Hi,
On Wed, Mar 05, 2025 at 11:05:48AM +0900, Michael Paquier wrote:
> Hi all,
>
> While working on I/O statistics, I have noticed that it is not
> complicated to break the set of rows returned by pg_stat_io if one is
> not careful when updating pgstat_tracks_io_object().
>
> Attached is a patch that I've found useful as a sanity check,
> returning all the combinations supported for BackendType, IOContext
> and IOObject, so as it is easily possible to evaluate if the
> information returned is relevant.
>
> Thoughts?
That would mean changing the test each time pgstat_tracks_io_object() is
modified in such a way that the output is changed. That's a good thing as
the writer will need to double check if the new output makes sense according
to his changes. So I don't see any reason not to add this test.
+SELECT backend_type, object, context FROM pg_stat_io ORDER BY 1, 2, 3;
What about adding some extra paranoia like?
SELECT backend_type, object, context FROM pg_stat_io ORDER BY backend_type, object, context COLLATE "C";
Regards,
--
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com
