lustre/scripts/lustre

   1 #!/bin/sh
   2 #
   3 # lustre   This shell script takes care of starting and stopping Lustre
   4 #
   5 # chkconfig: - 99 1
   6 # description: Lustre Lite network File System.
   7 #              This starts both Lustre client and server functions.
   8 # processname: lconf
   9 # config: /etc/lustre/config.xml
  10 # pidfile: /var/run/lustre.pid
  11 ### BEGIN INIT INFO
  12 # Provides: lustre
  13 # Required-Start: $network +sshd
  14 # Required-Stop: $network
  15 # Should-Start:
  16 # Should-Stop:
  17 # Default-Start:
  18 # Default-Stop: 0 1 2 3 4 5 6
  19 # Short-Description: Lustre Lite network File System.
  20 # Description: This starts both Lustre client and server functions.
  21 ### END INIT INFO
  22
  23
  24 SERVICE=${0##*/}
  25
  26 : ${LUSTRE_CFG:=/etc/lustre/lustre.cfg}
  27 [ -f ${LUSTRE_CFG} ] && . ${LUSTRE_CFG}
  28 [ -f /etc/sysconfig/lustre ] && . /etc/sysconfig/lustre
  29
  30 : ${LUSTRE_CONFIG_XML:=/etc/lustre/config.xml}
  31 : ${LCONF:=/usr/sbin/lconf}
  32 : ${LCTL:=/usr/sbin/lctl}
  33 # Some distros use modprobe.conf.local
  34 if [ -f /etc/modprobe.conf.local ]; then
  35    : ${MODPROBE_CONF:=/etc/modprobe.conf.local}
  36 else
  37    : ${MODPROBE_CONF:=/etc/modprobe.conf}
  38 fi
  39 # Be sure the proper directories are in PATH.
  40 export PATH="/sbin:$PATH"
  41
  42 case "$SERVICE" in
  43     [SK][[:digit:]][[:digit:]]lustre | lustre)
  44         SERVICE="lustre"
  45         : ${LCONF_START_ARGS:="${LUSTRE_CONFIG_XML}"}
  46         : ${LCONF_STOP_ARGS:="--force --cleanup ${LUSTRE_CONFIG_XML}"}
  47         ;;
  48     *)
  49         : ${LCONF_START_ARGS:="--group ${SERVICE} --select ${SERVICE}=${HOSTNAME} ${LUSTRE_CONFIG_XML}"}
  50         : ${LCONF_STOP_ARGS:="--group ${SERVICE} --select ${SERVICE}=${HOSTNAME} --failover --cleanup ${LUSTRE_CONFIG_XML}"}
  51         ;;
  52 esac
  53 LOCK=/var/lock/subsys/$SERVICE
  54
  55 # Source function library.
  56 if [ -f /etc/init.d/functions ] ; then
  57         . /etc/init.d/functions
  58 fi
  59
  60 # Source networking configuration.
  61 if [ -f /etc/sysconfig/network ] ; then
  62         . /etc/sysconfig/network
  63 fi
  64
  65 check_start_stop() {
  66         # Exit codes now LSB compliant
  67         # Check that networking is up. - exit 'not running'
  68         [ "${NETWORKING}" = "no" ] && exit 7
  69
  70         # exit 'not installed'
  71         [ -x ${LCONF} -a -x ${LCTL} ] || exit 5
  72
  73         if [ ${LUSTRE_CONFIG_XML:0:1} = "/" ] ; then
  74                         if [ ! -f ${LUSTRE_CONFIG_XML} ] ; then
  75                         echo "${0##*/}: Configuration file ${LUSTRE_CONFIG_XML} not found; skipping."
  76                         # exit 'not configured'
  77                         exit 6
  78                 fi
  79         fi
  80
  81         # Create /var/lustre directory
  82         # This is used by snmp agent for checking lustre services
  83         #    status online/offline/online pending/offline pending.
  84
  85         [ -d ${STATUS_DIR:=/var/lustre} ] || mkdir -p $STATUS_DIR
  86         STATUS=${STATUS_DIR}/sysStatus
  87 }
  88
  89 start() {
  90         if [ -x "/usr/sbin/clustat" -a "${SERVICE}" = "lustre" ] ; then
  91                 if [ ! -f "/etc/lustre/start-despite-clumanager" ] ; then
  92                 cat >&2 <<EOF
  93 This script was run directly, which can be dangerous if you are using
  94 clumanager to manage Lustre services.
  95
  96 If you are not using clumanager for Lustre services, run the following
  97 command to have this script start Lustre instead:
  98
  99 touch /etc/lustre/start-despite-clumanager
 100 EOF
 101                 RETVAL=6  # program not configured
 102                 return
 103             fi
 104         fi
 105         check_start_stop
 106         echo -n "Starting $SERVICE: "
 107         if [ $UID -ne 0 ]; then
 108                 echo "Lustre should be started as root"
 109                 RETVAL=4 # insufficent privileges
 110                 return
 111         fi
 112         # Cat the modprobe file and place all lines that follow a trailing backslash on the same line
 113         ROUTER=`cat ${MODPROBE_CONF} | sed ':a;N;$!ba;s#\\\[:space:]*\\n##g' | grep lnet | grep forwarding=\"enabled\"`
