In case of a cluster split into 2 partitions (i.e. 2+2) where both sides will go in DEGRADED mode (DENY_READ_WRITES) the behavior for the user is confusing.

Start a new node, which try to join the cluster will fail with a state-transfer timeout , which is unexpected and leads to confusion.
It would be better if the node can finish the state-transfer and log a WARN message that the ST will be delayed, the node is not functional because of the DEGRADED state

The same will happen if one of the nodes in DEGRADED mode will be restarted for some reason, also if the machine driving the instances has crashed.
In this case the nodes are not considered as the members of the last stable cluster-view and will join like every other new node.

Logging and behaviour should be enhanced that

a) new nodes will not fail but show DEGRADED mode and log a WARN message
b) as soon as the cluster becomes AVAILABLE the nodes are merged into the cluster