On Tue, 2004-08-10 at 19:30, lec wrote:
> Chris Travers wrote:
> The first hardware that got replaced was the SCSI controller. After
> that there were still hardware "hanging" and eventually the system
> engineers replaced the whole server and they weren't sure what the fault
> was. The lost transactions happened not on the first server hang. I'm
> more curious to know why the transactions in the middle got lost but the
> last transaction was there.
With questionable hardware (failing memory, scsi / raid controller,
etc.) all failure modes are pretty much possible. PostgreSQL is good,
but it can't make up for bad hardware.