> While searching I found coda. I thought that it might help me but I am
> not sure of that.

	I was going to use CODA for this exact purpose (Maildir files).
My conclusion was that it was too complex, poorly documented, and
administratively-intensive to compete with other systems out there like
rsync and unison.

	I also looked at nbd and enbd (Network Block Device, which can be
set up to do RAID1 over a network connection.)  It was too risky; sending
block operations acrossed the network does not sound very fault-tolerant
to me.  enbd (Enhanced nbd) seemed to address some of the tolerance issues
but was also poorly documented and difficult to deploy.

	SANs were too expensive for my project.  There was also IBM's
mirroring software but I can't remember the name of it... ALSA?  Hrm.  I
found the IBM stuff to be basically the same as CODA but without as many
cool features (it's only one-way, has no offline cache, etc.)  On the flip
side the IBM thing was very well documented.

	I don't mean to sound negative on CODA, it's supposed to be
constructive criticism.  Note that I found examples of successful
deployments for all of the above technologies, so they CAN be made to work
if you invest into the learning curve and build some comprehensive unit
tests.

	I'm now using some custom C code that uses Linux's DNotify kernel
feature (very much like SGI's File Alteration Monitor, or FAM) to trigger
a GNU "unison" sync on individual files as they come in.  I don't have
long-term test results yet but it looks promising.  I originally used
rsync as the mirroring program but you then wind up with a race condition
where files can get lost if the sync process gets confused as to who the
"master" copy is.  Unison handles this case for you by maintaining a "last
known state" database between all nodes in the cluster.

	Try the same thing you were doing with rsync, but with unison.
Or else, talk to the rsync mailing list about your problem and get the
file list bug fixed.  (Are you using the very very latest rsync?  It has
some important fixes.)

	Good luck.


Thanks,
Derek Simkowiak