Add on_success_succeed_dagrun option to teardown tasks

[kadubhumika]

Add on_success_succeed_dagrun to teardown tasks, so that a successful teardown
no longer hides a failed work task.

closes: #74727

Problem

With @teardown(on_failure_fail_dagrun=True) the teardown is the only task that
decides the Dag run state. If a work task fails but the teardown succeeds, the
Dag run is marked SUCCESS and the failure is hidden.

Fix

New teardown option on_success_succeed_dagrun (default True, so existing Dags
behave exactly as before). With on_failure_fail_dagrun=True and
on_success_succeed_dagrun=False:

Work	Teardown	Dag run
failed	success	FAILED
success	failed	FAILED
success	success	SUCCESS

What changed

• airflow-core/.../models/dagrun.py: _tis_for_dagrun_state also evaluates the
  tasks before a teardown that sets on_success_succeed_dagrun=False.
• task-sdk (@teardown, as_teardown(), _internal/abstractoperator.py,
  _internal/mixins.py, xcom_arg.py, bases/decorator.py, bases/operator.py,
  mappedoperator.py): new argument, property and validation, with support for
  mapped tasks and .override().
• airflow-core/.../serialization/ (schema.json, serialized_objects.py,
  definitions/baseoperator.py, definitions/mappedoperator.py,
  dag_version_diff.py): the new field is serialized and deserialized.
• Tests: test_dagrun.py covers all six flag and state combinations;
  test_dag_serialization.py registers the new serialized field.
• Newsfragment added.

Testing

test_dagrun.py, test_dag_serialization.py and the task-sdk
`test_setup_t

Was generative AI tooling used to co-author this PR?

• Yes (Chat Gpt)

[shahar1]

Thanks for your contribution! I'd like to ask for the following:

1. Add AI disclosure to the PR's body.
2. Undo unrelated Dataproc commits
3. Add documentation for using the new feature (howto/setup-and-teardown.rst)
4. Add tests to task-sdk
5. Fix the static checks
6. Fix the title according to the guidelines

[kadubhumika]

Thank You ✨

…

On Sat, Oct 17, 2026, 5:20 PM Shahar Epstein ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In providers/google/src/airflow/providers/google/cloud/hooks/dataproc.py <#74916 (comment)>: > @@ -26,7 +26,7 @@ from collections.abc import MutableSequence, Sequence Great! I'll take a look after CI is done — Reply to this email directly, view it on GitHub <#74916?email_source=notifications&email_token=BMB777KZMHUVRDMWGBZDBN35UNMJPA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTKNJVG43DGNBWHA22M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#discussion_r4302556526>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BMB777LAFAFZHHVCKUWJSAL5UNMJPAVCNFSNUABEKJSXA33TNF2G64TZHMZTGOBYGQ4DSMJ3JFZXG5LFHM2TQOJXGMYDCMBZG6QXMAQ> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BMB777N6LUROVEAXDMDB7EL5UNMJPA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTKNJVG43DGNBWHA22M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/BMB777N7Z2RDCFPA42KMZCL5UNMJPA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTKNJVG43DGNBWHA22M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you authored the thread.Message ID: ***@***.***>

Blocking concerns were addressed

[yuseok89]

I left one question inline about test coverage for mapped teardown tasks across a Dag serialization round trip.

[yuseok89]

For a teardown declared with on_failure_fail_dagrun=True, has applying .as_teardown(on_success_succeed_dagrun=False) after .expand() also been checked?

If this path isn't covered yet, would it make sense to add a test checking that

• the configured value is preserved after the Dag is serialized and deserialized
• the restored value is reflected in the Dag run state evaluation