[jira] [Created] (AIRFLOW-6535) add exception subclass to fail immediately without retrying

["Jon Stern (Jira)" <ji...@apache.org>]

Jon Stern created AIRFLOW-6535:
----------------------------------

             Summary: add exception subclass to fail immediately without retrying
                 Key: AIRFLOW-6535
                 URL: https://issues.apache.org/jira/browse/AIRFLOW-6535
             Project: Apache Airflow
          Issue Type: New Feature
          Components: models
    Affects Versions: master
            Reporter: Jon Stern


I would like to be able to configure certain tasks to retry, but to also have a way to bypass retry if I can detect a condition that is unlikely to be change for the better.

For example, imagine I have a DAG with a large number of tasks that fetch data from an API ... sometimes the API returns a 200 and everything is fine, sometimes the API returns a 500 and I know this means that API is failing under load and I want to retry later, and sometimes the API returns a 400 indicating that I've configured my requests in a way that will never succeed. If the API returns a 400 for some/all of my tasks, then I have to wait for all of them to get through all their retries before the run fails.

My proposal is to add another subclass of AirflowException called AirflowFailException, and to update the exception handling in TaskInstance.run_raw_task such that when this exception is seen, the resulting behavior is that same as if we entered handle_failure and is_eligible_for_returned false (except with adjusted logging so it does not look like we simply hit our retry limit). 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)