Nick Rupley <nickr@mirthcorp.com> writes:
> We also tracked down every single duplicate entry and noted the time it got
> inserted. What we found is that this issue only ever started happening
> after we upgraded from 9.0.13 to 9.3.4. The upgrade itself doesn't seem to
> be the culprit, but messages received *after* the upgrade encountered this
> issue.
Oh dear. You were never on any earlier 9.3.x release? If not, then
there's still some unfixed bug in this area in 9.3.4.
There's been one post-9.3.4 fix in this same general area:
http://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=c0bd128c8
but according to the commit message, at least, that bug would not have led
to the symptom you're seeing, namely rows disappearing from indexes while
they're still visible to seqscans.
> Thanks in advance for any input, we're trying to make a self-contained test
> case in the meantime...
A test case would definitely help.
regards, tom lane