Retries, Timeouts & Failure Handling

Robust workflows need resilient task execution. auto-workflow offers per-task retries with exponential backoff + jitter, timeouts, and flow-level failure policies.

Task-Level Retries

@task(retries=3, retry_backoff=1.0, retry_jitter=0.3)
async def unreliable():
    ...

Sequence (no jitter): 1.0s, 2.0s, 4.0s delays before final attempt.

Timeout

@task(timeout=5)
async def slow(): ...

If the inner coroutine does not finish in 5 seconds, a custom TimeoutError is raised (subject to retry logic if retries remain).

Failure Policies (Flow Run)

Set at invocation:

flow.run(failure_policy="fail_fast")       # default
flow.run(failure_policy="continue")        # downstream tasks attempt even if upstream failed
flow.run(failure_policy="aggregate")       # collect all failures, raise AggregateTaskError at end

Error Surfaces

• TaskExecutionError: wrapper for a task failure.
• RetryExhaustedError: raised after final attempt fails (unless policy continues).
• AggregateTaskError: contains list of TaskExecutionError when using aggregate policy.

Observability Hooks

Events emitted: task_started, task_retry, task_failed, task_succeeded. Subscribe via:

from auto_workflow import subscribe

def on_retry(evt):
    if evt.get("task") == "unreliable":
        print("retrying", evt)

subscribe("task_retry", on_retry)