Document the issue from the original bug. QE will provide text.
In Red Hat JBoss Data Grid, after network partition (split-brain) is resolved, nodes start reconnecting through a series of merge events. In some of these merges a node may be reported to temporarily leave the cluster; if such node is executing a transaction spanning other nodes, this transaction is not executed on the remote node. However, the transaction can be confirmed and succeeds on the originating node. The result is stale value on the node not committing this transaction. This inconsistency is not resolved until the entry is updated or removed; reads can return both stale and committed value. After the merge is finished on all nodes, this situation cannot happen any more. There is no workaround for this issue.
I've added a question in the original bug to clarify whether we wanted this information in just docs (this is what I thought) or docs+RN, or just RN. I've added it to the docs here in a warning box: http://docbuilder.usersys.redhat.com/12532/#Detecting_and_Recovering_from_a_Split-Brain_Problem Setting this to martin. Does this satisfy the requirments for that bug, martin?
Thanks, Martin. I'll keep this on ASSIGNED for now then to remind myself that this may still need work.
Verified the latest version.