I've noticed today that if one tries to pg_dump a database cluster
running under pgpool, one gets the error message:
pg_dump: query to get table columns failed: ERROR: kind mismatch
between backends
HINT: check data consistency between master and secondary
Looking at the SQL that pg_dump sends to be relying on object OIDs for
the dump. Would it be reasonable at some date to design pg_dump to work
by joining whatever query would get the OID with the query that would
use it as where criteria so that the OID itself is never reported?
Would there be any interest in such changes, were they made, making
their way into the backend?
Does my question even make sense? It's been a long couple of weeks...