Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
26.10.0
-
None
-
MXS-SPRINT-281, MXS-SPRINT-282
-
Not for ES Release Notes
-
Description
If MaxScale is shut down while an ODBC call is in flight in a session that has already been closed, call_t() cannot hand its shared_ptr back to the session's worker, because the worker no longer accepts messages. The failed post deletes the task in the thread pool thread, which then destroys the session there. A debug build aborts on the assertion in Worker::Callable::~Callable() (worker.cc:335); a release build runs the destructor in the wrong thread. t_return() has the same problem. Found by the intermittent failure of exasolrouter_test. Possible fixes: cancel the ODBC calls and drain the thread pool before the workers are stopped, or make a failed post safe.