Craig Ringer <craig@postnewspapers.com.au> writes:
> [ much good info snipped ]
> The only thing that's likely to make things go pear-shaped is if the
> postmaster is forcibly killed ( kill -9 ) while the worker children are
> running, then the postmaster is re-started.
Actually, even that doesn't cause problems in itself. The new
postmaster will refuse to start up until all the old child processes are
gone, but there's no risk of data corruption. The way you can seriously
break things is if you try to make the new postmaster start by removing
the lock file that ensures this behavior :-(
> I don't know if writes to temp tables go through the WAL or not, nor if
> they're opened O_SYNC for synchronous writes.
No for both.
regards, tom lane