The best solution that I have found is kick all connections and execute a select for update to /dev/null in the affected tables, i.e.:
psql -o /dev/null -c "select * from table for update" database
After this, the vacuum is executed w/o problems again.
Best regards,
Alexandre
On 11/5/18 5:41 PM, Adrien NAYRAT wrote:
> We "solved" with a dump/restore on another server, also we kept previous
> cluster to investigate. I want to be sure we encountered the bug solved
> in 9.6.9 and it is not a new one.
FYI, I am not sure I will be able to keep previous cluster many days.
>
> If we confirm it, maybe we should advise users to perform integrity
> check? I was surprised 9.6.9 avoid new appearance of corruption but
> nothing to ensure if database is already corrupted.
If I understand correctly, it seems related to vacuum skipping page
already frozen. Maybe we should advise to do a VACUUM FREEZE with
DISABLE_PAGE_SKIPPING option?
Regards,