Re: pg_upgrade and wraparound

On 06/10/2018 02:45 PM, Alexander Shutyaev wrote:

Comments inline.

> The error log is like this. Here's its tail:

Is this the regular Postgres log or the pg_upgrade log which should be
something like pg_upgrade_server.log?

> pg_restore: [archiver (db)] could not execute query: ERROR:  database is
> not accepting commands to avoid wraparound data loss in database with OID 0
> HINT:  Stop the postmaster and vacuum that database in single-user mode.

How did you get into the 10 cluster to report on the database OID's and
names?

> You might also need to commit or roll back old prepared transactions.
>     Command was: ALTER LARGE OBJECT 1740737402 OWNER TO bof_user;
>
> Before that there is a lot of similar messages - the only things
> chainging are the "executing BLOB nnn" number and "must be vacuumed
> within nnn transactions" number.
>

Which database has the large objects?

> As for the prepared transactions - no, I don't have them, our
> application doesn't use this functionality.

Did you check this view to confirm?:

https://www.postgresql.org/docs/10/static/view-pg-prepared-xacts.html

Just trying to eliminate possibilities.

--
Adrian Klaver
adrian(dot)klaver(at)aklaver(dot)com