In Red Hat JBoss Data Grid, when the cluster splits into multiple parts (usually due to networking issues such as a crashed router between two sites), JBoss Data Grid can detect this situation and reduce availability (prohibit reads or writes) to keep the data consistent until the cluster is re-connected. If a write to the cache is issued soon after the cluster split is detected, this update may be lost or different nodes may read different versions of the entry. This is a known issue in JBoss Data Grid 6.4 Beta and the workaround for this issue is to avoid updating the cache contents before the cluster topology stabilizes.