Rafael Martinez Guerrero <r.m.guerrero@usit.uio.no> writes:
> To me this looks like a cache somewhere that upon read contained some
> incorrect data. This cache was somehow flushed two-hours later, and
> fresh data was read from disk.
Very possible.
> Could this be postgres problem, or is it hardware/kernel related?
I'd wonder about bad RAM in the address range that the kernel uses for
disk block caching. Or we could return to the theory I suggested a
bit ago: once in a while the disk controller messes up the last byte
of a transfer.
> The problems started about a week ago after stable production for
> about 9 months.
If you hadn't changed any software immediately beforehand, then that
sure sounds like hardware going bad.
regards, tom lane
