# Conflicts:
#	src/common/gres.c

Regression from enhancement 4506.

Bug 6680

Name change was made before 17.11.

Bug 6662

Previously the state reason could remain as WAIT_NO_REASON even after
backfill evaluation. This should improve system feedback/responsiveness
user perception.

Bug 6594.

Continuation of 69d78159

Bug 6500

Bug 6513

This allows this function to print to any log level requested instead of
having to make multiple copies of this function.

Bug 6513

where _agent_defer was throwing away rpcs if _batch_launch_defer() or
_signal_defer() returned 1.

Bug 6704

Caught by test17.31 failing.

Bug 6726.

Issue was from ade9101e.

The problem was an over simplified if statement which when ran multiple
times would set the start time incorrectly.

Bug 6697

Bug 6664.

Even if main scheduler doesn't allocate resources for hetjobs, the
queue list should be composed by all types of jobs. Otherwise, lower
priority regular jobs could be allocated resources by main scheduler
while higher priority hetjobs are waiting for a backfill cycle.

Bug 6593.

CID 196331.

CID 196330.

Fix the initialization through the declaration instead.

Update test to NO_VAL64, and use PRIu64 for for format string.

Not used at this point, so drop them from the sbatch/srun CLI.

Leaving them in struct job_details for now, they may be removed at
a future point if the rest of the checkpoint infrastructure is dropped.

Change mem_per_cpu / mem_per_gpu / pn_min_memory / pn_min_tmp_disk
to uint64_t in slurm_opt_t, and use NO_VAL64 as the sentinel value
throughout.

Rework str_to_mbytes2() to return a uint64_t, and NO_VAL64 as an invalid
value instead of -1.

Not needed for now, may come back at a later time.

Only used as part of an info() message hidden behind the --verbose
option in salloc/sbatch/srun, which isn't worth keeping here as
the rest of the slurm_opt_t structure is being cleaned up.

# Conflicts:
#	NEWS
#	doc/man/man5/slurm.conf.5
#	src/plugins/sched/backfill/backfill.c
#	src/slurmctld/job_scheduler.c

This should help insure that jobs are not prevented from running
at the same time on the same node due to memory limits.

bug 6675

Bug 6708.

when acct_policy_job_runnable_post_select() is call for job without
job_resrcs (eg.: from _attempt_backfill() or when select_nodes() can't
find free recource) but with req_node_bitmap.

This patch will change reason of jobs like in test 21.39 from
"JobState=PENDING Reason=QOSGrpNodeLimit" to
"JobState=PENDING Reason=Resources"

bug 6675