Pranaykarvi opened a new pull request, #64484:
URL: https://github.com/apache/airflow/pull/64484
## Description
Closes #64476
When LocalExecutor runs a task in a subprocess, the result (including any
exception) is passed back to the scheduler via a `multiprocessing.Queue`.
Python serializes queue entries using pickle.
Some exceptions — such as `httpx.HTTPStatusError` from the httpx library —
are not pickle-safe. Their `__init__` requires keyword arguments (`request`,
`response`) that cannot be reconstructed during deserialization. This causes:
TypeError: HTTPStatusError.__init__() missing 2 required keyword-only
arguments: 'request' and 'response'
This exception propagates out of `_read_results()`, crashes the scheduler
loop, and takes down the entire scheduler pod. The only recovery is to
disable the offending DAG.
## Root Cause
In `_execute_work_in_subprocess`, raw exception objects were placed directly
onto the result queue:
output.put((key, TaskInstanceState.FAILED, e))
Any exception whose class is not trivially picklable would cause
deserialization to fail on the receiving end.
## Fix
Wrap the exception in a plain `Exception` before putting it on the queue,
preserving the original type name, message, and full traceback as a string:
safe_exc = Exception(f"{type(e).__name__}:
{str(e)}\n{traceback.format_exc()}")
output.put((key, TaskInstanceState.FAILED, safe_exc))
This is applied to both the `ExecuteTask` and `ExecuteCallback` branches.
A plain `Exception` with a string message is always pickle-safe regardless
of the original exception type.
## Impact
- Scheduler no longer crashes when a task raises a non-picklable exception
- Full debugging information (type, message, traceback) is preserved
- Fix is minimal and does not affect any other executor behaviour
- Applies to both task execution and callback execution paths
## Testing
Reproduced the crash locally using a `PythonOperator` DAG that triggers an
`httpx.HTTPStatusError` (as reported in #64476). After this fix the
scheduler remains stable and the task is correctly marked as FAILED with
the full traceback visible in logs.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
