Re: ALTER DATABASE SET TABLESPACE vs crash safety

Tom Lane wrote:
> Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> > Do we need to fsync the directory itself? My fsync(2) manpage says
>
> > Calling fsync() does not necessarily ensure that the entry in the directory
> > containing the file has also reached disk. For that an explicit fsync() on a
> > file descriptor for the directory is also needed.
>
> Hmm ... I see that in the Linux manpage, but not on Darwin, HPUX, or in
> the Single Unix Spec. I'm inclined to argue that we've always expected
> the filesystem to take care of its own metadata, and we've never seen
> any indication that that's unsafe. We don't try to "fsync the
> directory" after a normal table create for instance.

I dimly recall the Postfix guys got burned by this some time ago (mails
got lost after a crash because they didn't fsync the directory on which
they had just created the files before acknowledging the email delivery
to the remote server). I guess this is the reason we require a
filesystem that journals metadata.

http://osdir.com/ml/file-systems.reiserfs.general/2003-09/msg00120.html

--
Alvaro Herrera http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support