On Fri, 10 Oct 2003, Vivek Khera wrote:
> >>>>> "sm" == scott marlowe <scott.marlowe> writes:
>
> >> will ungracefully kill the DB process(es). Doesn't matter what DB (or
> >> any other application) you're running, you *can* lose data this way.
>
> sm> While it is possible to lose a non-committed transaction, WAL prevents the
> sm> database from becoming corrupted. Assuming proper fsyncing of your hard
> sm> drives (i.e. SCSI, or IDE with write cache disabled)
>
> So you're saying it is not possible to corrupt the WAL if the process
> is ungracefully killed by the OS?
No, but it doesn't matter if it is corrupted, because the corrupted part
would be at the end, where a transaction was starting, and would just get
ignored. i.e. postgresql would replay only the parts of the WAL that
were complete and showed as committed.