diff --git a/mcp/pacemaker.in b/mcp/pacemaker.in index aa643f619d..ce6fbffbe2 100644 --- a/mcp/pacemaker.in +++ b/mcp/pacemaker.in @@ -1,232 +1,276 @@ #!/bin/bash # Authors: # Andrew Beekhof # Fabio M. Di Nitto # # License: Revised BSD # chkconfig: - 99 01 # description: Pacemaker Cluster Manager # processname: pacemakerd # ### BEGIN INIT INFO # Provides: pacemaker # Required-Start: $network corosync # Should-Start: $syslog # Required-Stop: $network corosync # Short-Description: Starts and stops Pacemaker Cluster Manager. # Description: Starts and stops Pacemaker Cluster Manager. ### END INIT INFO desc="Pacemaker Cluster Manager" prog="pacemakerd" -cman=0 # set secure PATH PATH="/sbin:/bin:/usr/sbin:/usr/bin:@sbindir@" checkrc() { if [ $? = 0 ]; then success else failure fi } success() { echo -ne "[ OK ]\r" } failure() { echo -ne "[FAILED]\r" } +log() +{ + logger -t pacemaker -p daemon.notice "$*" +} + +notify() +{ + log "$*" + echo -n "$*" +} + status() { pid=$(pidof $1 2>/dev/null) rtrn=$? if [ $rtrn -ne 0 ]; then echo "$1 is stopped" else echo "$1 (pid $pid) is running..." fi return $rtrn } # rpm based distros if [ -d @sysconfdir@/sysconfig ]; then [ -f @INITDIR@/functions ] && . @INITDIR@/functions [ -f @sysconfdir@/sysconfig/pacemaker ] && . @sysconfdir@/sysconfig/pacemaker [ -z "$LOCK_FILE" ] && LOCK_FILE="@localstatedir@/lock/subsys/pacemaker" fi # deb based distros if [ -d @sysconfdir@/default ]; then [ -f @sysconfdir@/default/pacemaker ] && . @sysconfdir@/default/pacemaker [ -z "$LOCK_FILE" ] && LOCK_FILE="@localstatedir@/lock/pacemaker" fi # Unless specified otherwise, assume cman is in use if cluster.conf exists if [ x = "x$PCMK_STACK" -a -f @sysconfdir@/cluster/cluster.conf ]; then PCMK_STACK=cman fi start() { - echo -n "Starting $desc: " + notify "Starting $desc" # most recent distributions use tmpfs for $@localstatedir@/run # to avoid to clean it up on every boot. # they also assume that init scripts will create # required subdirectories for proper operations mkdir -p @localstatedir@/run if status $prog > /dev/null 2>&1; then success else $prog > /dev/null 2>&1 & # Time to connect to corosync and fail sleep 5 if status $prog > /dev/null 2>&1; then touch $LOCK_FILE pidof $prog > @localstatedir@/run/$prog.pid success else failure rtrn=1 fi fi echo } cman_pre_start() { - pid=$(pidof corosync 2>/dev/null) + # start cman if it's not running + service cman status >/dev/null 2>&1 if [ $? -ne 0 ]; then service cman start - rc=$? - - if [ $rc != 0 ]; then - echo "Not starting $desc" - exit $rc + if [ $? -ne 0 ]; then + notify "Aborting startup of $desc" + echo + exit 1 fi - - sleep 2 fi + + # start cman's friends if they're not running but were configured to start automatically + for cservice in clvmd cmirror gfs2; do + chkconfig --list $cservice 2>/dev/null | grep -q ":on" + if [ $? -eq 0 ]; then + service $cservice status >/dev/null 2>&1 + if [ $? -ne 0 ]; then + log "Attempting to start $cservice" + service $cservice start + fi + fi + done } cman_pre_stop() { - pid=$(pidof fenced 2>/dev/null) + # if cman is not running there is nothing we need to do here + service cman status >/dev/null 2>&1 if [ $? -ne 0 ]; then - : CMAN is not running, nothing to do here return fi + + has_lvm=`crm_resource -c | grep Resource: | grep LVM` + + # migrate resources to another node or shut them down cname=`crm_node --name` crm_attribute -N $cname -n standby -v true -l reboot - logger -t pacemaker -p daemon.notice "Waiting for shutdown of managed resources" - echo -n "Waiting for shutdown of managed resources" + notify "Waiting for shutdown of managed resources" + while [ 1 = 1 ]; do # 0x0000000000000002 means managed active=`crm_resource -c | grep Resource: | grep 0x...............[2367] | awk '{print $9}' | grep $cname | wc -l` if [ $active = 0 ]; then break; fi - sleep 1 + sleep 2 echo -n "." done success echo - logger -t pacemaker -p daemon.notice "Leaving fence domain" - echo -n "Leaving fence domain" + if [ -d /sys/kernel/dlm/ ]; then + lockspace="$(ls -1 /sys/kernel/dlm/)" + if [ -n "$lockspace" ]; then + notify "DLM lockspace still in use" + echo "" + + for cservice in gfs2 clvmd cmirrord; do + service $cservice status >/dev/null 2>&1 + if [ $? -eq 0 ]; then + if [ -n "$has_lvm" ] && [ "$cservice" = "clvmd" ]; then + # allow HA-LVM to take a lock on vg/lv before clvmd can exit + notify "Waiting for LVM services to start somewhere else" + sleep 15 + success + fi + log "Attempting to shutdown $cservice" + service $cservice stop + fi + done + fi + fi + + notify "Leaving fence domain" fence_tool leave -w 10 checkrc - logger -t pacemaker -p daemon.notice "Stopping fenced" fenced=$(pidof fenced) - echo -n "Stopping fenced $fenced" + notify "Stopping fenced $fenced" kill -KILL $fenced > /dev/null 2>&1 checkrc } stop() { shutdown_prog=$prog if ! status $prog > /dev/null 2>&1; then shutdown_prog="crmd" fi if status $shutdown_prog > /dev/null 2>&1; then - echo -n "Signaling $desc to terminate: " + notify "Signaling $desc to terminate" kill -TERM $(pidof $prog) > /dev/null 2>&1 - success + checkrc echo - echo -n "Waiting for cluster services to unload:" + notify "Waiting for cluster services to unload" while status $prog > /dev/null 2>&1; do sleep 1 echo -n "." done else echo -n "$desc is already stopped" fi rm -f $LOCK_FILE rm -f @localstatedir@/run/$prog.pid killall -q -9 'crmd stonithd attrd cib lrmd pacemakerd' success echo } rtrn=0 case "$1" in start) # For consistency with stop [ "$PCMK_STACK" = cman ] && cman_pre_start start ;; restart|reload|force-reload) stop start ;; condrestart|try-restart) if status $prog > /dev/null 2>&1; then stop start fi ;; status) status $prog rtrn=$? ;; stop) # # stonithd needs to be around until fenced is stopped - # fenced can't be stopped until any cluster filesystems are unmounted + # fenced can't be stopped until any resource using dlm is active # # So: # 1. put the node into standby # 2. wait for all resources to be stopped # 3. stop fenced and anything that needs it (borrowed from the cman script) # 4. stop pacemaker # 5. stop the rest of cman (so it doesn't end up half up/down) # [ "$PCMK_STACK" = cman ] && cman_pre_stop stop [ "$PCMK_STACK" = cman ] && service cman stop ;; *) echo "usage: $0 {start|stop|restart|reload|force-reload|condrestart|try-restart|status}" rtrn=2 ;; esac exit $rtrn