If an asynchronous command fails at a remote node, it should be retried.

I'm not sure how feasible this really is. One possible solution could be this: having NACK style implementation where by default the originator assumes an asynchronous command has been executed, but if the receiver tells it that the topology is outdated, the originator retries?

This is related to ISPN-8027 where we've discovered that some updates are not applied when asynchronous commands to update the Hibernate 2L timestamp cache fail as a result of an outdated topology.