If an activation group has decided that it can go inactive, it sets local state to disallow incoming newInstance calls, and then attempts to notify rmid by calling the inactivateGroup method of the monitor. Suppose a newInstance call comes in after the local state is set but before rmid is notified that the group is inactive. The activation group rejects the newInstance call by throwing an ActivationException, but rmid treats that ActivationException as fatal and rethrows it to the client ActivatableRef, which also treats it as fatal and wraps it in an ActivateFailedException thrown to the client. This is wrong. Somehow rmid needs to know that the problem is a "stale" group, so that it can spawn a new group and call newInstance on it instead.