[potiuk]

Deadline Alert callback kwargs were decoded in the scheduler and the API server whenever a
serialized Dag's deadline alerts were loaded. Neither component calls the callback: the decoded
callback is only used to pick the callback model by type and to copy serialize() back into
Callback.data, which is an identity round-trip. Decoding there means constructing user objects
in components that never use them, and it is also lossy for sync callbacks, whose kwargs went
through BaseSerialization and a JSON workload before reaching the worker.

This keeps the kwargs in their serde-encoded form on the server and decodes them at the point
of use, on both consumption paths:

• Server: _decode_deadline_callback keeps kwargs encoded (only a legacy
  {"__type": "dict", "__var": ...} container is unwrapped). Re-encoding a decoded
  SerializedDeadlineAlert writes them back unchanged instead of serializing them again.
• AsyncCallback (triggerer): TriggererCallback.queue stores the encoded kwargs on the
  trigger row as-is via a new Trigger.from_serialized_kwargs, giving the same stored form as
  before. The triggerer decodes them with its existing pipeline when it creates the
  CallbackTrigger, so the triggerer itself is unchanged.
• SyncCallback (worker): execute_callback decodes the kwargs in the callback subprocess
  right before calling the user callable. The scheduler-provided context is passed through.
• SyncCallback version note: the executor callback data now carries
  "kwargs_encoding": "serde". BaseExecutor.run_workload passes it to supervise_callback only
  when set, and the Task SDK decodes the kwargs only when it is present, so callbacks created
  before upgrade (no marker, already-decoded kwargs) run unchanged. A worker with an upgraded
  apache-airflow and an older Task SDK fails such callbacks with an error instead of running
  them with encoded arguments. Workers should upgrade apache-airflow and
  apache-airflow-task-sdk together; see the significant newsfragment.

Both the current serde encoding and the legacy {"__type", "__var"} encoding are handled, and
Callback rows created before upgrade (holding already-decoded kwargs) keep working on both paths.

Tests cover:

• kwargs naming a class are not constructed during server-side decode, and round-trip
  byte-identically;
• AsyncCallback kwargs arrive decoded in the CallbackTrigger built the way the triggerer
  builds it;
• SyncCallback kwargs arrive decoded in the callable after a full
  store -> workload JSON -> execute_callback round-trip;
• each of the above for current and legacy encodings;
• with the encoding marker the worker decodes, without it the kwargs are passed as-is, and an
  unknown marker fails the callback;
• the async path is unchanged.

The test that asserted the previous decode-on-load behaviour is replaced.

---

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

• Yes (please specify the tool below)

Generated-by: Claude Opus 5 following the guidelines

---

• Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
• For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
• When adding dependency, check compliance with the ASF 3rd Party License Policy.
• For significant user-facing changes create newsfragment: {pr_number}.significant.rst, in airflow-core/newsfragments. You can add this file in a follow-up commit after the PR is created so you know the PR number.

[potiuk]

@ferruzzi @ramitkataria Could you take a look at this when you have a moment? It's targeted at 3.4.0 and needs a review to make the cut on 2026-10-19. CI is green.

[vatsrahul1001]

Looks good to me. Callback kwargs stay encoded on the server and are decoded where the callback runs, the encoding marker keeps callbacks created before the upgrade working, and the triggerer now fails a single undecodable trigger instead of stopping the runner. Could use reviews from @ferruzzi @ramitkataria before merging.

[github-actions]

Backport successfully created: v3-4-test

Note: As of Merging PRs targeted for Airflow 3.X
the committer who merges the PR is responsible for backporting the PRs that are bug fixes (generally speaking) to the maintenance branches.

In matter of doubt please ask in #release-management Slack channel.

Status	Branch	Result
✅	v3-4-test

[dondaum]

@potiuk I’m actually not sure whether this change might lead to any issues:
https://github.com/apache/airflow/blob/main/airflow-core/src/airflow/jobs/scheduler_job_runner.py#L1436 in _enqueue_executor_callbacks
As far as I understand, the scheduler uses the decoded callback.data; therefore, they must be decoded for everything to function correctly. Or am I missing something important here @ferruzzi ?

(Apologies for the tag – I just saw this change and realized that it is part of 3.4.0.)