GitHub user seunggihong added a comment to the discussion: Clearing tasks from old DAG runs resets start_date instead of preserving original
### This behavior is a bug and not intended - Reported in **GitHub Issue #30124** as a bug. It was later fixed through **PR #30125**. - The key message was: *“When clearing tasks, the DagRun’s `state` and `start_date` should not be reset.”* This behavior has since been corrected. ### In which version was it fixed? - The bug was observed in **Airflow 2.5.1**, and according to the issue discussion, it was fixed by **PR #30125**. - Therefore, starting from **version 2.5.1 and above** (e.g., 2.6.x, 2.7.x, 3.x), this issue should no longer occur. ### Workarounds or alternative fields - In **GitHub Discussions #25440**, similar confusion was raised. Many users agreed that it was counterintuitive for `start_date` to be updated on task clear/retry, and the discussion concluded that it should be fixed. - As alternatives, some fields were suggested: - For **manual triggers**, you can rely on `execution_date`. - For **scheduler-based runs**, you can use `data_interval_end`. ### Recommendations - Upgrade to **Airflow 2.5.1 or higher** (e.g., 2.6.x or later). This prevents `start_date` from being reset when clearing tasks, preserving the original execution history. - Additionally, if you need a reliable reference field, consider using `execution_date` or `data_interval_end`. GitHub link: https://github.com/apache/airflow/discussions/55206#discussioncomment-14303247 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
