deno-lambda VS xgboost-node

Using Deno with AWS Lambda functions requires a custom runtime. You can build your own runtime or use one that already exists. If you’re concerned about security, I suggest maintaining a copy of an existing runtime and carefully inspecting updates or creating your own runtime. For this proof of concept, I deployed the Serverless Application Repository (SAR) app for Deno into my AWS account. I used the included Lambda layer and the provided.al2 Lambda runtime to create my Deno Lambda function. I created a file called index.ts with some basic JavaScript code that makes requests to two different websites and returns the HTTP status code of the response or a caught error. I then updated the function’s configuration to reference the exported handler function.

I'm doing a simple PoC / testing with Deno using https://github.com/hayd/deno-lambda and specifically the CDK instructions.

As I had mentioned, it requires using a Lambda Layer. See: https://github.com/hayd/deno-lambda/blob/2d90756a0f18b57f16e...

Using your own image (i.e. without using the base AWS image with the layer) you'd get even worse cold start times.

Decorators and TypeScript aren't supported out of the box in Lambda (without using deno) so we'll also need a transpilation step if we go this route. Fortunately this is a mostly solved problem for AWS CDK, AWS SAM and Serverless Framework users. If you want or need to roll your own, esbuild is a great place to start and seems to be the bundler of choice for this purpose.

In addition to the stuff posted here, there's the deno-lambda project for deploying to AWS Lambda

Cloud hosting support. Deno runs fine in a container, but lambda/cloud function support is difficult on most providers. deno-lambda exists but it only applies to AWS and can't be used with all CICD tools. Deno Deploy also exists but it's pretty new.

This chips away at one of the showstoppers for Deno for me, which is good.

But while "the vast majority" of npm packages don't require a gyp build step for native addons, some of those modules are pretty important, and I see no indication in the announcement that they're also going to be implementing the Node C API or the gyp build process.

Right now I'm working with a machine learning project, and XGBoost [0] is a direct Node.js extension [1] through the binary interface.

So this does bring things a step closer to being generally usable, but there are still significant roadblocks.

A WebAssembly build of XGBoost could work with Deno, but aside from some guy's unsupported side project/proof-of-concept for use in a browser, I'm not seeing an XGBoost WebAssembly build. And generally when deploying something like a machine learning model I'd rather use well-supported tools than to need to dive into the rabbit hole of maintaining my own.

And yes, XGBoost will likely eventually have that kind of support for Deno, but then the next bleeding-edge project will come along and only support Node.

Even assuming Deno eventually hits a tipping point in popularity where everyone wants to release Node _and_ Deno support in their bleeding-edge projects, there are still things that I miss from package.json that don't seem to exist in the Deno ecosystem.

Things like the "scripts" block: A nice centralized place to find all of the things that need to be done to a project, plus auto-run script entries that can trigger when a project is installed. And inheritable, overridable dependency maps (see the yarn "resolutions" block).

I'd love to jump into Deno, but I think there has been far too much "baby thrown out with the bathwater" to its design. It's the classic development problem of looking at a system and seeing a ton of complexity, but not really understanding that all of that complexity was there for a reason. Maybe when it re-evolves 80% of Node's and npm's features I'll be convinced to make the jump. I'm a huge TypeScript fan after all. But it still strikes me as a violation of "As simple as possible, but no simpler."

[0] XGBoost is a _very_ promising approach to machine learning, training models much faster and with much more accuracy than traditional approaches.

[1] https://github.com/nuanio/xgboost-node