Re: VACUUM FULL versus system catalog cache invalidation

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> With HOT, there is very little need to perform a VACUUM FULL on any
> shared catalog table. Look at the indexes...

This is not really a useful argument. People do do VAC FULL on
catalogs, whether we think they should or not. Also, it's not only
"shared" catalogs that are at stake.

We could avoid fixing the bug if we forbade both VAC FULL and CLUSTER
on all system catalogs, period, no exceptions. That doesn't seem like
a workable answer though. Some usage patterns do see bloat on the
catalogs, especially if you disable or dial down autovacuum.

> In the unlikely event we do actually have to VACUUM FULL a shared
> catalog table, nuke any cache entry for the whole shared catalog.

You're still not getting the point. We *do* nuke all cache entries
after a VAC FULL. That does not avoid this bug.

regards, tom lane