Tobias Meyer <t9m@qad.com> writes:
> I was able to correlate the transaction ids (xmin) of the duplicate rows in
> pg_largeobject_meta with times stored in another table, and found the
> duplicates were all added on one specific day, while the originals nicely
> spread out over time. I was further able to trace that date back to a
> cluster upgrade (pg_upgradecluster) we performed with that machine, so I
> guess we have a pretty good candidate. After the upgrade, no more
> duplicates were produced.
Interesting. Can you nail down the software versions that were in
use here? That'd be the old PG server version upgraded from, the
new server version upgraded to, the versions of pg_upgrade and
pg_dump (these probably should match the new server version, but
I'm not certain we enforce that), and the pg_upgradecluster
script version?
regards, tom lane