uranusjr commented on code in PR #75184:
URL: https://github.com/apache/airflow/pull/75184#discussion_r4397335314
##########
task-sdk/src/airflow/sdk/coordinators/_subprocess.py:
##########
@@ -480,6 +544,222 @@ def _handle_set_task_state_store(
)
return super()._handle_set_task_state_store(msg, log, req_id)
+ def _monitor_subprocess(self):
+ try:
+ super()._monitor_subprocess()
+ finally:
+ if self._xcom_backend is not None:
+ self._xcom_backend.stop()
+
+ def _start_xcom_backend_call(
+ self, call: XComBackendCall, finish: Callable[[XComBackendResult],
None]
+ ) -> None:
+ """Send ``call`` to the XCom backend process, which starts on the
first call it handles."""
+ if self._xcom_backend is None:
+ self._xcom_backend = XComBackendProcess.start(
+ id=self.id,
+ client=self.client,
+ logger=self.process_log,
+ selector=self.selector,
+ on_result=self._finish_xcom_backend_call,
+ )
Review Comment:
If the helper dies mid-task, `_on_socket_closed` fails the calls that are
waiting, but `_xcom_backend` still points at the dead process. The next call is
queued on it and never answered, so the runtime hangs until its own request
timeout. Check `_xcom_backend._exit_code is not None` here, and either restart
the helper or fail the call right away.
--
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]
