NunoACHenriques <nach@fct.unl.pt> writes:
> Neither I believe it because the machine is well tested (including a
> 24h memtest). But there is something I can't get of my mind: once a day my
> app "forces" PG to "play" with some 3GB of disk data in a ext2 fs. It is
> known that sometimes ext2 corrupts data...
Unless you've got things set up so that the temporary files created for
the sorting step are in the ext2 partition, this doesn't seem like it
could be the source of the problem. A more plausible theory is that a
segment of main RAM is bad, but you happen not to use that part of RAM
except under heavy load (ie, while running this daily batch job).
Or it could just be a bug. If you can get a reproducible test case I'll
be happy to dig into it.
regards, tom lane