If a job preemption involves multiple preemption types (cancel +
requeue + suspend) and more than one job is preempted with one
of the preemption types being suspend and the other something else,
the preemptee initiation may be delayed longer than needed.
bug 377

such that  1) specifying --without-rrdtool or --with-rrdtool=no works as
expected, disabling rrdtool build and  2) the link test also tests for
rrd_info_t which was not present in the old rrdtool-devel-1.2.27 which I
happen to have installed. It is present in the newer rrdtool I used from
rpmforge for centos5 rrdtool-devel-1.4.7-1.el5.rf.x86_64.
rrd_info_t was chosen because it was in the output of make when it failed.

That is I saw this error:
/bin/sh ../../../../libtool --tag=CC   --mode=compile gcc -DHAVE_CONFIG_H -I. -I../../../.. -I../../../../slurm -I../../../.. -I../../../../src/common  -I/usr/include    -DNUMA_VERSION1_COMPATIBILITY -g -O2 -pthread -fno-gcse -MT ext_sensors_rrd.lo -MD -MP -MF .deps/ext_sensors_rrd.Tpo -c -o ext_sensors_rrd.lo ext_sensors_rrd.c
libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I../../../.. -I../../../../slurm -I../../../.. -I../../../../src/common -I/usr/include -DNUMA_VERSION1_COMPATIBILITY -g -O2 -pthread -fno-gcse -MT ext_sensors_rrd.lo -MD -MP -MF .deps/ext_sensors_rrd.Tpo -c ext_sensors_rrd.c  -fPIC -DPIC -o .libs/ext_sensors_rrd.o
ext_sensors_rrd.c: In function '_rrd_get_last_one':
ext_sensors_rrd.c:198: error: 'rrd_info_t' undeclared (first use in this function)
ext_sensors_rrd.c:198: error: (Each undeclared identifier is reported only once
ext_sensors_rrd.c:198: error: for each function it appears in.)
ext_sensors_rrd.c:198: error: 'data' undeclared (first use in this function)
ext_sensors_rrd.c:198: error: 'data_p' undeclared (first use in this function)

See commit b156ce54

Add uid to container_g_add_pid() and container_p_add_pid() functions
Resturcture calls to job container in multiple slurmd/stepd locations

Split container_g_add() into 2 functions, One to add a proctrack
container (PAGG) and a new one to add a PID (which creates the
PAGG and adds the PID to it first.

Conflicts:
	configure

Conflicts:
	src/common/slurmdb_defs.c

Avoid referencing memory out of bounds when setting CUDA_VISIBLE_DEVICES
if FastSchedule=2 is configured AND the node has fewer GPUs configured
in gres.conf than in slurm.conf AND those GPUs have specific associated
files

Removed the previous sorting functions.

that was fixed in f1c74d6e

the process is reran for a step/job that was running it when the slurmctld
was shutdown.

when a step is totally done.  This is needed for some systems that have an
internal epilog type process that has to run from the slurmctld and need to
free the resources after it finishes in a plugin.