The GitHub Actions job "Check newsfragment PR number" on airflow.git/1fanwang/aip97-failure-details has failed. Run started by GitHub user 1fanwang (triggered by 1fanwang).
Head commit for run: 8e1060dc0783775c551eb1e7cc20f5abb0a2632f / Stefan Wang <[email protected]> AIP-97: add FailureDetails primitive for infrastructure-side failure context Foundation for AIP-97 (Infrastructure-Aware Task Execution). Today the on_task_instance_failed listener only sees the worker-side error exception; failure causes that originate outside the worker process — OOMKilled / PodEvicted on Kubernetes, WorkerLost / SoftTimeLimit on Celery, oom-killer SIGKILL on the local executor, preemption / eviction on resource-managed clusters — are visible only to the executor and never reach the listener. This PR adds the executor-agnostic FailureDetails type (executor_kind, infra_reason, infra_metadata) and extends the on_task_instance_failed hookspec to accept it as an optional keyword argument. Per-executor wiring is intentionally deferred to follow-up PRs so each executor's surfacing PR can iterate against a fixed contract. pluggy dispatches by parameter name, so existing hookimpls that don't declare failure_details keep working unchanged. Tests cover the FailureDetails dataclass surface (parametrized over five realistic executor / reason / metadata shapes, default factory empty dict, frozen immutability) and verify both the new listener-receives-failure-details path and back-compat for legacy listeners that don't declare the parameter. Report URL: https://github.com/apache/airflow/actions/runs/25368821149 With regards, GitHub Actions via GitBox --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
