The scenario we encountered on EAC is:
1. A node gets kicked from the cluster
2. The node then rejoins. IMap states are merged
3. The merging process gets the cluster safety number out of sync with the database
4. (boom), cluster panic. Even worse is the fact that this can get the IMap in a bad state where all nodes panic
To avoid this, on a cluster merge, we should have the cluster safety number map clear the stored entry. That way it will run a fresh cluster safety check without the (potentially) polluted value
For the same reason this IMap should not have backups.