The GitHub Actions job "Tests (AMD)" on 
airflow.git/databricks-repair-from-airflow-3 has succeeded.
Run started by GitHub user PrakshiGoyal10 (triggered by eladkal).

Head commit for run:
b621d4306cf289698091d687d27ac2ab2f58b5cd / PrakshiGoyal10 
<[email protected]>
Fix Databricks workflow repair failing on retried tasks and explicit task keys

Two defects made the repair button fail against real Databricks runs while unit
tests stayed green.

get_run_failed_task_keys iterated the per-attempt task list returned by the
Databricks API, so a retried or already-repaired task produced its task_key more
than once. Databricks rejects rerun_tasks with duplicate keys, so "Repair All
Failed Tasks" failed with a 502 for any run with retries, and a task whose
latest attempt had succeeded was still repaired. It now keeps only the latest
attempt per key before judging state, matching how the operator resolves a task.

An explicit databricks_task_key does not survive Dag serialization, so the API
server reconstructed md5(dag_id__task_id) and mismatched the real key: 
single-task
repair targeted a non-existent task and repair-all skipped clearing that task's
instance. The launch task now records the Airflow task_id to Databricks task_key
mapping in its run-metadata XCom, which the endpoint reads server-side; runs
launched before this fall back to the md5 derivation.

Report URL: https://github.com/apache/airflow/actions/runs/31714826151

With regards,
GitHub Actions via GitBox


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to