Pterodactyl

Pterodactyl is a javascript sdk for Flyte

Quick Start

Pterodactyl allows you to author and register flyte workflows in javascript.

Assuming that you have a flyte installation already setup you can follow these steps to get a workflow registered in flyte.

If you don’t already have a flyte installation you can use kind + helm to set up a local kubernetes cluster running flyte.

Create a workflow

Create a workflow using the pterodactyl library in javascript. An example is provided below. Be sure to name it workflow.js so the remaining steps work correctly or replace workflow.js with your chosen file name in the remaining steps.

import {
  task,
  workflow,
} from "https://raw.githubusercontent.com/NotMatthewGriffin/pterodactyl/main/pterodactyl.js";

const sum = task(function sum(x, y) {
  return x + y;
});

const square = task(function square(z) {
  return z * z;
});

const myWorkflow = workflow(function myWorkflow(x, y) {
  return sum(square(x), square(y));
});

Create a Dockerfile

Create a dockerfile based on a deno image that copies your script into the container. Be sure to name this file Dockerfile.

FROM denoland/deno:distroless-1.21.0
COPY workflow.js workflow.js

Build the container image

Using the dockerfile above build the image. Its fine to change the tag provided to docker with -t to your prefered tag but you’ll need to replace jsworkflow:v1 in future steps with your chosen tag (you may wish to do this if you have a private container registry to host containers changing the tag to something like: <private-registry-url>/jsworkflow:v1). From the directory containing both your javascript workflow and Dockerfile run:

docker build -t jsworkflow:v1 .

Push your container to a registry/load into cluster

This step is very dependent upon how your flyte installation is setup. If you have a private container registry configured pushing the image you created in the last step will provide the best user experience. Assuming that in the last step you tagged your image with your <private-registry-url>/jsworkflow:v1, this can be done like:

docker push <private-registry>/jsworkflow:v1

If you’re running a local installation using kind you can use the following to load the image into your cluster without deploying a private container registry:

kind load docker-image jsworkflow:v1

Register to flyte using pterodactyl_register.js

If the steps above worked correctly then you’re ready to register your workflow with flyte. Be sure to replace localhost:30081 with the endpoint for your flyte installation if it is not also hosted there. Run the following to register with the same version of pterodactyl as was used in the example workflow:

deno run --allow-read --allow-net https://raw.githubusercontent.com/NotMatthewGriffin/pterodactyl/main/pterodactyl_register.js --pkgs workflow.js --image jsworkflow:v1 --endpoint localhost:30081 --project flytesnacks --domain development --version v1

Run the workflow in flyte console

After all above steps run successfully you can look in your flyte console under the flytesnacks project and development domain to find the workflow myWorkflow. At this point it can be run like any other flyte workflow. The flyte console interface will say that the inputs are all of type string but they will be fed to JSON.parse and the result of that sent as the argument to your workflow task functions. For this example try 2 for input0 and 2 for input1.