Peter Mount <peter(at)retep(dot)org(dot)uk> writes:
> Ah, try running first vacuum; then vacuum analyze. The first one will
> truncate the table removing the wasted space. The second simply rebuilds
> some statistics about the table.
This is not right --- AFAICT, vacuum *always* runs the compaction phase.
You can allow or skip the stats-gathering phase by specifying "analyze"
or not, but "vacuum analyze" implies "vacuum".
It sounds like Pablo has a problem with vacuum not doing what it is
supposed to ... has anyone else ever seen that? It's a new one on me.
regards, tom lane