High availability question on re. 5.6

I am running the following cluster:
Node A and C have XtraDB-Cluster-Shared 5.6.20-25.7.888
Node B has XtraDB-Cluster-Shared 5.6.15-25.5.759

If I test failover (shut down) with nodes A or C (shut the node down … add a record … then restart the node) everything replicates as expected.
If I test failover (shutdown) with node B I immediately get a too many connections error and I must clear hosts before I can continue.

The question here is … would this be because node b is an earlier version of XtraDB-Cluster-Shared?
Should I upgrade Node B so it is running the same version as the other two?
When upgrading should I uninstall the current artifact or can I just upgrade to 5.6.20-25 ?

Can you show PXC version for each node show status like 'wsrep_provider%' and select @@version

If you have different wsrep_provider_versions this cwould affect replication and consistency of the cluster.

pxc 5.6.15 uses galera replicator 3.4 and wsrep API 25.5
pxc 5.6.20 uses galera replicator 3.7 and wsrep API 25.7
pxc 5.6.22 uses galera replicator 3.9 and wsrep API 25.8

It is recommended to have the same version of PXC across all nodes of the same cluster. Upgrading to latest PXC version would be ideal.