Multiple workers

Workers belong to your task backend, not to Goatbeat. Scale them as you normally would: more Django Tasks workers, more Celery or Dramatiq processes, more RQ workers.

Goatbeat affects workers in two ways:

• Routing. A schedule's queue= (and priority= where the backend supports it) decides which workers receive its tasks. Unsupported routing is rejected rather than ignored: RQ has no priority; Dramatiq honors priority on RabbitMQ only; no backend lets a schedule pick a different broker except Django Tasks' backend=.
• Idempotency. In the at-least-once window a task can be delivered twice, possibly to two different workers. Use the idempotency key (Celery task id, Dramatiq message id, RQ job id, Django goatbeat_key parameter) to make the second delivery a no-op.

Make sure a worker consumes every queue your schedules route to: a message sent to a queue no worker reads waits forever.