 114         if [[ ! -z ${ROUTER} ]]; then
 115                 modprobe lnet
 116                 ${LCTL} network configure
 117         else
 118                 ${LCONF} ${LCONF_START_ARGS}
 119         fi
 120         RETVAL=$?
 121         echo $SERVICE
 122         if [ $RETVAL -eq 0 ]; then
 123                 touch $LOCK
 124                 echo "online" >$STATUS
 125         else
 126                 echo "online pending" >$STATUS
 127         fi
 128 }
 129
 130 stop() {
 131         check_start_stop
 132         echo -n "Shutting down $SERVICE: "
 133         if [ $UID -ne 0 ]; then
 134                 echo "Lustre should be stopped as root"
 135                 RETVAL=4 # insufficent privileges
 136                 return
 137         fi
 138         # Cat the modprobe file and place all lines that follow a trailing backslash on the same line
 139 +       ROUTER=`cat ${MODPROBE_CONF} | sed ':a;N;$!ba;s#\\\[:space:]*\\n##g' | grep lnet | grep forwarding=\"enabled\"`
 140         if [[ ! -z ${ROUTER} ]]; then
 141                 MODULE_LOADED=`lsmod | awk ' { print $1 } ' | grep lnet`
 142                 if [[ ! -z ${MODULE_LOADED} ]]; then
 143                         ${LCTL} network unconfigure
 144                 fi
 145                 ${LCTL} modules | awk '{ print $2 }' | xargs rmmod >/dev/null 2>&1
 146                 # do it again, in case we tried to unload ksocklnd too early
 147                 ${LCTL} modules | awk '{ print $2 }' | xargs rmmod
 148
 149         else
 150                 ${LCONF} ${LCONF_STOP_ARGS}
 151         fi
 152         RETVAL=$?
 153         echo $SERVICE
 154         rm -f $LOCK
 155         if [ $RETVAL -eq 0 ]; then
 156                 echo "offline" >$STATUS
 157         else
 158                 echo "offline pending" >$STATUS
 159         fi
 160 }
 161
 162 restart() {
 163         stop
 164         start
 165 }
 166
 167 status() {
 168         STATE="stopped"
 169         # LSB compliance - return 3 if service is not running
 170         # Lustre-specific returns
 171         # 150 - partial startup
 172         # 151 - health_check unhealthy
 173         # 152 - LBUG
 174         RETVAL=3
 175         egrep -q "libcfs|lvfs|portals" /proc/modules && STATE="loaded"
 176
 177         # check for any routes - on a portals router this is the only thing
 178         [ "`cat /proc/sys/lnet/routes 2> /dev/null`" ] && STATE="running" && RETVAL=0
 179
 180         # check for any configured devices (may indicate partial startup)
 181         if [ -d /proc/fs/lustre ]; then
 182                 [ "`cat /proc/fs/lustre/devices 2> /dev/null`" ] && STATE="partial" && RETVAL=150
 183
 184                 # check for either a server or a client filesystem
 185                 MDS="`ls /proc/fs/lustre/mds/*/recovery_status 2> /dev/null`"
 186                 OST="`ls /proc/fs/lustre/obdfilter/*/recovery_status 2> /dev/null`"
 187                 LLITE="`ls /proc/fs/lustre/llite/fs* 2> /dev/null`"
 188                 [ "$MDS" -o "$OST" -o "$LLITE" ] && STATE="running" && RETVAL=0
 189         else
 190                 # check if this is a router
 191                 if [ -d /proc/sys/lnet ]; then
 192                         ROUTER="`cat /proc/sys/lnet/routes | head -1 | grep -i -c \"Routing enabled\"`"
 193                         if [[ ! -z ${ROUTER} && ${ROUTER} -ge 1 ]]; then
 194                                 STATE="running"
 195                                 RETVAL=0
 196                         fi
 197                 fi
 198         fi
 199
 200         # check for server disconnections
 201         DISCON="`grep -v FULL /proc/fs/lustre/*c/*/*server_uuid 2> /dev/null`"
 202         [ "$DISCON" ] && STATE="disconnected" && RETVAL=0
 203
 204         # check for servers in recovery
 205         [ "$MDS$OST" ] && grep -q RECOV $MDS $OST && STATE="recovery" && RETVAL=0
 206
 207         # check for error in health_check
 208         HEALTH="/proc/fs/lustre/health_check"
 209         [ -f "$HEALTH" ] && grep -q "NOT HEALTHY" $HEALTH && STATE="unhealthy" && RETVAL=151
 210
 211         # check for LBUG
 212         [ -f  "$HEALTH" ] && grep -q "LBUG" $HEALTH && STATE="LBUG" && RETVAL=152
 213
 214         # If Lustre is up , check if the service really exists
 215         # Skip this is we are not checking a specific service
 216         if [ $RETVAL -eq 0 ] && [ $SERVICE != 'lustre' ]; then
 217                 DUMMY=`lctl dl | grep $SERVICE`
 218                 [ $? -ne 0 ] && STATE="not_found" && RETVAL=3
 219         fi
 220
 221         echo $STATE
 222 }
 223
 224 # See how we were called.
 225 case "$1" in
 226   start)
 227         start
 228         ;;
 229   stop)
 230         stop
 231         ;;
 232   restart)
 233         restart
 234         ;;
 235   status)
 236         status $SERVICE
 237         ;;
 238   *)
 239         echo "Usage: $SERVICE {start|stop|restart|status}"
 240         exit 1
 241 esac
 242
 243 exit $RETVAL