On Tue, Feb 20, 2018 at 6:27 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Claudio Freire <klaussfreire@gmail.com> writes:
>> I'm not 100% sure this is a pg_upgrade bug or a pg_dump
>> --binary-upgrade one, or some other thing, but at this point I'm
>> fairly certain there's something wrong in one of them.
...
> (2) independently of that, it sounds like REFRESH
> MATERIALIZED VIEW CONCURRENTLY is somehow preventing advancement of the
> matview's relfrozenxid in the source DB.
Not necessarily. I have vacuum_table_freeze_max_age set to 350M, so
it's not yet due for freezing.
>> I just tried to pg_upgrade a database from 9.5 to 10.2. I took a
>> snapshot off a replica, promoted it, and then did the pg_upgrade there
>> (to avoid breaking our production server).
>
> And that brings replication behavior into the mix, too :-(. I'd
> suggest seeing if you can duplicate these problems without any
> replication involved.
Indeed, I'll try.