request.

the backup controller and proc_req.c is the code to process incomming
RPCs. No changes in controller logic were made for this.
job_mgr.c was also modified to better handle bad job records on
controller restart's data recovery.

  - Enhanced xsignal.[ch] to more easily block and unblock sets of signals
    [ added: xsignal_unblock xsignal_sigset_create, changed: xsignal_unblock ]
  - Take advantage of above changes in srun and slurmd code
  - Only unblock signals that were specifically blocked in slurmd
    (instead of *all* signals) when exec'ing user tasks.

 o allow error() to return SLURM_ERROR so callers can do:
     return error (...);

state. Bug was resulting in update of node going to DRAINING instead
of DRAINED from DOWN state.

large job specifications

function.

particularly for the get_info RPCs.

 o slight code format improvements

change required by init.d.slurm (bash didn't like names containing "-".
Added slurm_tv_clean support to init.d.slurm.

change required by init.d.slurm (bash didn't like names containing "-".

the configuration file.

with other slurm commands.

could result in a node going from state DRAINED to ALLOCATED to COMPLETING
to IDLE (losing the DRAIN state).

return_to_service flag set.

error reported via agent.

Leave in state DRAIN*.

Explicitly kill any daemon still running when a new daemon is initiated.

common/read_config.h. The values of PIDFILEs were renamed to differentiate
between the files for slurmd and slurmctld. Their default values are
set in read_config:validate_config().