failed (presumably due to unkillable processes)
 o retry failed JOB_KILL rpcs

error).

RPC in agent.

   instead of just kill_wait seconds because slurmd sleeps for kill_wait
   seconds, so therefore slurmctld would never recv a reply.

   slurm_send_recv_node_msg(), slurm_send_recv_rc_msg(), etc.
 o Fixed fd leak in agent.c using slurm_send_recv_rc_msg() w/ timeout.

reference if a reconfigure RPC was active at the same time.

This would occur for example if we killed a job that never started any
steps on a node.

improved performance in production.

slurmctld sends REQUEST_KILL_TIMELIMIT when job reaches its time limit,
change job state to TIMEOUT | COMPLETING and node state to COMPLETING.

Add COMPLETING node and job states.
Major restructuring of slurmctld node and job state transition code.

its nodes are put in state COMPLETING (used to be IDLE) and a revoke
credential RPC is issued to all slurmd's. When the slurmd's respond
(after running the epilog and confirming that all tasks are completed),
the node's state is changed to IDLE and other jobs can run.

Initialize/free/set alloc_sid and alloc_node in API functions.
Pack/unpack/free new elements in job descriptor RPCs.
Load/dump/pack new elements into job table records.

Add support in slurmctld for SlurmctldDebug on startup and reconfig RPC.

Slurmctld add node list to job step (leave bitmap too, faster)	Jette
Slurmctld fix node state when job keeps running and node fails	Jette
Slurmctld job requests with min-max resource requirement pair	Jette
Slurmctld job requests with excluded node list (--exclude)	Jette

job time limit. Defined pack/unpack/free functions for RPC.
Slurmctld modified to broadcast REQUEST_UPDATE_JOB_TIME RPCs.

Slurmctld priority of zero treated like job hold (not scheduled).
Preserve priority order even if incomming job could use idle resources.

slurmctld now will ping slurmd periodically and flag non-responsive nodes as such.

Expanded error recovery logic on slurmd registration.

Terminate job if slurmd restart without it.