This reverts the following two commits:
d84bdaea
25051e48
While they work as designed, we probably do not want to provide
support for the DenyGroups option

The AllowGroups and DenyGroups configuration parameters only apply
to the effective group ID of the job. Previously any user with
access to any allowed group was permitted to use the partition.

My patch for 366 still has a bug in that the step number in the
job file is incorrect.
Bug 368

Bug 366

Before a task is attached to a PAGG and the PAGG attached to a
reservation, no tracking is done.  Any files and directories the task
creates at such a time will not be tracked.

When a task is removed from a PAGG (via job_detachpid()), its current
cleanup state is dropped, and no further files it creates will be tracked.

Put another way, job_detachpid() will cancel any cleanup state accrued by
that task, and make it as if that task had never been subject to CNCU.

Once the creating task exits while subject to CNCU, the CNCU-tracking of
their files is inherited by the PAGG, and then by the reservation when the
PAGG exits.

So if the slurmstepd task had spawned child tasks, and those tasks create
files and exit while still attached to the PAGG attached to the
reservation, then those files will be cleaned up when the reservation
ends.

Paul

AllowAccounts, AllowQOS, DenyAccounts, DenyGroups and DenyQOS.
Enforcement logic needs to be added as well as slurm.conf man page.

Previously a killpg was used, now a signal_container call is used and
this might result in the processes being signalled at very slightly
different times, possibly permitting a new process to start

Makes sure that we don't run the multifactor plugin using the
ticket based logic

This appears to have been added by accident

The most significant change is that we don't signal individual processes
unless there are some processes in the step which are currently core
dumping. This change signals the processes closer together in time
when possible.

message.

users.

Processes which can be signalled will be. Others will be signalled
when possible from a pthread.

It is not clear why this RPC exists in the first place. Convert
to use REQUEST_SIGNAL_CONTAINER RPC instead.

Conflicts:
	META
	NEWS
	RELEASE_NOTES