On Wed, Dec 15, 2021 at 12:15:27AM -0300, Martín Fernández wrote:
> The reindex went fine in the primary database and in one of our
> standby. The other standby that we also operate for some reason
> ended up in a state where all transactions were locked by the WAL
> process and the WAL process was not able to make any progress. In
> order to solve this issue we had to move traffic from the “bad”
> standby to the healthy one and then kill all transactions that were
> running in the “bad” standby. After that, replication was able to
> resume successfully.
You are referring to the startup process that replays WAL, right?
Without having an idea about the type of workload your primary and/or
standbys are facing, as well as an idea of the configuration you are
using on both (hot_standby_feedback for one), I have no direct idea,
but that could be a conflict caused by a concurrent vacuum.
Seeing where things got stuck could also be useful, perhaps with a
backtrace of the area where it happens and some information around
it.
> I’m just trying to understand what could have caused this issue. I
> was not able to identify any queries in the standby that would be
> locking the WAL process. Any insight would be more than welcome!
That's not going to be easy without more information, I am afraid.
--
Michael