Tom Lane wrote:
> I've been thinking more about Gaetano Mendola's report of a
> FlushRelationBuffers failure during VACUUM FULL. I still don't see how
> that could happen in 7.4, but I do see a real related risk in 8.0.
Just another information that could help your analysis.
The index's parent is a table that implement a "materialized view" so is
heavily updated by triggers.
Regards
Gaetano Mendola