Ensure server discovery and monitoring events are published in order

[jyemin]

• Ensure events are ordered according to how they are applied to driver state
• Publish all events to application listeners one at a time, on a separate thread

JAVA-4449

[jyemin]

Simplified based on in-person conversation. PTAL.

[stIncMale]

I like the new approach for decorating listeners 👍

[optional]
I propose to assert that there is exactly 1 listener in the constructor of BaseCluster and then just use it, instead of calling getClusterListener(settings). We know that it must be the case for all clusters and servers because of how DefaultClusterFactory works.

If you agree, the same can be done also in

• LoadBalancedCluster
• DefaultClusterableServerFactory
• LoadBalancedClusterableServerFactory
• DefaultServerMonitor

As a result, the only place where EventListenerHelper.getClusterListener/getServerListener/getServerMonitorListener are called will be DefaultClusterFactory. Additionally, we could even move the methods there and make them private to avoid using them accidentally in a new server/cluster/factory implementation.

[jyemin]

Since the vast number of clients never register a listener, avoiding thread/queue creation for that case.

[jyemin]

Small behavioral change here. Using multicaster when there is one listener registered, because the muilticaster will log exceptions thrown by the listeners, which should happen in all cases. Didn't seem worth it to add a non-multicasting exception-logging implementation.