Set the delay time for job requeue to the job credential lifetime (1200
second by default). This insures that prolog runs on every node when a
job is requeued. (This change will slow down launch of re-queued jobs).
Without this change, if a job is restated within 1200 seconds, the nodes
previously used would not run the prolog again, since the job ID is
still seen as active (from the previous execution). It is also advisable
to set the value of DEFAULT_EXPIRATION_WINDOW in src/common/slurm_cred.c
to the lowest value reasonable. We need to add a new configuration parameter
so this is easly changed in the future.

Conflicts:
	src/slurmctld/job_mgr.c

Display job's estimated NodeCount based off of partition's configured resources rather than the whole system's.

Bug 1478

This provides a better global view of what the limits and caps are.

Remove time from "capmc get_node_energy_counter" call. If no recent
  data is available, no data is being returned, so just get latest
  information.
Initialize a variable to avoid xfree of uninitialized variable.
Correct joule to watt calculation (">" changed to "<")
Read configuration once when slurmctld starts rather than twice
Compute a node's power consumption with more precision based upon
  time to the microsecond

Add links to burst buffer and power management pages.
Add JSON-C build/installation instructions.

Previously, there was no binding of tasks to the appropriate NUMA.
Based upon work by Josko Plazonic <plazonic@princeton.edu>.

Improved logging and some code restructuring. No change in logic.

This reverts commit e24a418b.

Improved logging and some code restructuring. No change in logic.

Mail notifications on job BEGIN, END and FAIL now apply to a job array as a
whole rather than generating individual email messages for each task in the
job array.

This reverts commit e24a418b.

This is a variation on commit 5391b8cc
Check $HOME/.my.cnf last rather than first to follow more standard search order

Bug 1469

The /root/.my.cnf would typically contain the login credentials for
root.  If those are needed for Slurm, then it should be checking
that directory.

(In reply to Nina Suvanphim from comment #0)
...
> const char *default_conf_paths[] = {
> "/root/.my.cnf", <<<<<<<<<<<<<<<<<------- add this line
> "/etc/my.cnf", "/etc/opt/cray/MySQL/my.cnf",
> "/etc/mysql/my.cnf", NULL };

I'll also note that typically the $HOME/.my.cnf file would be
checked last rather than first.

Fix some logic related to power distribution across nodes

don't support strong_alias

Update power management web page: Add notes about powering nodes down/up
Prevent underflow in power distribution logic
Add logic to identify nodes in "ready" state. Only ready nodes can have
  their power caps modified
Don't change power cap if node not in ready state
Various improvements to logging
Refactor code to eliminate duplicate/repeated building of full NID list
Plug some memory leaks

Modify test 12.7 so that we specify a reason when setting a node DOWN
A recent change to the Slurm code now requires a reason

Correct capmc arguments to set power cap.
Convert "capmc get_node_energy_counter" to use hostlist expressin rather
   than listing every node in a comma separated list.
Log commands and args run by the plugin via the power_run_script()
   function in src/plugins/power/common/power_common.c.
Use hostlist to build condenced nid list for power cap set/clear functions.

we came across the following error message in the slurmctld logs when
using non-consumable resources:

error: gres/potion: job 39 dealloc of node node1 bad node_offset 0 count
is 0

The error comes from _job_dealloc():

node_gres_data=0x7f8a18000b70, node_offset=0, gres_name=0x1999e00
"potion", job_id=46,
    node_name=0x1987ab0 "node1") at gres.c:3980
(job_gres_list=0x199b7c0, node_gres_list=0x199bc38, node_offset=0,
job_id=46,
    node_name=0x1987ab0 "node1") at gres.c:4190
job_ptr=0x19e9d50, pre_err=0x7f8a31353cb0 "_will_run_test", remove_all=true)
    at select_linear.c:2091
bitmap=0x7f8a18001ad0, min_nodes=1, max_nodes=1, max_share=1, req_nodes=1,
    preemptee_candidates=0x0, preemptee_job_list=0x7f8a2f910c40) at
select_linear.c:3176
bitmap=0x7f8a18001ad0, min_nodes=1, max_nodes=1, req_nodes=1, mode=2,
    preemptee_candidates=0x0, preemptee_job_list=0x7f8a2f910c40,
exc_core_bitmap=0x0) at select_linear.c:3390
bitmap=0x7f8a18001ad0, min_nodes=1, max_nodes=1, req_nodes=1, mode=2,
    preemptee_candidates=0x0, preemptee_job_list=0x7f8a2f910c40,
exc_core_bitmap=0x0) at node_select.c:588
avail_bitmap=0x7f8a2f910d38, min_nodes=1, max_nodes=1, req_nodes=1,
exc_core_bitmap=0x0)
    at backfill.c:367

The cause of this problem is that _node_state_dup() in gres.c does not
duplicate the no_consume flag.
The cr_ptr passed to _rm_job_from_nodes() is created with _dup_cr()
which calls _node_state_dup().

Below is a simple patch to fix the problem. A "future-proof" alternative
might be to memcpy() from gres_ptr to new_gres and
only handle pointers separately.

Bug 1471