job step.
fed driver waits for windows to become unloaded before attempting to load
the network table.

checkpoints. It sends an abort signal on timeout.

in order to avoid confusion with slurm partitions.

we know better.

Check _allocate_windows' return code.

Remove maximum processes limitation.

Stick the federation key in the ntbl job desc string.

node when using sn_all.
sn_single is not yet supported.

cleaned up env code once again.  all env for jobs happens in slurmd/smgr.c setup for env is done within slurmd/mgr.c to new variable in slurmd_job_t *job ->envtp which is an env_t *.  All variables are setup there except for when srun runs the script instead of slurmd.  For the srun case things are the same as before.

Andy Riebs/HP.

tolerance, and request timeouts are needed.