Re: BUG #15309: ERROR: catalog is missing 1 attribute(s) for relid 760676 when max_parallel_maintenance_workers > 0

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2018-Aug-06, Peter Geoghegan wrote:
>> Sure enough, that's what the bug is - a few debugging calls to
>> RelationMapFilenodeToOid() within nbtsort.c proves it.

> Uh, that's weird, isn't it?  I mean, why is the relfilenode changing
> underneath?

Because we're building a brand new index in a new file.

> Why isn't it blocked by the CREATE INDEX?  Or is CREATE
> INDEX inflicting that upon itself?

The problem is (or so I assume) failure to propagate process-local
state of relmapper.c into the worker processes.  So the leader knows
that we've assigned a new relfilenode to some mapped index, but the
workers don't.

            regards, tom lane