On Fri, Sep 24, 2021 at 10:42:36AM -0300, Alvaro Herrera wrote:
> I think the most reasonable action is to push the patch in
> https://postgr.es/m/20201123193957.GA21810@alvherre.pgsql to all
> branches, closing the immediate hole, and we can see about the xact-hook
> stuff (when we have it) to master only, which I view as a better
> mechanism to protect work-items going forward. That's what I understand
> Justin was suggesting in his last reply.
I suggested to use the 20 line solution rather than invent new infrastructure
just to avoid corner case errors.
But I also don't understand why changing to use an INFO message doesn't work.
vacuum_open_relation() outputs a WARNING for manual vacuum and an INFO for
autovacuum when log_min_duration >= 0. Why couldn't this do the same thing?
I found a few instances of this in my logs, indicating that my understanding of
the logic is not wrong.
log_time | 2021-01-13 15:23:40.687-05
session_line | 2
session_start_time | 2021-01-13 15:23:19-05
error_severity | LOG
sql_state_code | 42P01
message | skipping vacuum of "pg_toast_984781820" --- relation no longer exists
backend_type | autovacuum worker
--
Justin