Poll a Sume job from a Node script: deadline, backoff and exit code

Run the script with a job id and it polls getApiJobStatus until the job is terminal. It waits for the time that Sume asks for in next_poll_after_seconds, and uses its own doubling backoff, capped at 30 seconds, when the field is missing.

import { createSumeClient, getApiJobStatus } from "@sume-com/sdk";

const client = createSumeClient({ apiKey: process.env.SUME_API_KEY });

export async function waitForJob(id, { timeoutMs = 20 * 60_000 } = {}) {
  const start = Date.now();
  let backoff = 2;
  for (;;) {
    const { data, error } = await getApiJobStatus({ client, path: { id } });
    if (error) throw new Error(`status read failed: ${JSON.stringify(error)}`);
    if (data.terminal) return data;
    if (Date.now() - start > timeoutMs) {
      throw new Error(`job ${id} is ${data.sume_status}; it keeps running`);
    }
    const seconds = data.next_poll_after_seconds ?? backoff;
    await new Promise((resolve) => setTimeout(resolve, seconds * 1000));
    backoff = Math.min(backoff * 2, 30);
  }
}

const job = await waitForJob(process.argv[2]);
console.log(job.sume_status);

How to run it

Set SUME_API_KEY, install @sume-com/sdk, and run node poll.mjs <job_id>. It prints the final sume_status. On a deadline it throws and says the job keeps running: stopping the loop does not cancel anything on Sume's side.

Why it is written this way

The table lists the choices in the code.

Failed jobs return

The loop treats completed, failed and canceled all as terminal. Read sume_status after it returns; a failed job returns normally and carries a public error, so do not assume success.

Next step

After a completed job, fetch the output from result_url when result_ready is true. If you prefer to be told, use a signed webhook and keep this loop as the backup.