JavaScript client sdk for runpod

const { RUNPOD_API_KEY, ENDPOINT_ID } = process.env;
import runpodSdk from "runpod-sdk";

const runpod = runpodSdk(RUNPOD_API_KEY);
const endpoint = runpod.endpoint(ENDPOINT_ID);
const result = await endpoint.runSync({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
});

Once an endpoint has been created, you can send requests to the queue:

const { id } = await endpoint.run({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
});

You can check on the status of this request once you have the id:

const status = await endpoint.status(id);

If the request has been completed, the status object returned will contain the output of the request.

If you don't want to manage polling for request completion yourself, you can simply call runSync, which will enqueue the request and then poll until the request completes, fails or times out.

const result = await endpoint.runSync({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
});

If you have the id of a request, you can cancel it if it's taking too long or no longer necessary:

await endpoint.cancel(requestId);

For long running applications or troubleshooting, you may want to check the health of the endpoint workers:

const health = await endpoint.health();

Some Runpod endpoints are configured to yield multiple results for a single input, which can be streamed locally:

const { id } = await endpoint.run({
  input: {
    prompt: "7 photos of a horse the size of a Boeing 787",
  },
});

for await (const result of endpoint.stream(id)) {
  console.log("Stream yielded another photo:");
  console.log(result.output);
}

(this horse photo endpoint is fictitious, but it shows why you might want a stream of results)

All endpoint functions have a timeout parameter which is the maximum amount of time, in milliseconds, to wait for the function to complete.

For example,

const result = await endpoint.runSync({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
}, 5000);

will timeout after 5 seconds.

Note that this is the timeout on the local function call, not the underlying request in the queue. So this:

const result = await endpoint.run({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
}, 3000);

will not enqueue a request which can only take 3 seconds on the server, it will only timeout if the enqueue itself is slower than 3 seconds.

To affect the maximum time a request can take on the backend, you can provide a policy in the request input:

const result = await endpoint.run({
  input: {
    prompt: "a photo of a horse the size of a Boeing 787",
  },
  policy: {
    executionTimeout: 3000,
  }
});