Background Workers

When to use a worker#

• Consuming a queue (RabbitMQ, Redis, Kafka) and processing jobs.
• Telegram, Discord or Slack bots on long polling.
• Streaming or ETL processes that read, transform and write continuously.
• Schedulers written inside the app rather than as separate cron entries.

Creating a worker#

1. 1
   Create the service

   In console.muerte.cloud press Create service → Background worker.
2. 2
   Connect the repository

   Pick the repo and branch, or a Docker image. Auto-deploy on push works exactly as it does for web services.
3. 3
   Set the start command

   For example node worker.js, python worker.py or celery -A app worker -l info.
4. 4
   Add variables and resources

   Queue and database credentials go into environment variables. Workers are usually RAM-bound rather than CPU-bound.

No port, no public URL#

A worker gets no public address and no health check on an HTTP port. It is considered healthy while the process is alive. If your process exits, the platform restarts it — an exit is treated as a failure, not as a normal end of work.

import { consume } from "./queue.js";

async function main() {
  for await (const job of consume("emails")) {
    await handle(job);
  }
}

main().catch((err) => {
  console.error(err);
  process.exit(1); // let the platform restart the worker
});

Graceful shutdown

On redeploy the container receives SIGTERM and a drain window. Stop taking new jobs, finish the current one, then exit — otherwise a deploy can interrupt work mid-flight.

let stopping = false;
process.on("SIGTERM", () => { stopping = true; });

while (!stopping) {
  const job = await queue.take();
  if (job) await handle(job);
}
await queue.close();

Logs, state and scaling#

• Everything the process writes to stdout and stderr appears in the service logs — this is the primary way to observe a worker.
• Local files disappear on redeploy unless you attach a persistent disk.
• Running several workers on the same queue works only if the queue guarantees a job goes to one consumer — check idempotency before scaling out.
• Deploy history and rollbacks apply to workers the same way as to web services.

Common pitfalls#