Andres Freund wrote:
> I'm not entirely clear why pg_restore appears to use a separate
> transaction for each large object, surely exascerbating the problem.
To make sure that per-object locks don't fill up the shared
lock table?
There might be hundreds of thousands of large objects.
If it had to restore N objects per transaction, would it know
how to compute N that is large enough to be effective
and small enough not to exhaust the shared table?
Best regards,
--
Daniel Vérité
PostgreSQL-powered mailer: http://www.manitou-mail.org
Twitter: @DanielVerite