Re: vacuumdb is failing with NUMBER OF INDEX TUPLES NOT

Denis Braekhus <denis(at)startsiden(dot)no> writes:
> Indicating that they should produce the same results, but that they work
> differently. I am not sure what that implies, but maybe someone else knows ?

The only difference the docs are talking about is what kind of lock is
held while the rebuild proceeds.

A reindex builds a new index file from scratch, and AFAICS should give
the same results as dropping/recreating the index --- at least in terms
of what's in the file proper. The only theory I can come up with for
your experience is that there was some corruption in the system catalog
rows describing the index. That would not get fixed by a reindex.
However, I haven't the foggiest idea what sort of corruption might
allow the index to seem to work (and not, say, crash the reindex itself
which is going to use that information...) yet allow problems to appear
much later on. Too bad the evidence is gone now.

regards, tom lane