On Wed, 12 Jan 2000, Mark Peglow wrote:
> I tried this, but unfortunately it did not do the trick.  What I am seeing
> is
> corruption in the cyrus.index and cyrus.cache files when they are written by
> deliver and imapd.  The cyrus reconstruct works (and corrects the mailboxes)
> because it is just doing file I/O operations, and not memory mapping the
> index,
> as does deliver and imapd.  I am working on decoding the cyrus structures to
> 
> see if I can find where the writes/reads are going wrong. 
> 

  Apparently Cyrus depends on file locking to prevent such disasters from
happening. AFS and Coda do not provide this - they both use session
semantics (update on close) which somewhat reduces the usefulness of
file locking, and the number of programs that need it are are small
compared to the number that will benefit from the better caching
session semantics allow. Not to mention the trouble disconnection would 
cause. The downside is that programs that need locking (mostly databases
or similar) will have to do it outside of the file system. 

  So, unfortunately, your setup probably won't work. The Cyrus team
was looking at replication, but I don't know what exacly they're planning.

Good luck
Shafeeq