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 lxc_hostname() {
  44         inside=`cat $(lxc_rootfs $1)/etc/hostname`
  45         config=`grep lxc.utsname /var/lib/lxc/$name/config | cut -d= -f2`
  46         echo "$config [$inside]";
  47 }
  48
  49 lxc_ip() {
  50         grep lxc.network.ipv4 /var/lib/lxc/$name/config | cut -d= -f2 ||
  51         grep address $(lxc_rootfs $name)/etc/network/interfaces | sed 's/.*address //'
  52 }
  53
  54 lxc_status() {
  55         ( find /var/lib/lxc/ -name "config" | cut -d/ -f5 | sort -u | xargs -i lxc-info -n {} | sed "s/'//g" | while read name is status ; do
  56                 boot="-"
  57                 test -s /var/lib/lxc/$name/on_boot && boot="boot"
  58                 echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
  59         done ) | column -t
  60 }
  61
  62
  63 cleanup_init_scripts() {
  64         rootfs=$(lxc_rootfs $1)
  65
  66         ls \
  67                 $rootfs/etc/rc?.d/*umountfs \
  68                 $rootfs/etc/rc?.d/*umountroot \
  69                 $rootfs/etc/rc?.d/*hwclock* \
  70                 $rootfs/etc/rc?.d/*udev* \
  71         2>/dev/null | xargs -i rm -v {}
  72
  73         echo $1 > $rootfs/etc/hostname
  74 }
  75
  76
  77 setup_inittab() {
  78         rootfs=$(lxc_rootfs $1)
  79         remove=$2
  80         add=$3
  81
  82         # let container respond to kill -SIGPWR
  83         inittab=$rootfs/etc/inittab
  84         if test -e $inittab && ! grep "$add" ${inittab} >/dev/null ; then
  85                 grep -v "$remove" ${inittab} > ${inittab}.new
  86                 echo $add >> ${inittab}.new
  87                 mv ${inittab}.new ${inittab}
  88                 echo "$inittab modified with $add"
  89         fi
  90 }
  91
  92
  93 lxc_log() {
  94         echo `date +%Y-%m-%dT%H:%M:%S` $*
  95 }
  96
  97
  98 lxc_kill() {
  99         name=$1
 100         sig=$2
 101
 102         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
 103         if [ -z "$init_pid" ] ; then
 104                 lxc-info -n $name
 105                 exit 1
 106         fi
 107         lxc_log "$name kill $sig $init_pid"
 108         /bin/kill $sig $init_pid
 109 }
 110
 111 lxc_stop() {
 112         lxc_log "$name stop"
 113         lxc_kill $name -SIGPWR
 114         lxc-wait -n $name -s STOPPED
 115         lxc_log "$name stoped"
 116 #       rm -f /var/lib/lxc/${name}/on_boot
 117 }
 118
 119
 120 lxc_start() {
 121         name=$1
 122         rootfs=$(lxc_rootfs $1)
 123
 124         if [ ! -e $rootfs ] ; then
 125                 echo "ERROR $name rootfs $rootfs not found"
 126                 return
 127         fi
 128
 129         if ! lxc-info -n $name | grep RUNNING ; then
 130                 lxc_log "$name start"
 131                 lxc-start -n $name -o /tmp/${name}.log -d
 132                 lxc-wait  -n $name -s RUNNING
 133                 lxc-info  -n $name
 134                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 135         fi
 136 }
 137
 138 lxc_watchdog() {
 139 name=$1
 140 rootfs=$(lxc_rootfs $1)
 141
 142 while true; do
 143         vps_utmp=${rootfs}/var/run/utmp
 144         tasks=`wc -l < /cgroup/${name}/tasks`
 145         test -z "$tasks" && exit 1
 146         if [ "$tasks" -eq 1 ]; then
 147
 148                 runlevel="$(runlevel ${vps_utmp})"
 149                 lxc_log "$name runlevel $runlevel"
 150
 151                 case $runlevel in
 152                 N*)
 153                         # nothing for new boot state
 154                 ;;
 155                 ??0)
 156                         lxc_log "$name halt"
 157                         lxc-stop -n "${name}"
 158                         lxc-wait -n ${name} -s STOPPED
 159                         break
 160                 ;;
 161                 ??6)
 162                         lxc_log "$name reboot";
 163                         lxc-stop -n ${name}
 164                         lxc-wait -n ${name} -s STOPPED
 165                         lxc-start -d -n ${name} -o /tmp/${name}.log
 166                 ;;
 167                 *)
 168                         # make sure vps is still running
 169                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 170                         [ "$state" = "RUNNING" ] || break
 171                 ;;
 172                 esac
 173         else
 174                 lxc_log "$name $tasks tasks"
 175         fi
 176
 177         # time of 5 minutes on it JUST IN CASE...
 178         inotifywait -qqt 300 ${vps_utmp}
 179 done
 180
 181 lxc_log "$name watchdog exited"
 182
 183 }
 184
 185
 186 usage() {
 187         echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
 188         exit 3
 189 }
 190
 191 command_on_lxc() {
 192 command=$1
 193 shift
 194
 195 echo "# $command $1"
 196
 197 case "$command" in
 198
 199 start)
 200         lxc_exists $1
 201         cleanup_init_scripts $1
 202         setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
 203         setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
 204         setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
 205         lxc_start $1
 206         # give container 5 seconds to start more than one process
 207         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 208         ;;
 209 stop|halt)
 210         lxc_exists $1
 211         lxc_stop $1
 212         ;;
 213 reload|force-reload|restart|reboot)
 214         lxc_kill $1 -SIGINT
 215         ;;
 216 watchdog)
 217         lxc_watchdog $1
 218         ;;
 219 boot)
 220         echo $1 > /var/lib/lxc/$1/on_boot
 221         ;;
 222 disable)
 223         echo -n > /var/lib/lxc/$1/on_boot
 224         ;;
 225 *)
 226         usage
 227         ;;
 228
 229 esac
 230
 231 }
 232
 233 command=$1
 234 test -z "$command" && usage
 235 test "$command" = "status" && lxc_status && exit
 236 shift
 237
 238 if [ -z "$1" ] ; then
 239         ls /var/lib/lxc/*/on_boot | while read path ; do
 240                 name=`echo $path | cut -d/ -f5`
 241                 if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
 242                         command_on_lxc $command $name
 243                 else
 244                         echo "# skip $command $name"
 245                 fi
 246         done
 247 else
 248         while [ ! -z "$1" ] ; do
 249                 command_on_lxc $command $1
 250                 shift
 251         done
 252 fi
 253