srun erroneously expected replies from these hosts (gnats:291)

scalability. An arbitrary number of requests may be queued and they
are processed one per second until the queue is empty or pending
requests were last attempted recently (configuration parameters set
to 60 seconds as a minimum retry interval).

These jobs are reported by slurmd on node registration. They are logged
but otherwise ignored by slurmctld. Several changes to slurmd logging
messaged to report job id and step id using %u format rather than %d
format (which shows no-allocate job id values as negative numbers).

and --nodes options.

--relative option is lower than the node count specified. The --relative
option takes precedence.

that reach the slurm inactivity time limit.

control (it needs to complete all pending RPCs and save state before
the primary reads state and takes over).

responsibilities (backup was routinely reading at startup).

server to shutdown request.

SIGPWR, and SIGLOST.

and when returned to service went improperly back into state DRAINING
(job counter was inconsistent).

transition to DRAINED state.

data will not be used and the process is too slow anyway.

EPILOG_COMPLETE_MESSAGE. At this time the job is COMPLETED and all
associated nodes available.

   on nodes relative to the current allocation.
 o srun no longer sends SIGKILL to job if one task is killed except
   if --no-allocate is used. (the job will otherwise be killed by
   the controller anyway)

   function.

   memory appears to be full.

   write() call.

  - instead of attempting to kill pending threads, immediately
    exit wait_for_procs if a thread is already waiting for job.
  - if wait_for_procs fails (thread already waiting), exit w/out
    sending epilog complete rpc.

type debug.

descriptors. This was needed in several grouped functions (e.g.
slurm_send_recv_rc_msg and slurm_send_only_node_msg, which combine
open, send, receive, and close functions for simplicity).

RPC server threads to that number - 2. This should slightly reduce the
incomming RPC load.

and may have been exhausting virtual memory, resulting in the death of
slurmctld.

the left-over slurmctld process on abnormal termination.

clarity. No change in functionality or logic.

If the "-c" option is not specified then only the jobs and some node
state information will be preserved. Specifically the state of DOWN,
DRAINED, or DRAINING nodes and the associated reason field for those nodes.