In Stash Data Center 3.11.0 a hidden hibernate.cache.mode property allowed selection of LOCAL Hibernate L2 cache mode instead of DISTRIBUTED (the default). In Bitbucket Data Center 4.0, LOCAL Hibernate L2 cache mode became the default.

If a cluster goes into a "split brain" state, where nodes' connections to each other are lost (e.g., through network interruption or software problem such as excessively long JVM GC pauses) but they are still accessing the shared DB and filesystem and processing requests, there is a risk that the L2 caches may become inconsistent. If the "split brain" situation later heals, this inconsistency does not repair itself and may persist indefinitely. Different nodes may display different state of pull requests, users, groups, permissions, and so on; and updates may be lost or encounter spurious "You are attempting to modify a pull request based on out-of-date information" errors.