Tom Lane <tgl@sss.pgh.pa.us> writes:
> Maybe you can duplicate the problem with a synthesized or anonymized
> data set? It's unlikely anybody will spend much time on such a vague
> report as this.
I reloaded the same data set overnight, and I could not reproduce the
error (which was present in one in 3.6 million rows). However, I got a
"duplicate key value violates unique constrant" on another table,
without actually duplicating any key value on insert (this was for one
row in 3.3 million).
Because of this inconsistency and the fact that I ran these tests on a
laptop (no ECC), on a hunch I ran memtest86+. And lo and behold:
Faulty memory :-/
> Not the same query, surely? AFAIR the rbtree code is only used in GIN
> indexes at the moment, so this would presumably be related to an update
> of that GIN index. But again, it's hard to expect somebody to go
> looking for this without a more definite way of producing the problem.
Right, this error occured on insert.
The problems I reported were most probably caused by memory errors,
and I apologize for posting a bug report without checking that first.
I also apologize for the vague report; however unfortunately it's
quite difficult to construct a test case if you can't reliably
reproduce the error, and the (rather big) data set which caused it in
the first place cannot be shared.
In any case, I'll try to do my homework better before posting another
bug report :-)
I'd like to use this opportunity to commend you and the other
developers for your excellent work on PostgreSQL!
--
Best regards,
Dag Lem