> Later versions of BerkeleyDB have a row-level lock available which works > pretty transparently.
That's what I'm talking about. Multiple writers can cause deadlocks unless you run their deadlock daemon or implement their deadlock handling API yourself. > However, this works using mmap() so it won't work if > your BerkeleyDB file(s) is/are mounted via NFS. Cluster-wide storage is a whole different animal. I'd probably lean toward a database with network capabilities at that point. Joshua Chamas has had some success with Samba though. - Perrin
