In slurmctld communication agent, make the thread timeout be the configured
value of MessageTimeout (or 30 seconds, whichever is larger) rather than
30 seconds.

Fix scancel bug which could return an error on attempt to signal a job step.
A simple "scancel 12.3" to signal a specific job step would fail. Adding
another option (say "-i", "--partion=", etc.) would fix this.

before ending the job.

will make it so the slurmctld will not signal the apid's in a batch job.
Instead it relies on the rpc coming from the slurmctld to kill the job to
end things correctly.

Refactor scancel so that all pending jobs are cancelled before starting
cancellation of running jobs. Otherwise they happen in parallel and the
pending jobs can be scheduled on resources as the running jobs are being
cancelled.

Add SchedulerParameters option of sched_min_interval that controls the
minimum time interval between any job scheduling action. The default value
is zero (disabled).
bug 1623

Initialize some variables used with the srun --no-alloc option that may
cause random failures.

job.  This is already handled on the stepd when the script finishes.

often an inventory request is handled.

free afterwards would not have zeroed out memory on the variables that
didn't get unpacked.

sbatch to stop parsing script for "#SBATCH" directives after first command.
It keeps parsing so long as lines contain only white space or comments
(first non-white space character is '#').

Add SchedulerParameters option of "sched_max_job_start="  to limit the
number of jobs that can be started in any single execution of the main
scheduling logic.

Bug 1601

Bug 1603

and update NEWS

the same time, don't double bill the down time.

database.

list, but only the column list.

Prevent slurmdbd error if cluster added or removed while rollup in progress.
Removing a cluster can cause slurmdbd to abort. Adding a cluster can cause
the slurmdbd rollup to hang.

When allocating resources with resolution of sockets, charge the job for a
CPUs on allocated sockets rather than just the CPUs on used cores.

bug 1595

The error was being triggered by logic to collect accounting
information starting and not being completed when testing to
start a node ping RPC. In other words, the logic wasn't testing
for the completion of one RPC against the start of that same
RPC, but against the start of a different RPC. This change
moves all of the timeout logic into the ping_nodes.c module
where we can make sure that the timing of different RPCs do
not get confused with each other.
bug 1190

slurmd was logging:
"Error reading step 2031886.0 memory limits" (with various job/step IDs)
because it was treating SLURM_SUCCESS as an error rather than SLURM_ERROR
bug 1589

Disable changes to GRES count while jobs are running on the node.
Previous logic could clear GRES count data and generate counter
underflow then when attemtping to deallocate the job's resources
(including future job terminations when computing when and where
pending jobs will start as part of backfill scheduling logic).
bug 1589

if many steps/jobs finish at once.

Bug 1587