On Tue, 5 Dec 2000, Leandro Fanzone wrote:
> I'm having problems with transactions, lately. I'm checking
> power-failure conditions on my psql-based programs, and found that a
> simple loop with
>
> BEGIN TRANSACTION
> INSERT INTO test VALUES('any #n');
> COMMIT TRANSACTION
>
> leaves me with no records inserted at all after an abrupt reset on the
> server. The loop is outside the transaction, id est, the BEGIN/COMMIT is
> executed every time. Is there any way to assure this, I mean, to
> actually write the records and to not rely on the cache to do it?
> Because I think it is somewhere floating in the Linux cache and it's
> never wrote physically, and when the power is down, everything is lost.
> Thank you in advance,
Do you accidently pass the "-F" option to the postgres backends? This will
disable fsync (and boost performance). Remove the "-o -F" option to
postmaster from your start-up script and everything should be fine.
You will still experience data loss if you do not use a journalling
filesystem; get the ext3 patches or try reiserfs.
cu,
Helge
--
Hi! I'm a .signature virus! Put me into your .signature and help me spread!
% rm * .o
rm: cannot remove '.o': No such file or directory