Re: Monitoring gaps in XLogWalRcvWrite() for the WAL receiver

On Wed, Mar 05, 2025 at 08:04:44AM +0000, Bertrand Drouvot wrote:
> On Wed, Mar 05, 2025 at 12:35:26PM +0900, Michael Paquier wrote:
>> Perhaps there's a point in backpatching a portion of what's in the
>> attached patch (the wait event?), but I am not planning to bother much
>> with the stable branches based on the lack of complaints.
>
> We're not emitting some statistics, so I think that it's hard for users to
> complain about something they don't/can't see.

Hmm, not exactly actually. I've missed that ff99918c625a mentions
that WAL receiver was discarded on purpose, but this was still when
pgstats was not in shared memory and still file-based. We scale much
better now. It is not that difficult to make XLogWrite() hot enough
that it would trigger a lot of calls of pgstat_count_io_op_time() per
ms, either, like the WAL receiver, so as long as the timings are
behind track_wal_io_timing we're fine.

Let's do this at the end, without a backpatch. At least we'll be anle
to get better IO metrics for replication setups.
--
Michael