o add calls to interconnect_node_init/fini() in slurmd.
 o move interconnect_postfini() above wait_for_io()

   waiting for processes once slurmctld closes the connection.

   is not inadvertently killed.
 o slight changed in slurmd "task %d exited ..." debug message

   interconnect_node_fini() to allow for interconnect specific node
   initializations.
 o Elan interconnect module now runs the network error resolver thread
   (requires /etc/elanhosts config file)

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

 o Do not terminate job if job is already terminating, just exit launch thd.

   has been no IO activity since job ended/failed/etc.

specified.

 o do not call fatal() for lsd_fatal_error(), just call error()

recover qsw state saved (if any and if "-c" option not used) and use as
argument to qsw_init(). If no state to be preserved, call qsw_init(NULL)
to initialize data structures.

   permissions are 0755.

Add a quick-load capability to the plugins so that the init() and fini() code aren't invoked if all you want to do is query the plugin type and version.

Previously slurmctld was not filtering out nodes failing to satisfy the
job's mincpus, mem, or tmp constraints.

Add step_id filtering as needed if a user requests info about a step_id
of zero (e.g. 123.0, it used to report all steps for the specified job_id).

   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.

   allocation of a job step.

configuration information downloaded.

html/slurm_overview.html.

job_id, but the full step_id.

   such as MPIR_debug_state.