, you know the overall performance of application may not necessarily be much slower. You know, just of course the true benefit of using quickjs and webassembly.

A

Is that it's much much smaller, so v8 is like 40 megabytes in memory, but uh quickjs is like less than one megabyte, so uh even adding web assembly, it's still much smaller than say, you know um a v8 inside docker right. You know, and the um quickjs in web assembly is also safer, because um um web assembly runtime itself is a security sandbox and uh it has much smaller attack service, for um you know for security issues and it's also more manageable. You know because whether somebody runtime itself is uh is oci.

A

Containers can be managed by kubernetes and the same. uh You know in order for v8 to do that. You have to wrap it around in in linux container, which make it much much bigger and heavier right. You know so, of course, there's other approaches to run javascript inside um inside the web assembly. You know one of the leading exam alternatives is spider monkey.

A

You know, that's uh the javascript engine, the javascript engine, that's used in firefox, so there's articles um and and also code on how to do that and then there's uh shopify has a project called javi and it's also a set of rust, wrappers and tools around quickjs that shopify has developed. You know, that's uh that allows um you know um a pretty much uh it's it's a similar approach that than the one that that I have just uh that I've just that I've just described so check it out.

A

So the other way to do that is really to um incorporate v8 into webassembly using v8 as a host function. You know, meaning that you pass javascript evaluation directly into v8, but that's something you know, that's something that we are experimenting with and you know if you are interested, you know we we could talk more after this talk yeah. So you know that's the same thing about web assembly. Now we have a couple minutes left, so we want to talk a little bit about python and uh um and uh and ruby as well.

A

So for python you know so. For javascript we took a javascript interpreter and compiled into web assembly. We can do the same thing for python, so python, there's a python interpreter written in rust right. You know it's called rust python. You know, so you can take that project. You can take that project and compile it into whether something towards it. You know using use your favorite use your you know rust compiler to do that and then you can just run. um You know um a python command shell.

A

You know you can run any pure python applications in there right. um However, you know. That's uh um you know, um there's still a lot of work that needs to be done because most python applications out there are not pure python applications. Those are python wrappers around c applications. You know it's uh it's pretty much. You know same approach that we have just described right. You know we have javascript application wraps around rust applications right.

A

You know, so they cannot simply be interpreted by russell tyson because they have native libraries and, uh um however, those c libraries could potentially be compiled into web somebody as well. You know using the same approach that we have just described with javascript. So this is pretty much. uh uh You know a very exciting area. That's that's! uh You know. I think people should be working on. You know, of course, there's uh um you know, um because the the networking uh socket support in what is lacking. So we can't quite do python networking applications.

A

Yet we have to incorporate if we want to run the e-waste match we have to do. We have we can incorporate the wasa mesh socket api into python and uh and then supports python frameworks like fast api, to turn that into a microservice framework right. You know so so there are things like that. You know so um you know. That's uh that's still pretty much working progress and we'd love to collect collaborate with the community, then there's a another. Interesting approach is just to compile python to c right.

A

You know, turn the python code into c code and then turn c code into webassembly and run inside web somewhere on time. So that is another. um You know interesting approach. People have been trying that, but uh you know we can also see some roadblocks because you know primarily because um you know not all c apis are supported in wazi, so you're gonna find something that things that are missing and um you know so. Basically you may your.

A

Your management may vary uh using this approach and then one of the exciting um you know happenings last year was um the ruby team. The c ruby team has figured out how to compile the c based ruby interpreter into webassembly, so they even provide a very simple tool for that. You know so from their repository. You can just uh you know, compile you can compile that and run that team was made and you would be able to. You know, run ruby, pure ruby applications that way.

A

uh Of course you know, that's uh um that's um you know, ruby has more language features than weber. Somebody can support, for instance, that's threat, it has operating system processes, and you know things like that, so those things has to um has to be. You know worked out before this scan can be. um It can be a generic, ruby, runtime right.

A

So yeah, that's uh um you know, um you know we have got a very quick overview of you know um of scripting language support in web assembly using was managed as an example, you know, that's uh um and, uh like I said, javascript is probably the most um it's the one that has most application, but and also the one has the most mature support and, uh um and but python are getting there.

A

You know, I think, because there's lots of python developers and lots of people are interested in that and there's a clear roadmap of you know, adding more features to both javascript and python. You know to have them run better um in web assembly.

A

Right, you know, and uh ruby is not far behind as well, so you know I think, um hopefully, by by this time next year by the cloud of awesome day next year, that we would be able to see um a whole slew of you know: um scripting language applications running inside the web assembly. These are running inside as serverless functions or as extensions for other platforms. You know so. um Thank you very much.

A

So um I think that's, that's all the time I have um so um come check out the watchmatch and I'll talk to you later. Thanks.