can be waited on in shutdown.

Slurmctld treats prolog failure (reported as error code in node_registration
  RPC) as sign to set node DOWN.
Slurmd reports prolog failure on srun step task launch back to slurmctld
  to set node down.

Slurmctld treats prolog failure (reported as error code in node_registration
  RPC) as sign to set node DOWN.

   (probably just an srun detaching)
 o srun: disable stdin reading if in background
         enable proper termination in attach mode

allocation is release only upon termination of the last step.

   - fix some EXTRA_DIST declarations so that CVS/ directories don't get
     inlcuded in a make dist
   - add forced target to rebuild any dependent convenience or api libs
     (if necessary)

ALLOCATED to IDLE.
Save creds to file on each cred revoke and on slurmd shutdown.

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.