How the Delivery of Events Is Guaranteed

There is a total ordering of event generation within the cluster that is preserved in the order of delivery to each client. In other words, if event A is generated within the cluster before event B, then client X receives event A before that client receives event B. However, the total ordering of event delivery to all clients is not preserved. That is, client Y could receive both events A and B before client X receives event A. In this way, slow clients do not hold up delivery to all clients.

All events that the server delivers (except the first event for a subclass and events that follow server errors) occur in response to the actual events that the cluster generates, except if the server experiences an error that causes it to miss cluster-generated events. In this case, the server generates an event for each event type that represents the current state of the system for that type. Each event is sent to clients that registered interest in that event type.

Event delivery follows the “at least once” semantics. That is, the server is allowed to send the same event to a client more than once. This allowance is necessary in cases in which the server goes down temporarily, and when it comes back up, cannot determine if the client has received the latest information.