Re: REINDEX INDEX results in a crash for an index of pg_class since 9.6

Andres Freund <andres(at)anarazel(dot)de> writes:
> I see the bug. Turns out we need to figure out another way to solve the
> assertion triggered by doing catalog updates within
> RelationSetNewRelfilenode() - we can't just move the
> SetReindexProcessing() before it. When CCA is enabled, the
> CommandCounterIncrement() near the tail of RelationSetNewRelfilenode()
> triggers a rebuild of the catalog entries - but without the
> SetReindexProcessing() those scans will try to use the index currently
> being rebuilt.

Yeah. I think what this demonstrates is that REINDEX INDEX has to have
RelationSetIndexList logic similar to what REINDEX TABLE has, to control
which indexes get updated when while we're rebuilding an index of
pg_class. In hindsight that seems glaringly obvious ... I wonder how we
missed that when we built that infrastructure for REINDEX TABLE?

I'm pretty sure that infrastructure is my fault, so I'll take a
whack at fixing this.

Did you figure out why this doesn't also happen in HEAD?

regards, tom lane