Re: DISTINCT is not quite distinct

* Tom Lane:

>> According to EXPLAIN, an index scan on the bar column is used (using
>> the underlying B-tree index).
>
> Do you mean an indexscan followed immediately by a Unique node? If
> so, yeah, that would depend entirely on correct ordering of the
> indexscan output to produce distinct results.

Yes.

>> If I drop the DISTINCT, the output is not correctly ordered, either.
>> Perhaps this is an index corruption issue? The hardware itself seems
>> fine.
>
> Perhaps. Do you want to save off a physical copy of the index and then
> try REINDEXing?

The duplicate row is gone.

> If that fixes it, I'd be interested to compare the two versions of
> the index.

The index files are about 155 MB and 98 MB, compressed. How shall we
transfer them? (Their contents is not super-secret, but I don't want
to distribute them widely, either.)

--
Florian Weimer <fweimer(at)bfk(dot)de>
BFK edv-consulting GmbH http://www.bfk.de/
Kriegsstraße 100 tel: +49-721-96201-1
D-76133 Karlsruhe fax: +49-721-96201-99