On Tue, Apr 7, 2009 at 5:46 PM, Kevin Fox <Kevin.Fox@xxxxxxx> wrote:
> Not exactly. I think the whole ext4 uber thread on
> create/write/rename/no fsync thing showed that there are cases when that
> can have bad results. I think your pretty safe doing it with ext3 (but
> there was some mention of a 5 second window that may be a problem).
i haven't tested ext4 yet; but some of those issues are shrugged off
with a "XFS does that since forever". and in the end all the blame is
to the application, and rsync is the best behaved copy i've seen.
i've done several disaster tests with both 'cp -a' and rsync, using
ext3, XFS and JFS, both with local and AoE disks, pulling the plug on
the initiator, the target and the switches.
ext3 is the most resilient, by far. the journal rollback (with
data=ordered) goes a long way in the past to guarantee only correctly
copied files are there.
cp loses data on all three filesystems
rsync never loses data on ext3, and only one file on ~3% cases on XFS
jfs always loses thousands of files.
so, i prefer ext3, and if need more than 8TB in one filesystem, i use
XFS without real worries. always copying with rsync, of course. it
seems that cp does the 'braindead' (quoting Linus) method.
i guess ext4 will be like XFS, but with data=ordered, it should be a
little like ext3. in any case, rsync does it correctly.
Xen-users mailing list