Before I add a second level cache to my product, I'm trying to understand the implications completely. At the moment, I want to understand more details about possible race conditions. Without JTA, it seems that the safest concurrency strategy available is "read/write". With this strategy, given a scenario of concurrent writes, is it possible that a race condition could lead to one server in the cluster having stale data? If not, how is the situation of out of order execution handled, in which the database writes and the invalidation notifications happen in a different order? Does the cache provider affect the outcome?

Thank you for the information.

Scott

[b]Hibernate version:[/b] 2.1.8 and above