Currently when a node failure occurs on a job with a semaphore it is not getting released properly. This is only recoverable by a scheduler restart. Change-Id: Ifa463824f4a394e015a6ee11fcd51bee163492f8