Re: Lost rows/data corruption?

"Andrew Hall" <temp02@bluereef.com.au> writes:
> We have a long running DB application using PG7.4.6. We do a VACUUM FULL
> every night and a normal 'maintenance' VACUUM every hour. We do nothing with
> any indexes. Every now and then we get errors from the database whereby an
> update will fail on a table saying that there is duplicate violation on a
> primary keyed row. Theoretically this is impossible as the constraint should
> prevent any duplication, ever. When the next vacuum is run, we see an error
> saying that there are more rows than the database can account for (or words
> to that effect) and we should either REINDEX or TRUNCATE the table.

Could we see the exact log trace, rather than handwaving?  Also the
schemas of the table(s)/index(es) involved might be useful.

            regards, tom lane