On 06/10/2018 02:09 PM, Alexander Shutyaev wrote:
> Some more notes on databses.
>
> Although the pg_upgrade failed, I've decided to check the databases in
> the new cluster (10.4). There is no database with oid 0 either. Also to
> be noted that some system databases changed the oids while others
> retained them.
If I am following the source code for pg_upgrade correctly that is
expected. Pretty sure because the order of object creation is different.
>
> And of my databases - sslentry. It had a very big oid (can that seem
> strange?) and its oid has changed.
OID's are added at time of object creation so I would say the ssslentry
database was created some time after the other databases in the 9.6
cluster. Actually probably more accurate to say after 1016305714 -
16400(bof db) objects that have OID's where created.
When the upgrade failed pg_upgrade should have pointed you at an error log.
Did it and is there anything useful there?
From your OP post:
"You might also need to commit or roll back old prepared transactions."
Do you have any of those in the 9.6 cluster?
See:
https://www.postgresql.org/docs/10/static/view-pg-prepared-xacts.html
>
> select oid, datname from pg_database;
> oid | datname
> -------+-----------
> 13011 | template0
> 16400 | bof
> 13012 | postgres
> 16401 | sslentry
> 1 | template1
> (5 rows)
>
>
--
Adrian Klaver
adrian.klaver@aklaver.com