Cluster node selection process do not respect whether the nodes are able to handle the EJB request

The problem happen only if a cluster of servers contains nodes with different deployments.
An application or EJB might be missed or deployed only once, this can be the case during application upgrade or HASingleton deployments.

In that situation the ejb-client API does not take into account during the cluster node selection that not all nodes can handle the invocation.
In that case the invocation will fail randomly if such a node is selected.