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

   1 #! /bin/sh
   2 ### BEGIN INIT INFO
   3 # Provides:          lxc-watchdog
   4 # Required-Start:    $remote_fs $named $network $time
   5 # Required-Stop:     $remote_fs $named $network
   6 # Required-Start:
   7 # Required-Stop:
   8 # Default-Start:     2 3 4 5
   9 # Default-Stop:      0 1 6
  10 # Short-Description: Manage Linux Containers startup/shutdown
  11 # Description:       Uses clever inotify hack to monitor container's
  12 #                    halt/reboot events watching /var/run/utmp
  13 ### END INIT INFO
  14
  15 # Author: Dobrica Pavlinusic <dpavlin@rot13.org>
  16 #
  17 # based on Tony Risinger post to lxc-users mailing list
  18 # http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00074.html
  19 #
  20 # Install with:
  21 # ln -sf /srv/sysadmin-cookbook/recepies/lxc/lxc-watchdog.sh /etc/init.d/lxc-watchdog
  22 # update-rc.d lxc-watchdog defaults
  23
  24
  25 which inotifywait >/dev/null || apt-get install inotify-tools
  26
  27
  28 lxc_exists() {
  29         name=$1
  30
  31         if [ ! -e /var/lib/lxc/$name/config ] ; then
  32                 echo "Usage: $0 name"
  33                 lxc_status
  34                 exit 1
  35         fi
  36 }
  37
  38
  39 lxc_rootfs() {
  40         grep '^ *lxc\.rootfs *=' "/var/lib/lxc/$1/config" | cut -d= -f2 | sed 's/^ *//'
  41 }
  42
  43
  44 lxc_status() {
  45         find /var/lib/lxc/ -name "config" | cut -d/ -f5 | sort -u | xargs -i lxc-info -n {} | sed "s/'//g" | while read name is status ; do
  46                 boot="    "
  47                 test -s /var/lib/lxc/$name/on_boot && boot="boot"
  48                 echo "$name $status $boot $(lxc_rootfs $name)"
  49         done
  50 }
  51
  52
  53 cleanup_init_scripts() {
  54         rootfs=$(lxc_rootfs $1)
  55
  56         ls \
  57                 $rootfs/etc/rc?.d/*umountfs \
  58                 $rootfs/etc/rc?.d/*umountroot \
  59                 $rootfs/etc/rc?.d/*hwclock* \
  60         2>/dev/null | xargs -i rm -v {}
  61 }
  62
  63
  64 setup_inittab() {
  65         rootfs=$(lxc_rootfs $1)
  66         remove=$2
  67         add=$3
  68
  69         # let container respond to kill -SIGPWR
  70         inittab=$rootfs/etc/inittab
  71         if ! grep "$add" ${inittab} >/dev/null ; then
  72                 grep -v "$remove" ${inittab} > ${inittab}.new
  73                 echo $add >> ${inittab}.new
  74                 mv ${inittab}.new ${inittab}
  75                 echo "$inittab modified with $add"
  76         fi
  77 }
  78
  79
  80 lxc_log() {
  81         echo `date +%Y-%m-%dT%H:%M:%S` $*
  82 }
  83
  84
  85 lxc_kill() {
  86         name=$1
  87         sig=$2
  88
  89         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
  90         if [ -z "$init_pid" ] ; then
  91                 lxc-info -n $name
  92                 exit 1
  93         fi
  94         lxc_log "$name kill $sig $init_pid"
  95         /bin/kill $sig $init_pid
  96 }
  97
  98 lxc_stop() {
  99         lxc_log "$name stop"
 100         lxc_kill $name -SIGPWR
 101         lxc-wait -n $name -s STOPPED
 102         lxc_log "$name stoped"
 103 #       rm -f /var/lib/lxc/${name}/on_boot
 104 }
 105
 106
 107 lxc_start() {
 108         name=$1
 109
 110         if ! lxc-info -n $name | grep RUNNING ; then
 111                 lxc_log "$name start"
 112                 lxc-start -n $name -o /tmp/${name}.log -d
 113                 lxc-wait  -n $name -s RUNNING
 114                 lxc-info  -n $name
 115                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 116         fi
 117 }
 118
 119 lxc_watchdog() {
 120 name=$1
 121 rootfs=$(lxc_rootfs $1)
 122
 123 while true; do
 124         vps_utmp=${rootfs}/var/run/utmp
 125         tasks=`wc -l < /cgroup/${name}/tasks`
 126         test -z "$tasks" && exit 1
 127         if [ "$tasks" -eq 1 ]; then
 128
 129                 runlevel="$(runlevel ${vps_utmp})"
 130                 lxc_log "$name runlevel $runlevel"
 131
 132                 case $runlevel in
 133                 N*)
 134                         # nothing for new boot state
 135                 ;;
 136                 ??0)
 137                         lxc_log "$name halt"
 138                         lxc-stop -n "${name}"
 139                         lxc-wait -n ${name} -s STOPPED
 140                         break
 141                 ;;
 142                 ??6)
 143                         lxc_log "$name reboot";
 144                         lxc-stop -n ${name}
 145                         lxc-wait -n ${name} -s STOPPED
 146                         lxc-start -d -n ${name} -o /tmp/${name}.log
 147                 ;;
 148                 *)
 149                         # make sure vps is still running
 150                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 151                         [ "$state" = "RUNNING" ] || break
 152                 ;;
 153                 esac
 154         else
 155                 lxc_log "$name $tasks tasks"
 156         fi
 157
 158         # time of 5 minutes on it JUST IN CASE...
 159         inotifywait -qqt 300 ${vps_utmp}
 160 done
 161
 162 lxc_log "$name watchdog exited"
 163
 164 }
 165
 166
 167 usage() {
 168         echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
 169         exit 3
 170 }
 171
 172 command_on_lxc() {
 173 command=$1
 174 shift
 175
 176 echo "# $command $1"
 177
 178 case "$command" in
 179
 180 start)
 181         lxc_exists $1
 182         cleanup_init_scripts $1
 183         setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
 184         setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
 185         setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
 186         lxc_start $1
 187         # give container 5 seconds to start more than one process
 188         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 189         ;;
 190 stop|halt)
 191         lxc_exists $1
 192         lxc_stop $1
 193         ;;
 194 reload|force-reload|restart|reboot)
 195         lxc_kill $1 -SIGINT
 196         ;;
 197 watchdog)
 198         lxc_watchdog $1
 199         ;;
 200 boot)
 201         echo $1 > /var/lib/lxc/$1/on_boot
 202         ;;
 203 disable)
 204         echo -n > /var/lib/lxc/$1/on_boot
 205         ;;
 206 *)
 207         usage
 208         ;;
 209
 210 esac
 211
 212 }
 213
 214 command=$1
 215 test -z "$command" && usage
 216 test "$command" = "status" && lxc_status && exit
 217 shift
 218
 219 if [ -z "$1" ] ; then
 220         ls /var/lib/lxc/*/on_boot | while read path ; do
 221                 name=`echo $path | cut -d/ -f5`
 222                 if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
 223                         command_on_lxc $command $name
 224                 else
 225                         echo "# skip $command $name"
 226                 fi
 227         done
 228 else
 229         while [ ! -z "$1" ] ; do
 230                 command_on_lxc $command $1
 231                 shift
 232         done
 233 fi
 234