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 $name 2>/dev/null | grep state: | cut -d: -f2`     # 0.7.5
  59                 test -z "$status" && status=`lxc-info -n $name | sed -e 's/^.* is //'` # 0.7.2
  60                 boot="-"
  61                 test -s /var/lib/lxc/$name/on_boot && boot="boot"
  62                 echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
  63         done ) | column -t
  64 }
  65
  66
  67 cleanup_init_scripts() {
  68         rootfs=$(lxc_rootfs $1)
  69
  70         ls \
  71                 $rootfs/etc/rc?.d/*checkroot* \
  72                 $rootfs/etc/rc?.d/*umountfs \
  73                 $rootfs/etc/rc?.d/*umountroot \
  74                 $rootfs/etc/rc?.d/*hwclock* \
  75                 $rootfs/etc/rc?.d/*udev* \
  76                 $rootfs/etc/rc?.d/*checkfs* \
  77         2>/dev/null | xargs -i rm -v {}
  78
  79         echo $1 > $rootfs/etc/hostname
  80         grep $1 $rootfs/etc/hosts || echo "$(lxc_ip $1) $1" >> $rootfs/etc/hosts
  81 }
  82
  83
  84 setup_inittab() {
  85         rootfs=$(lxc_rootfs $1)
  86         remove=$2
  87         add=$3
  88
  89         # let container respond to kill -SIGPWR
  90         inittab=$rootfs/etc/inittab
  91         if test -e $inittab && ! grep "$add" ${inittab} >/dev/null ; then
  92                 grep -v "$remove" ${inittab} > ${inittab}.new
  93                 echo $add >> ${inittab}.new
  94                 mv ${inittab}.new ${inittab}
  95                 echo "$inittab modified with $add"
  96         fi
  97 }
  98
  99
 100 lxc_log() {
 101         echo `date +%Y-%m-%dT%H:%M:%S` $*
 102 }
 103
 104
 105 lxc_kill() {
 106         name=$1
 107         sig=$2
 108
 109         ver=`lxc-version | cut -d: -f2 | sed 's/\.//g'`
 110         opts=''
 111         test $ver -ge 075 && opts='--'
 112
 113         init_pid=`lxc-ps $opts -C init -o pid | grep "^$name" | cut -d" " -f2-`
 114         if [ -z "$init_pid" ] ; then
 115                 lxc-info -n $name
 116                 exit 1
 117         fi
 118         lxc_log "$name kill $sig $init_pid"
 119         /bin/kill $sig $init_pid
 120 }
 121
 122 lxc_stop() {
 123         lxc_log "$name stop"
 124         lxc_kill $name -SIGPWR
 125         lxc-wait -n $name -s STOPPED
 126         lxc_log "$name stoped"
 127 #       rm -f /var/lib/lxc/${name}/on_boot
 128 }
 129
 130
 131 lxc_start() {
 132         name=$1
 133         rootfs=$(lxc_rootfs $1)
 134
 135         if [ ! -e $rootfs ] ; then
 136                 echo "ERROR $name rootfs $rootfs not found"
 137                 return
 138         fi
 139
 140         if ! lxc-info -n $name | grep RUNNING ; then
 141                 lxc_log "$name start"
 142                 dev=`df -P $rootfs | tail -1 | cut -d" " -f1`
 143                 mount $dev -o remount,rw # fix debian upgrade which remounts dir ro
 144                 lxc-start -n $name -o /tmp/${name}.log -d
 145                 lxc-wait  -n $name -s RUNNING
 146                 lxc-info  -n $name
 147                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 148         fi
 149 }
 150
 151 lxc_watchdog() {
 152 name=$1
 153 cgroup=$(mount -t cgroup | awk '{ print $3 }')
 154 test -d "$cgroup/lxc/$1" && cgroup="$cgroup/lxc"
 155 rootfs=$(lxc_rootfs $1)
 156 run=$rootfs/var/run
 157 test -L $run && run=$rootfs/`readlink $run` # recent Debian have symlink to /run
 158 cd $run || echo "can't cd watchdog into $run"
 159
 160 while true; do
 161         tasks=`wc -l < $cgroup/${name}/tasks`
 162         stop_on=1 # init
 163         sulogins=`lxc-ps --name $name | grep sulogin | wc -l`
 164         if [ "$sulogins" -gt 0 ] ; then
 165                 stop_on=`expr $stop_on + $sulogins`
 166         fi
 167
 168         test -z "$tasks" && exit 1
 169         if [ "$tasks" -eq $stop_on ]; then
 170
 171                 runlevel="$(runlevel utmp)"
 172                 lxc_log "$name runlevel $runlevel"
 173
 174                 case $runlevel in
 175                 N*)
 176                         # nothing for new boot state
 177                 ;;
 178                 ??0|unknown)
 179                         lxc_log "$name halt"
 180                         lxc-stop -n "${name}"
 181                         lxc-wait -n ${name} -s STOPPED
 182                         break
 183                 ;;
 184                 ??6)
 185                         lxc_log "$name reboot";
 186                         lxc-stop -n ${name}
 187                         lxc-wait -n ${name} -s STOPPED
 188                         lxc-start -d -n ${name} -o /tmp/${name}.log
 189                 ;;
 190                 *)
 191                         # make sure vps is still running
 192                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 193                         [ "$state" = "RUNNING" ] || break
 194                 ;;
 195                 esac
 196         else
 197                 lxc_log "$name $tasks tasks $sulogins console"
 198         fi
 199
 200         # time of 5 minutes on it JUST IN CASE...
 201         inotifywait -qqt 300 utmp
 202 done
 203
 204 lxc_log "$name watchdog exited"
 205
 206 }
 207
 208
 209 usage() {
 210         echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
 211         exit 3
 212 }
 213
 214 command_on_lxc() {
 215 command=$1
 216 shift
 217
 218 echo "# $command $1"
 219
 220 case "$command" in
 221
 222 start)
 223         lxc_exists $1
 224         cleanup_init_scripts $1
 225         setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
 226         setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
 227         setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
 228         lxc_start $1
 229         # give container 5 seconds to start more than one process
 230         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 231         ;;
 232 stop|halt)
 233         lxc_exists $1
 234         lxc_stop $1
 235         ;;
 236 reload|force-reload|restart|reboot)
 237         lxc_kill $1 -SIGINT
 238         ;;
 239 watchdog)
 240         lxc_watchdog $1
 241         ;;
 242 boot)
 243         echo $1 > /var/lib/lxc/$1/on_boot
 244         ;;
 245 disable)
 246         echo -n > /var/lib/lxc/$1/on_boot
 247         ;;
 248 *)
 249         usage
 250         ;;
 251
 252 esac
 253
 254 }
 255
 256 command=$1
 257 test -z "$command" && usage
 258 test "$command" = "status" && lxc_status && exit
 259 shift
 260
 261 if [ -z "$1" ] ; then
 262         ls /var/lib/lxc/*/on_boot | while read path ; do
 263                 name=`echo $path | cut -d/ -f5`
 264                 if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
 265                         command_on_lxc $command $name
 266                 else
 267                         echo "# skip $command $name"
 268                 fi
 269         done
 270 else
 271         while [ ! -z "$1" ] ; do
 272                 command_on_lxc $command $1
 273                 shift
 274         done
 275 fi
 276