Fix invalid memory access during the shutdown of the parallel apply worker.
The callback function pa_shutdown() accesses MyLogicalRepWorker which may
not be initialized if there is an error during the initialization of the
parallel apply worker. The other problem is that by the time it is invoked
even after the initialization of the worker, the MyLogicalRepWorker will
be reset by another callback logicalrep_worker_onexit. So, it won't have
the required information.
To fix this, register the shutdown callback after we are attached to the
worker slot.
After this fix, we observed another issue which is that sometimes the
leader apply worker tries to receive the message from the error queue that
might already be detached by the parallel apply worker leading to an
error. To prevent such an error, we ensure that the leader apply worker
detaches from the parallel apply worker's error queue before stopping it.
Reported-by: Sawada Masahiko
Author: Hou Zhijie
Reviewed-by: Sawada Masahiko, Amit Kapila
Discussion: https://postgr.es/m/CAD21AoDo+yUwNq6nTrvE2h9bB2vZfcag=jxWc7QxuWCmkDAqcA@mail.gmail.com
Branch
------
master
Details
-------
https://git.postgresql.org/pg/commitdiff/3d144c6c86025272e1711539f5fafb6fb85c4feb
Modified Files
--------------
.../replication/logical/applyparallelworker.c | 30 ++++++++++++----------
src/backend/replication/logical/launcher.c | 24 ++++++++++++++---
src/include/replication/worker_internal.h | 2 +-
3 files changed, 38 insertions(+), 18 deletions(-)
