On Wed, Apr 4, 2018 at 05:29:46PM -0500, Jerry Sievers wrote:
> We have a large >20TB system just pg_upgraded from 9.5 to 9.6 as per the
> versions shown below.
>
> The system does <5M transactions/day based on sum(commit + abort) from
> pg_stat_database.
>
> Autovac is running all possible threads now and upon investigating I see
> that thousands of tables are now above the freeze threshold. Same
> tables all appear ~50M xids older than they did yesterday and the
> upgrade was less than 24 hours ago.
>
> I have a "safety" snap made of the system before upgrade that can be
> used for inspection.
>
> Any ideas why the age jump?
Uh, you can read how pg_upgrade handles frozen xids in pg_upgrade.c:
https://doxygen.postgresql.org/pg__upgrade_8c_source.html#l00543
I am not sure what would have caused such a jump. pg_upgrade brings
over the frozen values for each table, and sets the server's frozen
counter to match the old one.
If you run 'pg_dump --binary-upgrade' you will see the frozen xids being
transfered:
-- For binary upgrade, set heap's relfrozenxid and relminmxid
UPDATE pg_catalog.pg_class
SET relfrozenxid = '558', relminmxid = '1'
WHERE oid = 'public.test'::pg_catalog.regclass;
Is it possible that pg_upgrade used 50M xids while upgrading?
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +