git.rot13.org Git - sysadmin-cookbook/blob - recepies/lxc/lxc-watchdog.sh

   1 #!/bin/sh
   2
   3 # lxc-watchdog.sh
   4 #
   5 # Dobrica Pavlinusic <dpavlin@rot13.org> 2010-03-15
   6 #
   7 # this script can be used to start/stop Linux containers
   8 # using clever inotify hack to monitor halt/reboot from
   9 # Tony Risinger posted to lxc-users mailing list
  10 #
  11 # http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00074.html
  12
  13
  14 which inotifywait >/dev/null || apt-get install inotify-tools
  15
  16 lxc_status() {
  17         lxc-ls -1 | sort -u | xargs -i lxc-info -n {}
  18 }
  19
  20 lxc_exists() {
  21         name=$1
  22
  23         if [ ! -e /var/lib/lxc/$name/config ] ; then
  24                 echo "Usage: $0 name"
  25                 lxc_status
  26                 exit 1
  27         fi
  28 }
  29
  30
  31 lxc_rootfs() {
  32         grep lxc.rootfs "/var/lib/lxc/$1/config" | cut -d= -f2 | sed 's/^ *//'
  33 }
  34
  35
  36 cleanup_init_scripts() {
  37         rootfs=$(lxc_rootfs $1)
  38
  39         ls \
  40                 $rootfs/etc/rc?.d/*umountfs \
  41                 $rootfs/etc/rc?.d/*umountroot \
  42                 $rootfs/etc/rc?.d/*hwclock* \
  43         2>/dev/null | xargs -i rm -v {}
  44 }
  45
  46
  47 setup_inittab() {
  48         rootfs=$(lxc_rootfs $1)
  49
  50         # let container respond to kill -SIGPWR
  51         inittab=$rootfs/etc/inittab
  52         powerfail="pw::powerfail:/sbin/init 0"
  53         if ! grep "$powerfail" ${inittab} >/dev/null ; then
  54                 grep -v ::power ${inittab} > ${inittab}.new
  55                 echo $powerfail >> ${inittab}.new
  56                 mv ${inittab}.new ${inittab}
  57                 echo "$initab modified"
  58         fi
  59
  60 }
  61
  62
  63 lxc_stop() {
  64         name=$1
  65
  66         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
  67         if [ -z "$init_pid" ] ; then
  68                 lxc-info -n $name
  69                 exit 1
  70         fi
  71         echo "$name stop $init_pid"
  72         /bin/kill -SIGPWR $init_pid
  73         lxc-wait -n $name -s STOPPED
  74
  75 }
  76
  77
  78 lxc_start() {
  79         name=$1
  80
  81         if ! lxc-info -n $name | grep RUNNING ; then
  82                 echo "$name start"
  83                 lxc-start -n $name -o /tmp/${name}.log -d
  84                 lxc-wait  -n $name -s RUNNING
  85                 lxc-info  -n $name
  86         fi
  87 }
  88
  89 lxc_watchdog() {
  90 name=$1
  91 rootfs=$(lxc_rootfs $1)
  92
  93 while true; do
  94         vps_utmp=${rootfs}/var/run/utmp
  95         tasks=`wc -l < /cgroup/${name}/tasks`
  96         test -z "$tasks" && exit 1
  97         if [ "$tasks" -eq 1 ]; then
  98
  99                 runlevel="$(runlevel ${vps_utmp})"
 100                 echo "# $name runlevel $runlevel"
 101
 102                 case $runlevel in
 103                 N*)
 104                         # nothing for new boot state
 105                 ;;
 106                 ??0)
 107                         echo "$name halt"
 108                         lxc-stop -n "${name}"
 109                         lxc-wait -n ${name} -s STOPPED
 110                         break
 111                 ;;
 112                 ??6)
 113                         echo "$name reboot";
 114                         lxc-stop -n ${name}
 115                         lxc-wait -n ${name} -s STOPPED
 116                         lxc-start -d -n ${name} -o /tmp/${name}.log
 117                 ;;
 118                 *)
 119                         # make sure vps is still running
 120                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 121                         [ "$state" = "RUNNING" ] || break
 122                 ;;
 123                 esac
 124         else
 125                 echo "# $name $tasks tasks"
 126         fi
 127
 128         # time of 5 minutes on it JUST IN CASE...
 129         inotifywait -qqt 300 ${vps_utmp}
 130 done
 131
 132 echo "${name} exited"
 133
 134 }
 135
 136
 137 case "$1" in
 138
 139 start)
 140         lxc_exists $2
 141         cleanup_init_scripts $2
 142         setup_inittab $2
 143         lxc_start $2
 144         ( nohup $0 watchdog $2 >> /tmp/$2.log ) &
 145         ;;
 146 stop)
 147         lxc_exists $2
 148         lxc_stop $2
 149         ;;
 150 status)
 151         lxc_status
 152         ;;
 153 reload|force-reload|restart)
 154         lxc_stop $2
 155         lxc_start $2
 156         ;;
 157 watchdog)
 158         lxc_watchdog $2
 159         ;;
 160 *)
 161         echo "Usage: $0 {start|stop|restart|status}" >&2
 162         exit 3
 163         ;;
 164
 165 esac
 166