Andrew Sullivan <ajs@crankycanuck.ca> writes:
> Dunno if this is any help, but on a 7.2 system I saw a REINDEX which
> was interrupted leave the index at least partially working. We ended
> up with an index which seemed fine, but which didn't contain certain
> rows (so those rows were not visible when the SELECT criterion was
> the indexed field). This was extremely puzzling, but a DROP INDEX;
> CREATE INDEX pair solved it.
Hmm, I would expect that behavior for an overwrite-in-place REINDEX,
but 7.2 only seems to use overwrite-in-place for critical system
catalogs. What were you reindexing exactly? Were you running a
standalone backend?
regards, tom lane