Using the two-phase commit protocols, applications use the commit service of the syncpoint manager to make changes permanent, and the backout service to back out the changes. When the syncpoint manager receives a commit request, the syncpoint manager synchronizes commitment of resources.

A unit of recovery is an entity in which a resource manager can express interest, letting the syncpoint manager know that the unit of work is using some protected resources that it manages. The protocol verifies the all or nothing changes, even if the application program, the system, RRS, or a resource manager fails. Figure 1 shows how contexts become a series of units of recovery.

Figure 1. Contexts as a series of units of recovery

A context has only one active unit of recovery at a time. This unit of recovery must complete before another unit of recovery begins. The syncpoint manager knows about the context and tracks resource managers that express interest in a given unit of recovery. When the application invokes the commit service, the syncpoint manager initiates the two-phase commit protocol with the interested resource managers. The syncpoint manager reports the result back to the application.