ALLOCATED to IDLE

slurmctld accepts step kill RPC and forwards the signals to slurmd(s).
slurmd processes SIGKILL, SIGSTOP, and SIGCONT via  shm_signal_step(),
all others use killpg(). All signals used to go through killpg(), which
could kill the step specific slurmd and hose the job.

slurmctld accepts step kill RPC and forwards the signals to slurmd(s).

 o Try harder to clean up shm semaphore in slurmd
 o call daemon(1,0) instead of daemon(0,0) in slurmctld to preserve chdir()

 o Only terminate IO thread if IO is done or job is in FORCETERM state

   (may be a residual from cbuf operations)

 o launch.c : small cleanup, make sure return code is preserved

slurm_cancel_job RPC now takes signal argument.
slurmctld takes the signal number on slurm_kill RPC, but does nothing with it yet.

slurm_cancel_job RPC now takes signal argument.

   for scontrol
 o provide featureless readline alternative in case GNU readline is not
   available

cancel entire job).

steps are currently assigned to that node.

pthread_kill(tid, SIGHUP) to knock loose the hung poll()

   prolog is run as root and failure is reported as a non-zero return code
   from the launch tasks or launch batch job messages
   epilog is run as result of revoke credential message

   absolute pathname

Modify slurmd to dump/restore credentials at termination and restart.