Reliable DMAPI destroy events

A metadata destroy event is generated when the operating system has destroyed an object. This type of event is different from a remove event, which is a namespace event and is not related to the destruction of an object. A reliable destroy event supports synchronous destroy events in the same way that other synchronous events do. When a synchronous event is generated, a user process is suspended in the kernel. The process will be suspended until a DM application issues an explicit response to the event. The DM application at the session that supports the reliable destroy event must be capable of handling the synchronous destroy event. In other words, it must respond to the DM_EVENT_DESTROY event with DM_RESPOND_EVENT. Otherwise, the event will wait forever at the session node for a response. Based on this, it is recommended that the cluster not be made up of nodes that are running back-level code and new code, because the destroy event is not reliable in a mixed environment.