If a cluster has been down by using 'shutdown cluster' command the state is saved and the content persisted. To bring that back in a consistent way no cache, no matter whether its persistence is shared or not, should be available until all the cluster nodes are back and the state-transfer is enabled.

This is to prevent from additional state-transfer with shared stores or inconsistencies in case of unshared stores.

Each node should log a warning that it is waiting for all cluster nodes with a message like "wait for cluster startup. Nodes available A,B missing C,D" or similar.

Clients should get an AvailabilityException with a similar message logged.

The solution should apply despite the partition handling strategy.