Short answer: - A task will only retry if it ends in an error state (e.g., it raises an exception, times out, or otherwise fails/crashes) and you’ve configured

retries > 0

. - Simply returning

None

is a normal success (Completed) and will not trigger a retry. More detail for Prefect 3.x: - Retries apply when a task run finishes Failed or Crashed and there are attempts left. Prefect schedules the next attempt (state will move through AwaitingRetry → Retrying). - Common ways a task “fails”: - Your task function raises an exception - The task times out (

timeout_seconds

) which results in a failure - The run crashes (infrastructure/runtime error) - Returning any value, including

None

, is a Completed state and does not retry by itself. If you want to retry on “bad” values, raise an exception when you detect the condition. About retry_condition_fn: - Use

retry_condition_fn

to control which errors should be retried. - It’s called when your task raises an exception and is given the exception; return True to retry, False to stop retrying and mark the task Failed. - Example:

from prefect import task, flow
  import httpx

  @task(
      retries=5,
      retry_delay_seconds=10,
      retry_condition_fn=lambda exc: isinstance(exc, httpx.HTTPStatusError)
                                     and exc.response.status_code in (429, 503),
  )
  def fetch_url(url: str) -> str:
      r = httpx.get(url)
      r.raise_for_status()
      return r.text

  @flow
  def my_flow():
      fetch_url("<https://example.com>")

If you want to retry when a task returns

None

, raise explicitly:

@task(retries=3, retry_delay_seconds=5)
def compute() -> int:
    value = maybe_returns_none()
    if value is None:
        raise ValueError("Got None, retrying")
    return value

If you share what behavior you want to retry on, I can sketch the exact

retry_condition_fn

or task code for it.