The lack of real persistence in Redis (up until recently, of course) prevented me from using it. If I were to go with something else - Cassandra seems really interesting (I think I saw a cl-thrift somewhere ...). But, elephant/bdb for now ...
BDB has good recovery utilities, but just to be sure, try to keep a journal of all database updates to a flat text file, just in case, even if you push stuff to message queue (beanstalkd is my favorite and I will push my Lisp client for it when I have time to clean it up.)
Nothing like having a flat text file on a remote box, good for your sleep at night.
Hmm, do you mean saving more than the BDB .log files? Basically, I'm backing those up initially - then running a 'catastrophic recovery' (and some sanity chceks) on the data I've backed up. Once the backup checks out, I delete old .log files.
I wasn't familiar with beanstalkd, but it seems interesting. I already have an asynchronous thread-pool based job queue built on the back end that could work well for something like that ...
