pgsql: Use appropriate wait event when sending data in the apply worker
Use appropriate wait event when sending data in the apply worker.
Currently, we reuse WAIT_EVENT_LOGICAL_PARALLEL_APPLY_STATE_CHANGE in the
apply worker while sending data to the parallel apply worker via a shared
memory queue. This is not appropriate as one won't be able to distinguish
whether the worker is waiting for sending data or for the state change.
To patch instead uses the wait event WAIT_EVENT_MQ_SEND which has been
already used in blocking mode while sending data via a shared memory
queue.
Author: Hou Zhijie
Reviewed-by: Kuroda Hayato, Amit Kapila
Discussion: /messages/by-id/OS0PR01MB57161C680B22E4C591628EE994DA9@OS0PR01MB5716.jpnprd01.prod.outlook.com
Branch
------
master
Details
-------
https://git.postgresql.org/pg/commitdiff/d9d7fe68d35e1e10c7c8276d07f5abf9c477cb13
Modified Files
--------------
src/backend/replication/logical/applyparallelworker.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
On Mon, Feb 6, 2023 at 11:40 PM Amit Kapila <akapila@postgresql.org> wrote:
Use appropriate wait event when sending data in the apply worker.
Currently, we reuse WAIT_EVENT_LOGICAL_PARALLEL_APPLY_STATE_CHANGE in the
apply worker while sending data to the parallel apply worker via a shared
memory queue. This is not appropriate as one won't be able to distinguish
whether the worker is waiting for sending data or for the state change.To patch instead uses the wait event WAIT_EVENT_MQ_SEND which has been
already used in blocking mode while sending data via a shared memory
queue.
This is not right at all. You should invent a new wait state if you're
waiting in a new place.
--
Robert Haas
EDB: http://www.enterprisedb.com
On Thu, Feb 9, 2023 at 7:56 PM Robert Haas <robertmhaas@gmail.com> wrote:
On Mon, Feb 6, 2023 at 11:40 PM Amit Kapila <akapila@postgresql.org> wrote:
Use appropriate wait event when sending data in the apply worker.
Currently, we reuse WAIT_EVENT_LOGICAL_PARALLEL_APPLY_STATE_CHANGE in the
apply worker while sending data to the parallel apply worker via a shared
memory queue. This is not appropriate as one won't be able to distinguish
whether the worker is waiting for sending data or for the state change.To patch instead uses the wait event WAIT_EVENT_MQ_SEND which has been
already used in blocking mode while sending data via a shared memory
queue.This is not right at all. You should invent a new wait state if you're
waiting in a new place.
This is a misunderstanding on my part to reuse the wait_event for a
similar kind of wait but I got your point and will take care of this.
--
With Regards,
Amit Kapila.