> > meaning XA or is the Object PM not even atomic when using > > versionning/locking ? > If jackrabbit with an Object PM crashes while it is storing nodes, only > parts of the nodes might get written. those nodes will still be visible > when jackrabbit is started up again, even though the transaction has not > been completed. I guess fixing this would make a nice patch or > contribution ;) I get the message ;) The serverside just ran a story about a transactional connector to a non-tx filestore, maybe it's worth checking out although the performance hit might invalidate this scenario vs DB
> using a database backed PM with such a scenario, the database would > simply rollback the unfinished transaction. ok Just to sum it up : when using a database-backed PM, jackrabbit (currently) never uses clever SQL queries to access the content model. Ie, everything goes through lucene or UUID. In this case, I can't think of why fs would not be more performant than db. Is this your thinking ? Philippe
