Drain your workers on deploy without losing the job mid-flight

A deploy sends SIGTERM to every worker. Whatever task was mid-run either finishes or vanishes, depending on luck.

AdvancedRead 8 minDoing: an afternoon7 steps

▶ Start

What you'll have at the end

A worker that finishes its current job before it exits, proven by killing it mid-job and checking nothing is lost

Before you start

You need

A worker pool that already runs behind a real broker with late acknowledgment turned on and handlers safe to run twice if a job comes back around, so the only open question left is what happens to the process itself the moment something asks it to stop.

Not covered

Making a redelivered batch's own side effects safe to repeat if it runs twice; this recipe closes the gap in the worker's own exit, leaving what the job does once it lands again to a separate fix.

Leans on

7 steps

Where to go next

Make a background job idempotent

if a redelivered batch could double-send or double-charge instead of just repeating harmlessly, close that gap first, since a batch that comes back around here needs to already be safe to run twice.

Checked 19 Aug 2026

Part of the Background Job Processing Systems cookbook