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 | grep -v '^#' | cut -d= -f2 || \
  51         grep address $(lxc_rootfs $name)/etc/network/interfaces | grep -v '^#' | sed 's/.*address //' ) | \
  52         head -1 | \
  53         sed -e 's/ *//g' -e 's/\/.*$//'
  54 }
  55
  56 lxc_status() {
  57         ( find /var/lib/lxc/ -name "config" | cut -d/ -f5 | sort -u | while read name ; do
  58                 status=`lxc-info -n $name 2>/dev/null | grep state: | cut -d: -f2`
  59                 boot="-"
  60                 test -s /var/lib/lxc/$name/on_boot && boot="boot"
  61                 echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
  62         done ) | column -t
  63 }
  64
  65
  66 cleanup_init_scripts() {
  67         rootfs=$(lxc_rootfs $1)
  68
  69         ls \
  70                 $rootfs/etc/rc?.d/*checkroot* \
  71                 $rootfs/etc/rc?.d/*umountfs \
  72                 $rootfs/etc/rc?.d/*umountroot \
  73                 $rootfs/etc/rc?.d/*hwclock* \
  74                 $rootfs/etc/rc?.d/*udev* \
  75                 $rootfs/etc/rc?.d/*checkfs* \
  76         2>/dev/null | xargs -i rm -v {}
  77
  78         echo $1 > $rootfs/etc/hostname
  79         grep $1 $rootfs/etc/hosts || echo "$(lxc_ip $1) $1" >> $rootfs/etc/hosts
  80 }
  81
  82
  83 setup_inittab() {
  84         rootfs=$(lxc_rootfs $1)
  85         remove=$2
  86         add=$3
  87
  88         # let container respond to kill -SIGPWR
  89         inittab=$rootfs/etc/inittab
  90         if test -e $inittab && ! grep "$add" ${inittab} >/dev/null ; then
  91                 grep -v "$remove" ${inittab} > ${inittab}.new
  92                 echo $add >> ${inittab}.new
  93                 mv ${inittab}.new ${inittab}
  94                 echo "$inittab modified with $add"
  95         fi
  96 }
  97
  98
  99 lxc_log() {
 100         echo `date +%Y-%m-%dT%H:%M:%S` $*
 101 }
 102
 103
 104 lxc_kill() {
 105         name=$1
 106         sig=$2
 107
 108         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
 109         if [ -z "$init_pid" ] ; then
 110                 lxc-info -n $name
 111                 exit 1
 112         fi
 113         lxc_log "$name kill $sig $init_pid"
 114         /bin/kill $sig $init_pid
 115 }
 116
 117 lxc_stop() {
 118         lxc_log "$name stop"
 119         lxc_kill $name -SIGPWR
 120         lxc-wait -n $name -s STOPPED
 121         lxc_log "$name stoped"
 122 #       rm -f /var/lib/lxc/${name}/on_boot
 123 }
 124
 125
 126 lxc_start() {
 127         name=$1
 128         rootfs=$(lxc_rootfs $1)
 129
 130         if [ ! -e $rootfs ] ; then
 131                 echo "ERROR $name rootfs $rootfs not found"
 132                 return
 133         fi
 134
 135         if ! lxc-info -n $name | grep RUNNING ; then
 136                 lxc_log "$name start"
 137                 dev=`df -P $rootfs | tail -1 | cut -d" " -f1`
 138                 mount $dev -o remount,rw # fix debian upgrade which remounts dir ro
 139                 lxc-start -n $name -o /tmp/${name}.log -d
 140                 lxc-wait  -n $name -s RUNNING
 141                 lxc-info  -n $name
 142                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 143         fi
 144 }
 145
 146 lxc_watchdog() {
 147 name=$1
 148 rootfs=$(lxc_rootfs $1)
 149 run=$rootfs/var/run
 150 test -L $run && run=$rootfs/`readlink $run` # recent Debian have symlink to /run
 151 cd $run || echo "can't cd watchdog into $run"
 152
 153 while true; do
 154         tasks=`wc -l < /cgroup/${name}/tasks`
 155         stop_on=1 # init
 156         sulogins=`lxc-ps --name $name | grep sulogin | wc -l`
 157         if [ "$sulogins" -gt 0 ] ; then
 158                 stop_on=`expr $stop_on + $sulogins`
 159         fi
 160
 161         test -z "$tasks" && exit 1
 162         if [ "$tasks" -eq $stop_on ]; then
 163
 164                 runlevel="$(runlevel utmp)"
 165                 lxc_log "$name runlevel $runlevel"
 166
 167                 case $runlevel in
 168                 N*)
 169                         # nothing for new boot state
 170                 ;;
 171                 ??0|unknown)
 172                         lxc_log "$name halt"
 173                         lxc-stop -n "${name}"
 174                         lxc-wait -n ${name} -s STOPPED
 175                         break
 176                 ;;
 177                 ??6)
 178                         lxc_log "$name reboot";
 179                         lxc-stop -n ${name}
 180                         lxc-wait -n ${name} -s STOPPED
 181                         lxc-start -d -n ${name} -o /tmp/${name}.log
 182                 ;;
 183                 *)
 184                         # make sure vps is still running
 185                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 186                         [ "$state" = "RUNNING" ] || break
 187                 ;;
 188                 esac
 189         else
 190                 lxc_log "$name $tasks tasks $sulogins console"
 191         fi
 192
 193         # time of 5 minutes on it JUST IN CASE...
 194         inotifywait -qqt 300 utmp
 195 done
 196
 197 lxc_log "$name watchdog exited"
 198
 199 }
 200
 201
 202 usage() {
 203         echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
 204         exit 3
 205 }
 206
 207 command_on_lxc() {
 208 command=$1
 209 shift
 210
 211 echo "# $command $1"
 212
 213 case "$command" in
 214
 215 start)
 216         lxc_exists $1
 217         cleanup_init_scripts $1
 218         setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
 219         setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
 220         setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
 221         lxc_start $1
 222         # give container 5 seconds to start more than one process
 223         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 224         ;;
 225 stop|halt)
 226         lxc_exists $1
 227         lxc_stop $1
 228         ;;
 229 reload|force-reload|restart|reboot)
 230         lxc_kill $1 -SIGINT
 231         ;;
 232 watchdog)
 233         lxc_watchdog $1
 234         ;;
 235 boot)
 236         echo $1 > /var/lib/lxc/$1/on_boot
 237         ;;
 238 disable)
 239         echo -n > /var/lib/lxc/$1/on_boot
 240         ;;
 241 *)
 242         usage
 243         ;;
 244
 245 esac
 246
 247 }
 248
 249 command=$1
 250 test -z "$command" && usage
 251 test "$command" = "status" && lxc_status && exit
 252 shift
 253
 254 if [ -z "$1" ] ; then
 255         ls /var/lib/lxc/*/on_boot | while read path ; do
 256                 name=`echo $path | cut -d/ -f5`
 257                 if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
 258                         command_on_lxc $command $name
 259                 else
 260                         echo "# skip $command $name"
 261                 fi
 262         done
 263 else
 264         while [ ! -z "$1" ] ; do
 265                 command_on_lxc $command $1
 266                 shift
 267         done
 268 fi
 269