Re: Corrupt indexes on slave when using pg_bulkload on master

On Wed, Dec 5, 2012 at 5:17 AM, James Cowell <jcowell@btinternet.com> wrote:
> I'm using pg_bulkload to load large amounts of CSV data into a postgres
> database hourly.
>
> This database is replicated to a second node.
>
> Whenever a bulk load happens the indexes on the updated tables on the
> secondary node corrupt and are unusable until a reindex is run on the
> primary node.  I get the error below on node 2:
>
> ERROR: index "tablename" contains unexpected zero page at block 0
> SQL state: XX002
> Hint: Please REINDEX it.
>
> I'm assuming that this is because of the way pg_bulkload builds the index on
> the primary, and possibly has something to do with the way pg_bulkload
> overwrites rows in the event of a constraint violation,

If there are no constraint violations, do you still see the problem?

> but at the same time
> if something works on the primary shouldn't the replicated node be able to
> process the WAL log?
>
> I've tried this on 9.1.6 and 9.1.5 on RHEL 6.3 with pg_bulkload build
> 3.1.1-1.pg91.rhel6 and it happens every time.

Were there any older version on which it worked?  Can you post a
minimal schema and control file to reproduce the problem?

Cheers,

Jeff