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