@eddyq/queue

Node.js client for eddyq — a Rust job queue that runs on Postgres, Redis, or both.

Three classes you can import:

• Eddyq — Postgres backend. Transactional enqueue, durable batches, migrations.
• EddyqRedis — Redis backend (via Redis Functions). ~70k jobs/sec bulk ingest, { every: ms } schedules, no migrations.
• EddyqApp — multi-backend container. Route per-queue across both in one process (e.g. webhooks → Redis, payments → Postgres). See examples/redis-basic/multi.mjs.

See the main README for backend tradeoffs and benchmark numbers, or @eddyq/wakeboard for the dashboard UI.

Install

pnpm add @eddyq/queue

We recommend pnpm or yarn over npm — see the npm caveat for why NAPI-RS + npm has rough edges.

Quick start

import { Eddyq, CancelError, RetryError } from "@eddyq/queue";

const q = await Eddyq.connect(process.env.DATABASE_URL);

q.work("send.email", async ({ payload, id, attempt, signal }) => {
  const res = await fetch(payload.url, { signal }); // signal flips on shutdown

  if (res.status === 429) {
    const after = Number(res.headers.get("retry-after") ?? 60) * 1000;
    throw new RetryError("rate limited", { delayMs: after });
  }
  if (!res.ok && attempt >= 3) {
    throw new CancelError("permanent 5xx; giving up");
  }

  return { bytes: (await res.text()).length };   // stored in eddyq_jobs.result
});

await q.start();

// ... your app runs ...

await q.shutdown(10_000);   // 10s grace; fires abort signals

Migrations

eddyq does not auto-migrate at boot — slow migrations would stall every replica's startup. Run them as an explicit deploy step before booting workers.

The package ships a CLI:

DATABASE_URL=postgres://... npx eddyq migrate run
DATABASE_URL=postgres://... npx eddyq migrate list
DATABASE_URL=postgres://... npx eddyq migrate down --max-steps 1 --confirm

Or call from code:

import { Eddyq } from '@eddyq/queue';
const q = await Eddyq.connect(process.env.DATABASE_URL);
const report = await q.migrate();
console.log("applied:", report.applied.map(m => `${m.version}:${m.name}`));
await q.close();

eddyq.start() refuses to boot if migrations are pending. To bypass when schema is managed out-of-band, pass { skipMigrationCheck: true }.

migrate() is idempotent and holds a pg_advisory_lock per migration line, so running it twice (or from two deploy hosts at once) serializes safely.

Retry convention

Every failure persists the full { name, message, stack, directive, retryDelayMs } in eddyq_jobs.errors[].

Cancellation

Every handler receives call.signal — an AbortSignal that fires when eddyq.shutdown() is called. Pass it along:

q.work("download", async ({ payload, signal }) => {
  const res = await fetch(payload.url, { signal });
  return res.json();
});

shutdown(gracefulTimeoutMs?) defaults to 30 000 ms. It broadcasts abort to all in-flight handlers, waits up to the timeout, then force-cancels the runtime tasks. Orphaned jobs are recovered by the heartbeat sweeper on the next surviving worker.

Admin surface

Tuning

Defaults are sized for typical workloads — most users should leave them alone. Pass any of these to start() when you need to deviate:

await q.start({
  sweepIntervalMs:        30_000,    // heartbeat sweeper cadence
  staleAfterMs:           60_000,    // when a running job is considered orphaned
  heartbeatIntervalMs:    15_000,    // worker → DB heartbeat (must be ≪ staleAfterMs)
  cleanupIntervalMs:     300_000,    // retention sweep cadence
  completedRetentionSecs: 24 * 3600, // delete completed older than this
  failedRetentionSecs:    7 * 86400,
  cancelledRetentionSecs: 7 * 86400, // pass -1 to keep forever
  leaderLeaseSecs:        30,        // single-leader maintenance lease
  fetchPollIntervalMs:    1_000,     // ignored unless poll-only
});

When to touch them:

• High throughput (millions of completed jobs/day) — drop completedRetentionSecs to a few hours so cleanup keeps up.
• Long handlers (jobs that legitimately run > 60s) — raise staleAfterMs past your worst-case duration so the sweeper doesn't reclaim live jobs. Also raise heartbeatIntervalMs proportionally.
• Many replicas (50+ pods) — raise sweepIntervalMs so you're not hammering the sweep query from every pod.

License

MIT