Cancellation

Cancel a run with CancelRun or an SDK client, with an optional grace period of 0 to 3,600 seconds (default 300):

await client.cancel(runId, { gracePeriodSeconds: 60 })

grace := 60
run, err := c.Cancel(ctx, runID, workflows.CancelOptions{GracePeriodSeconds: &grace})

client.cancel(run_id, grace_period_seconds=60)

The answer is the run, with status cancelling. Cancelling a run that has already ended is 409 run_not_running.

What happens

1. Waiting stops at once. Sleeps, callbacks and child workflows the run is waiting on are cancelled, and so are its running children.
2. The workflow sees a cancellation error at the step it’s waiting on: CancelledError in TypeScript, an error matching workflows.ErrCancelled in Go, and RunCancelledError in Python. Its code is workflows.cancelled.
3. Cleanup runs within the grace period, which starts when the cancellation first reaches your code. In a catch or finally block (except in Python, a check of the error in Go), the workflow can run run steps, with their usual retries, to undo or report work. Any other step fails at once.
4. The run ends cancelled when the workflow returns, whatever it returns, or when the grace period ends.

try {
  await step.run("provision", () => provision(order))
  await step.sleep("trial", "14d")
} catch (e) {
  if (e instanceof CancelledError) await step.run("deprovision", () => deprovision(order))
  throw e
}

Steps already running

A cancel never interrupts a step that’s already running in your code. Its result is recorded if it arrives within the grace period. On a pull worker, the worker learns of the cancel from its heartbeats and signals the running step, so a long step can stop early:

• TypeScript: the step’s signal aborts with a CancelledError.
• Go: the StepContext is cancelled, and context.Cause is workflows.ErrCancelled.
• Python: ctx.step.cancelled() turns true.

No new step starts in that invocation; cleanup runs in the next one.