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 $name)/etc/network/interfaces | sed 's/.*address //' ) | \
  52         sed -e 's/ *//g' -e 's/\/.*$//'
  53 }
  54
  55 lxc_status() {
  56         ( 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
  57                 boot="-"
  58                 test -s /var/lib/lxc/$name/on_boot && boot="boot"
  59                 echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
  60         done ) | column -t
  61 }
  62
  63
  64 cleanup_init_scripts() {
  65         rootfs=$(lxc_rootfs $1)
  66
  67         ls \
  68                 $rootfs/etc/rc?.d/*umountfs \
  69                 $rootfs/etc/rc?.d/*umountroot \
  70                 $rootfs/etc/rc?.d/*hwclock* \
  71                 $rootfs/etc/rc?.d/*udev* \
  72         2>/dev/null | xargs -i rm -v {}
  73
  74         echo $1 > $rootfs/etc/hostname
  75         grep $1 $rootfs/etc/hosts || echo "$(lxc_ip $1) $1" >> $rootfs/etc/hosts
  76 }
  77
  78
  79 setup_inittab() {
  80         rootfs=$(lxc_rootfs $1)
  81         remove=$2
  82         add=$3
  83
  84         # let container respond to kill -SIGPWR
  85         inittab=$rootfs/etc/inittab
  86         if test -e $inittab && ! grep "$add" ${inittab} >/dev/null ; then
  87                 grep -v "$remove" ${inittab} > ${inittab}.new
  88                 echo $add >> ${inittab}.new
  89                 mv ${inittab}.new ${inittab}
  90                 echo "$inittab modified with $add"
  91         fi
  92 }
  93
  94
  95 lxc_log() {
  96         echo `date +%Y-%m-%dT%H:%M:%S` $*
  97 }
  98
  99
 100 lxc_kill() {
 101         name=$1
 102         sig=$2
 103
 104         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
 105         if [ -z "$init_pid" ] ; then
 106                 lxc-info -n $name
 107                 exit 1
 108         fi
 109         lxc_log "$name kill $sig $init_pid"
 110         /bin/kill $sig $init_pid
 111 }
 112
 113 lxc_stop() {
 114         lxc_log "$name stop"
 115         lxc_kill $name -SIGPWR
 116         lxc-wait -n $name -s STOPPED
 117         lxc_log "$name stoped"
 118 #       rm -f /var/lib/lxc/${name}/on_boot
 119 }
 120
 121
 122 lxc_start() {
 123         name=$1
 124         rootfs=$(lxc_rootfs $1)
 125
 126         if [ ! -e $rootfs ] ; then
 127                 echo "ERROR $name rootfs $rootfs not found"
 128                 return
 129         fi
 130
 131         if ! lxc-info -n $name | grep RUNNING ; then
 132                 lxc_log "$name start"
 133                 lxc-start -n $name -o /tmp/${name}.log -d
 134                 lxc-wait  -n $name -s RUNNING
 135                 lxc-info  -n $name
 136                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 137         fi
 138 }
 139
 140 lxc_watchdog() {
 141 name=$1
 142 rootfs=$(lxc_rootfs $1)
 143
 144 while true; do
 145         vps_utmp=${rootfs}/var/run/utmp
 146         tasks=`wc -l < /cgroup/${name}/tasks`
 147         test -z "$tasks" && exit 1
 148         if [ "$tasks" -eq 1 ]; then
 149
 150                 runlevel="$(runlevel ${vps_utmp})"
 151                 lxc_log "$name runlevel $runlevel"
 152
 153                 case $runlevel in
 154                 N*)
 155                         # nothing for new boot state
 156                 ;;
 157                 ??0)
 158                         lxc_log "$name halt"
 159                         lxc-stop -n "${name}"
 160                         lxc-wait -n ${name} -s STOPPED
 161                         break
 162                 ;;
 163                 ??6)
 164                         lxc_log "$name reboot";
 165                         lxc-stop -n ${name}
 166                         lxc-wait -n ${name} -s STOPPED
 167                         lxc-start -d -n ${name} -o /tmp/${name}.log
 168                 ;;
 169                 *)
 170                         # make sure vps is still running
 171                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 172                         [ "$state" = "RUNNING" ] || break
 173                 ;;
 174                 esac
 175         else
 176                 lxc_log "$name $tasks tasks"
 177         fi
 178
 179         # time of 5 minutes on it JUST IN CASE...
 180         inotifywait -qqt 300 ${vps_utmp}
 181 done
 182
 183 lxc_log "$name watchdog exited"
 184
 185 }
 186
 187
 188 usage() {
 189         echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
 190         exit 3
 191 }
 192
 193 command_on_lxc() {
 194 command=$1
 195 shift
 196
 197 echo "# $command $1"
 198
 199 case "$command" in
 200
 201 start)
 202         lxc_exists $1
 203         cleanup_init_scripts $1
 204         setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
 205         setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
 206         setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
 207         lxc_start $1
 208         # give container 5 seconds to start more than one process
 209         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 210         ;;
 211 stop|halt)
 212         lxc_exists $1
 213         lxc_stop $1
 214         ;;
 215 reload|force-reload|restart|reboot)
 216         lxc_kill $1 -SIGINT
 217         ;;
 218 watchdog)
 219         lxc_watchdog $1
 220         ;;
 221 boot)
 222         echo $1 > /var/lib/lxc/$1/on_boot
 223         ;;
 224 disable)
 225         echo -n > /var/lib/lxc/$1/on_boot
 226         ;;
 227 *)
 228         usage
 229         ;;
 230
 231 esac
 232
 233 }
 234
 235 command=$1
 236 test -z "$command" && usage
 237 test "$command" = "status" && lxc_status && exit
 238 shift
 239
 240 if [ -z "$1" ] ; then
 241         ls /var/lib/lxc/*/on_boot | while read path ; do
 242                 name=`echo $path | cut -d/ -f5`
 243                 if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
 244                         command_on_lxc $command $name
 245                 else
 246                         echo "# skip $command $name"
 247                 fi
 248         done
 249 else
 250         while [ ! -z "$1" ] ; do
 251                 command_on_lxc $command $1
 252                 shift
 253         done
 254 fi
 255