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