Aaron Krowne <akrowne(at)vt(dot)edu> writes:
> So, either it is broken, or doing a VACUUM FULL ANALYZE rather than just
> VACUUM ANALYZE made all the difference. Is this possible (the latter,
> we know the former is possible...)?
If your FSM parameters in postgresql.conf are too small, then plain
vacuums might have failed to keep up with the available free space,
leading to a situation where vacuum full is essential. Did you happen
to notice whether the vacuum full shrunk the database's disk footprint
noticeably?
regards, tom lane