timeout_seconds

applies to a single attempt. If a task times out and has retries configured, the retry starts with a fresh timeout window. It is not a cumulative limit across all attempts. So for example:

python
@task(timeout_seconds=30, retries=2)
def my_task():
    ...

This allows up to 3 attempts total, and each attempt can run for up to 30 seconds. I verified this against Prefect 3.x source/tests: - The task engine wraps execution in a timeout context for each run attempt - There is also a test specifically confirming that the timeout deadline resets on retry If helpful, I can also show what state transitions look like when a timeout triggers a retry.