Previous logic initialized a static variable once and forever. This
variable was used as a test to call or not log_job_resources and
dynamic changes through 'scontrol setdebugflags [+|-]SelectType' would
have no effect. New logic uses the standard select_debug_flags as
everywhere else and aligns with cons_tres logic.

Bug 7137.

Commit 2194edf3 was intended to change the default fairshare algorithm
to Fair Tree. There was an oversight with the conversion where Fair Tree
wouldn't be the default if PriorityFlags was commented out.

Bug 7154.

bug 7066

This is a continuation of commit edd09b55.

Bug 7071.

The init() function creates a separate thread that tries to index the
list of unindexed jobs (loaded from the state or newly finished) into
the elasticsearch server. This fix prevents this thread from trying to
do so until after slurm_jobcomp_set_location sets the log_url.

Bug 7153.

Also added TimeEligible as a field that can be requested.
Bug 7074

When fixing runaway jobs, submit time should always be set in the
database, so use that when rerolling usage. Avoids rerolling from 0.
Bug 7074

If bf_min_prio_reserve was not the last option listed, the setting would
not have been applied correctly, and an error message printed instead.

Bug 7242.

Bug 6221

Bug 7204

Bug 6221

Bug 7179.

Bug 6847
Bug 7190

For heterogeneous jobs, do not count the each component against the QOS or
association job limit multiple times.

bug 7190

Bug 6847

JobHeldUser state reason was properly set for user held jobs during the
course of job_allocate() execution until calling _select_nodes_parts()
which could override it under certain conditions (i.e. when accounting
policy limit is hit). The fix restores it back when appropriate.

Continuation of 8712a254.

Bug 7136.

Bug 7146.

Bug 7148.

Update NEWS, slurm.spec, and slurm.spec-legacy as well.

Previously with a configuration including FastSchedule=1 or 2,
GRES with files defined in gres.conf, and in the node configuration
either specified with a count of 0 or not included at all then
the node's GRES data would be built with a topology array containing
a GRES bitmap of zero size, resulting in an abort when trying to
operate on that bitmap.

bug 6997

Commit: 03165950

Bug 6466.

Bug 6466.

The name variable hasn't been set yet, so this is always NULL. Print the
uid/gid instead. While here, treat uid/gid as uint32_t, and use strtoul()
rather than atoi() to avoid issues with high-number uid/gid values.

Fixes GCC 9 warning.

Bug 7101.

Bug 7113

Coverity CID 198461

Bug 7112

Coverity CID 198464

Since the struct has an array, Clang complains about the use
of {0}. We decided to replace that with memset.

Bug 7096

Bug 6783

Co-authored-by: Alejandro Sanchez <alex@schedmd.com>

For testing purposes, disabling PrologFlags=Alloc seems reasonable.
If/when we actually need to support front-end systems in the future,
the underlying code will likely have changed and require additional
changes.

Bug 7057.

This reverts commit 123cb8e2.

The reason this code doesn't work is there are other flags set on the
job_state that make it not just Pending.

See test28.2.  It has the JOB_UPDATE_DB flag set.

Bug 7099.

commits:
e293204d
d5131780

Use RETRY_DELAY per to mirror job complete delay but without a max retry
count for the time being.

Bug 6970.

Bug 7040

Signed-off-by: Brian Christiansen <brian@schedmd.com>

Node state is 32bit. Have to wait till 20.02 to change packing routines.

See 845ff7d4

Bug 6964

Bug 6964

The reason was being set after the message was sent to the db. Also
clear the draing and reboot states before the message is sent so that
the event state will show DOWN.

Bug 6964

Bug 6964

so that new jobs can't get on the node.

Bug 6964