slurm-llnl/arch_rc.d.slurm

   1 #!/bin/bash
   2 #
   3 # description: SLURM is a simple resource management system which \
   4 #              manages exclusive access to a set of compute \
   5 #              resources and distributes work to those resources.
   6 #
   7 # processname: /usr/sbin/slurmd
   8 # pidfile: /run/slurmd.pid
   9 #
  10 # processname: /usr/sbin/slurmctld
  11 # pidfile: /run/slurmctld.pid
  12 #
  13 # config: /etc/default/slurm
  14 #
  15
  16 BINDIR=/usr/bin
  17 CONFDIR=/etc/slurm
  18 LIBDIR=/usr/lib
  19 SBINDIR=/usr/sbin
  20
  21 # Source main config
  22 if [ ! -f /etc/rc.conf ]; then
  23     echo "Could not find /etc/rc.conf. Are we using ARCH linux?"
  24     exit 1
  25 fi
  26 . /etc/rc.conf
  27
  28 # Source function library.
  29 if [ ! -f /etc/rc.d/functions ]; then
  30     echo "Could not find /etc/rc.d/functions. Is some other daemon launch mechanism used?"
  31     exit 1
  32 fi
  33 . /etc/rc.d/functions
  34
  35 STARTPROC=daemon
  36
  37 function rc_status() {
  38     RETVAL=$?
  39 }
  40 function rc_exit () {
  41     exit $RETVAL
  42 }
  43 RETVAL=0
  44
  45
  46 if [ ! -x $BINDIR/scontrol ]; then
  47    echo "Could not find $BINDIR/scontrol. Bad path?"
  48    exit 1
  49 fi
  50
  51 # Source slurm specific configuration
  52 # SLURMCTLD_OPTIONS defines slurmctld command line options. See "man slurmctld"
  53 # SLURMD_OPTIONS defines slurmd command line options. See "man slurmd"
  54 if [ -f /etc/default/slurm ] ; then
  55     . /etc/default/slurm
  56 else
  57     SLURMCTLD_OPTIONS=""
  58     SLURMD_OPTIONS=""
  59 fi
  60
  61 if [ ! -f $CONFDIR/slurm.conf ]; then
  62    echo "Could not find $CONFDIR/slurm.conf. Bad path?"
  63    exit 1
  64 fi
  65
  66 # setup library paths for slurm and munge support
  67 export LD_LIBRARY_PATH=$LIBDIR${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}
  68
  69 start() {
  70     prog=$1
  71     shift
  72     stat_busy "Starting $prog"
  73     unset HOME MAIL USER USERNAME
  74     $SBINDIR/$prog $*
  75     if [ $? -gt 0 ]; then
  76         stat_fail
  77     else
  78         add_daemon $prog
  79         stat_done
  80     fi
  81 }
  82
  83 stop() {
  84     stat_busy "Stopping $prog"
  85     PID=`pidof -o %PPID $prog`
  86     [ "$PID" ] && kill $PID &> /dev/null
  87     if [ $? -gt 0 ]; then
  88         stat_fail
  89     else
  90         rm_daemon $prog
  91         stat_done
  92     fi
  93 }
  94
  95 startall() {
  96     for prog in `$BINDIR/scontrol show daemons`; do
  97         optvar=`echo ${prog}_OPTIONS | tr "a-z" "A-Z"`
  98         start $prog ${!optvar}
  99     done
 100 }
 101
 102 #
 103 # status() with slight modifications to take into account
 104 # instantiations of job manager slurmd's, which should not be
 105 # counted as "running"
 106 #
 107 slurmstatus() {
 108     local base=${1##*/}
 109     local pid
 110     local rpid
 111     local pidfile
 112
 113     pidfile=`grep -i ${base}pid $CONFDIR/slurm.conf | grep -v '^ *#'`
 114     if [ $? = 0 ]; then
 115         pidfile=${pidfile##*=}
 116         pidfile=${pidfile%#*}
 117     else
 118         pidfile=/run/${base}.pid
 119     fi
 120
 121     pid=`pidof -o $$ -o $$PPID -o %PPID -x $1 || \
 122          pidof -o $$ -o $$PPID -o %PPID -x ${base}`
 123
 124     if [ -f $pidfile ]; then
 125         read rpid < $pidfile
 126         if [ "$rpid" != "" -a "$pid" != "" ]; then
 127             for i in $pid ; do
 128                 if [ "$i" = "$rpid" ]; then
 129                     echo $"${base} (pid $pid) is running..."
 130                     return 0
 131                 fi
 132             done
 133         elif [ "$rpid" != "" -a "$pid" = "" ]; then
 134 #           Due to change in user id, pid file may persist
 135 #           after slurmctld terminates
 136             if [ "$base" != "slurmctld" ] ; then
 137                echo $"${base} dead but pid file exists"
 138             else
 139                echo $"${base} is stopped"
 140             fi
 141             return 1
 142         fi
 143
 144     fi
 145
 146     if [ "$base" = "slurmctld" -a "$pid" != "" ] ; then
 147         echo $"${base} (pid $pid) is running..."
 148         return 0
 149     fi
 150
 151     echo $"${base} is stopped"
 152
 153     return 3
 154 }
 155
 156 #
 157 # stop slurm daemons,
 158 # wait for termination to complete (up to 10 seconds) before returning
 159 #
 160 slurmstop() {
 161     for prog in `$BINDIR/scontrol show daemons`; do
 162        stop $prog
 163
 164        for i in 1 2 3 4
 165        do
 166           sleep $i
 167           slurmstatus $prog
 168           if [ $? != 0 ]; then
 169              break
 170           fi
 171        done
 172     done
 173 }
 174
 175 #
 176 # The pathname substitution in daemon command assumes prefix and
 177 # exec_prefix are same.  This is the default, unless the user requests
 178 # otherwise.
 179 #
 180 # Any node can be a slurm controller and/or server.
 181 #
 182 case "$1" in
 183     start)
 184         startall
 185         ;;
 186     startclean)
 187         SLURMCTLD_OPTIONS="-c $SLURMCTLD_OPTIONS"
 188         SLURMD_OPTIONS="-c $SLURMD_OPTIONS"
 189         startall
 190         ;;
 191     stop)
 192         slurmstop
 193         ;;
 194     status)
 195         for prog in `$BINDIR/scontrol show daemons`; do
 196            slurmstatus $prog
 197         done
 198         ;;
 199     restart)
 200         $0 stop
 201         $0 start
 202         ;;
 203     condrestart)
 204         if [ -f /var/lock/subsys/slurm ]; then
 205             for prog in `$BINDIR/scontrol show daemons`; do
 206                  stop $prog
 207                  start $prog
 208             done
 209         fi
 210         ;;
 211     reconfig)
 212         for prog in `$BINDIR/scontrol show daemons`; do
 213             killproc $prog -HUP
 214         done
 215         ;;
 216     test)
 217         for prog in `$BINDIR/scontrol show daemons`; do
 218             echo "$prog runs here"
 219         done
 220         ;;
 221     *)
 222         echo "Usage: $0 {start|startclean|stop|status|restart|reconfig|condrestart|test}"
 223         exit 1
 224         ;;
 225 esac
 226
 227 rc_exit