when the job does not exist).

it is a duplicate record.

 o check for a job step state of STARTED before issuing kill_job rpc

was only going to 65500 for the job_id and the step_id was always zero.
This change does not elimiate the possibility of an error, but reduces
its probability by a factor of about 65000.
(gnats:276)

to job_kill request and slurmctld leaves node and job in COMPLETING state
until the slurmd issues an EPILOG_COMPLETE RPC on each node. This permits
better support for non-killable processes and/or long-running epilog scripts.
Several minor changes in node registration handling and slurmctld agent
logic to better address a flood of incomming RPC (typically when system
restarts).
(gnats:268)

gnats slurm/270.

   may result in multiple executions of system epilog for a single job
   (gnats:267)

of socket communications. Previously was sometimes overwriting legitimate
SLURM error code with fcntl error code of EINTR.

sort of slurm error.

on a job kill. Let the KILL_JOB RPC do all of the cleanup. This removes
a redundant RPC. - Moe

occur naturally if a srun, scontrol, scancel, sinfo, or squeue command is
killed by the user with a communication to slurmctld in progress. This
seems to occur fairly regularly as part of batch job termination.

send/receive, function (poll, timeout, send, recv, etc), and the error message
are all reported.

resources to satisfy the request).

This prevents an orphan job if srun dies after sending the request
or the network fails or the authenticaion mechanism fails.

logging the event and returning SLURM_SUCCESS.

It was picking zero nodes and failing.

   problem when debugging remote tasks. (and error should have only
   printed once anyway)

   not SIGXCPU on reaching timelimit.

It was formerly just killing the job step processes.

to match that of a job's run time (TIME).

hand if race condition starting all daemons).