On Sat, Oct 13, 2012 at 05:05:08PM +0200, Stephan von Krawczynski wrote: > On Sat, 13 Oct 2012 15:52:56 +0100 > Brian Candler <B.Candler at pobox.com> wrote: > > > In a distributed volume (glusterfs 3.3), files within a directory are > > assigned to a brick by a hash of their filename, correct? > > > > So what happens if you do "mv foo bar"? Does the file get copied to another > > brick? Is this no longer an atomic operation? > > > > Thanks, > > > > Brian. > > In fact it has never been atomic. > Take a look at my corresponding bug report from "back then"... > You can use a small script to show it is not. I hunted around, I couldn't find much in bugzilla apart from https://bugzilla.redhat.com/show_bug.cgi?id=762766 (seems to be more about replicated volumes). I couldn't find any bugs containing "skraw", but I'm probably just driving bugzilla wrongly. I did find this recent posting though: http://permalink.gmane.org/gmane.comp.file-systems.gluster.devel/2888 Is this what you're talking about? This is about an old version of glusterfs but the version used is not specified as far as I can see. Anyway, I think my question is answered. Normally I would create a file as xxx.tmp, sync and then rename it to xxx, to protect against a partially-created file; but for large files this looks like it's not a good approach for glusterfs with a distributed volume. Thanks, Brian.
