Add scheduling of federated batch jobs
Sibling jobs have to get lock from the origin cluster in order to attempt to allocate nodes. If it gets the allocation then it lets the origin cluster know and the origin cluster will set the siblings jobs, if any, into a REVOKED state and purge the jobs. If the sibling job is the only sibling then it assumes the lock and attempts to start the job to avoid extra communications. If nodes can't be allocated then the job releases the lock for another cluster to try.
Please register or sign in to comment