Regression from 7bed3120.

Bug 16141

Regression from 7bed3120.

Bug 16141

Bug 16181

Because the handling of job array is simplified we're not performing all
checks. This means that we may skip test for number of jobs tested
against bf_max_job_test. If we succeed to schedule job array element, we
need to check if the limit isn't hit before going to next task.

Bug 15685

Continuation of commit 16b51bf6.

Bug 16134

If the namespace (.ns) was already unmounted for some reason, then
job_container/tmpfs would error and not cleanup the rest of the
job directory. This is wanted for other errors that result in failing to
unmount the namespace, but in this situation where the namespace was
already unmounted, the job directory can be cleaned up.

Bug 16013

This reverts commit 9f41a5eb.

Bug 12500

Bug 7396

Instead of generating a formatted string off of NO_VAL value which is
confusing.

Bug 15953

Bug 15863

Bug 15863

Output format has changed.

Update slurm.spec as well.

Skip rc2, proceed straight to .0.

Bug 16119

Prior to this a hypothetical pair that would collide look like:

associd: 11 cluster: foo
associd: 1  cluster: 1foo

(While numbers as a leading part of the cluster name aren't common, they
aren't currently prohibited.)

Bug 16118

Co-authored-by: Tim Wickberg <tim@schedmd.com>

_arch_hash_rec_id() returns an xmalloc'ed key. uthash stores the pointer
to the key but xhash and uthash don't free the key (rightly so).
c15531c5 removed the reference to the key in xhash_item_t so it can't
be free'd by an xhash freefunc.

Because slurmdb_hierarchical_rec_t is used in a list and hash, it's
easier to just store the key in slurmdb_hierarchical_rec_t and free it
when it's destroyed -- rather than reverting c15531c5.

Bug 16118

Co-authored-by: Tim Wickberg <tim@schedmd.com>

declare is not a portable shell command. Match behavior with other
components like oneapi.

Bug 16109

Commit 5189704f changes the column used to determine when to archive
and purge to mod_time, which prior to this patch did not have an index.

This patch update the indexes based on the work done in c7c9c5fb to
maintain the performance benefit.

Bug 16108

This never worked previously, dating back to the introduction in cb901366.

Bug 14590.

Signed-off-by: Oriol Vilarrubi <jvilarru@schedmd.com>

If unset, do not attempt to change the GPU frequency if --gpu-freq has not
been set for the step.

Note that, before the next commit, GpuFreqDef does not actually work, and
was never implemented correctly.

Bug 14590.

Signed-off-by: Oriol Vilarrubi <jvilarru@schedmd.com>

Regression from commit ba4ed853.

Commit ba4ed853 added _gres_node_state_delete() which has the
identical for loop to free gres_ns->topo_*. Because
gres_ns->topo_type_name is NULL after _gres_node_state_delete(), the
slurmctld segfaults on this line on the dereference:

xfree(gres_ns->topo_type_name[i]);

Bug 16110

Bug 16113.

Bug 16086

Signed-off-by: Felip Moll <felip.moll@schedmd.com>

Avoid setting either parser to NULL. Instead swap to a NO_VAL model to
make the unset state explicit.

Bug 16073

Incorrect reference to v0.0.39_job_submission for RequestBody which
should point to DATA_PARSER_UPDATE_NODE_MSG.

Bug 16074

Regression from f141cd9f:

-		} else if (_multi_cluster(params.clusters) != 0)
+		} else if (!_multi_cluster(params.clusters, argc, argv))
 			rc = 1;

Bug 16104

In 23.02, the client doesn't send the user_id and group_id in the
job_desc_msg_t anymore (ca4bb950) and is set from the auth message
when received by the origin cluster. The origin cluster needs to send
this along to the siblings now since the job_desc_msg_t is just passed
along to the sibling clusters without re-packing the job_desc.

Bug 16104