[
https://issues.apache.org/jira/browse/AIRFLOW-123?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15285982#comment-15285982
]
Chris Riccomini commented on AIRFLOW-123:
-----------------------------------------
What exactly do you define as a timeout?
# Task is killed because it's a zombie
# A connection to some infrastructure times out
# A task runs for longer than it should
(1) seems to be the one that Airflow could do right now.
(2) is hard--it requires cooperation from the operator to throw some kind of
known TimeoutException (unless you want to use Python built-in TimeoutError).
(3) I supposed could be handled as a task-level SLA.
> Differentiate Between Failure and Timeout
> -----------------------------------------
>
> Key: AIRFLOW-123
> URL: https://issues.apache.org/jira/browse/AIRFLOW-123
> Project: Apache Airflow
> Issue Type: New Feature
> Components: db, operators, ui
> Reporter: Kevin Mandich
> Priority: Minor
>
> Currently, timeouts are classified as a failure (both task and DAG). It would
> be useful to differentiate states between a failure due to a timeout and, as
> an example, a failure due to an exception being raised in a PythonOperator.
> At a minimum, it would be useful to see a task which failed due to timeout
> visualized as a different color in the Tree View. It would also be nice to
> utilize this classification as part of DAG logic. Examples here include an
> option to retry on timeouts specifically, and the ability to take a different
> action downstream if a particular task failed due to timeout.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
