UnkillableStepProgram exposes environment variables to the script.
Those variables are now documented.

Bug 11231

Bug 11352

Bug 11352

Bug 11352

dealing with --mem-per-cpu and --threads-per-core.

The handling of --threads-per-core and --mem-per-cpu introduced by
49a7d7f9 is inconsistent with memory values calculated from job
credential, which resulted in memory over-allocation.

This needs to be revisited for proper solution since it relies on
non-signed information. This will require modification of job
credential.

Bug 11148

Update example for correctness and style.

Bug 7573

Previous 3 commits.

Bug 10980

The last task reuses array_job_id. So if job_ptr doesn't change, that
means we have already scheduled it and should never go to next_task.

job_launch() can requeue the job if it fails. This puts the job in the
completing and pending state, which could allow the last task to get
scheduled *again* without the check added here. Rescheduling a
completing job will destroy its node_bitmap and job_resrcs and cause
the job to stay completing forever.

Bug 10980

Co-authored-by: Brian Christiansen <brian@schedmd.com>

In practice, there are some rare situations where the job can get
corrupted and lose its job_resrcs object, preventing the controller from
starting due to a segfault in _step_dealloc_lps(). Instead of allowing
this segfault, simply emit an error and return.

Leaving xassert in place so that non-production builds will catch this
situation.

Bug 10980, 7757, 9474, 6837

Co-authored-by: Marshall Garey <marshall@schedmd.com>
Co-authored-by: Michael Hinton <hinton@schedmd.com>

Bug 10980

Remove the misleading statment in MaxArraySize documentation.

Bug 11317

Bug 10458

Bug 10458

Bug 10458

Bug 11331

Bug 11331

Bug 11331

Bug 11331

Bug 11333

Bug 11333

Bug 11333

Continuation of bde072c6.

Explicitly mention addition of --exact, and deprecation of --whole.

Bug 10914.

slurm_auth_init() may fail because of AuthAltTypes, not just AuthType,
so generalize this error message to make it less confusing in such cases.

Bug 11334.

Ensure an appropriate error message is printed. Only applies to slurmdbd
(which lacks an equivalent to StateSaveLocation).

Bug 11334.

Bug 11251

The plugins are only reloaded at restart of the daemons and
configuration changes will likely be missed otherwise.

Bug 10974

Bug 10974

Bug 10974

Bug 10974

Bug 10974

Bug 10974

Bug 10974

Update documentation for v2.x release of Influxdb. Differentiate how
current configuration values are different in v2.x database.
Correct formatting in several parts for readability.

Bug 10974

Set formatting same as source code to avoid needing inline
formatting macro instructions.

Bug 10974

Due to other changes for bug 9193, finding job_ptr here is no longer
needed, so remove it. Because job_ptr isn't used here, we can also move
where we lock the job_write_lock since we don't need it in the else
block anymore, but we still need it in the if block.

Bug 9193

The job burst buffer state was already state to BB_STATE_STAGING_IN
before this function so setting it here was redundant.

Bug 9193

The burst buffer record which is state saved was not created for a job
until the middle of the stage-in thread. If slurmctld was shutdown during
burst buffer stage-in for a job, the thread would be killed before
creating the burst buffer record, and therefore the burst buffer was
not state saved.

Bug 9193

Previously even though burst buffers were state saved, the state of the
burst buffer was always reset to PENDING. This commit gets the state of
the burst buffer when slurmctld was shutdown and then uses that
information to do whatever is needed: certain threads are restarted and
the burst buffer state may be changed, or the burst buffer may just be
cleaned up.

Previously a burst buffer that was state saved was ignored if the job
was completed. This commit removed that logic since the burst buffer may
not have completed stage-out and teardown.

Bug 9193