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
  21 which inotifywait >/dev/null || apt-get install inotify-tools
  22
  23
  24 lxc_exists() {
  25         name=$1
  26
  27         if [ ! -e /var/lib/lxc/$name/config ] ; then
  28                 echo "Usage: $0 name"
  29                 lxc_status
  30                 exit 1
  31         fi
  32 }
  33
  34
  35 lxc_rootfs() {
  36         grep lxc.rootfs "/var/lib/lxc/$1/config" | cut -d= -f2 | sed 's/^ *//'
  37 }
  38
  39
  40 lxc_status() {
  41         lxc-ls -1 | sort -u | xargs -i lxc-info -n {} | sed "s/'//g" | while read name is status ; do
  42                 on_boot="       "
  43                 test -s /var/lib/lxc/$name/on_boot && on_boot="on_boot"
  44                 echo "$name $status $on_boot $(lxc_rootfs $name)"
  45         done
  46 }
  47
  48
  49 cleanup_init_scripts() {
  50         rootfs=$(lxc_rootfs $1)
  51
  52         ls \
  53                 $rootfs/etc/rc?.d/*umountfs \
  54                 $rootfs/etc/rc?.d/*umountroot \
  55                 $rootfs/etc/rc?.d/*hwclock* \
  56         2>/dev/null | xargs -i rm -v {}
  57 }
  58
  59
  60 setup_inittab() {
  61         rootfs=$(lxc_rootfs $1)
  62         remove=$2
  63         add=$3
  64
  65         # let container respond to kill -SIGPWR
  66         inittab=$rootfs/etc/inittab
  67         if ! grep "$add" ${inittab} >/dev/null ; then
  68                 grep -v "$remove" ${inittab} > ${inittab}.new
  69                 echo $add >> ${inittab}.new
  70                 mv ${inittab}.new ${inittab}
  71                 echo "$inittab modified with $add"
  72         fi
  73 }
  74
  75
  76 lxc_kill() {
  77         name=$1
  78         sig=$2
  79
  80         init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
  81         if [ -z "$init_pid" ] ; then
  82                 lxc-info -n $name
  83                 exit 1
  84         fi
  85         echo "$name kill $sig $init_pid"
  86         /bin/kill $sig $init_pid
  87 }
  88
  89 lxc_stop() {
  90         lxc_kill $name -SIGPWR
  91         lxc-wait -n $name -s STOPPED
  92 #       rm -f /var/lib/lxc/${name}/on_boot
  93 }
  94
  95
  96 lxc_start() {
  97         name=$1
  98
  99         if ! lxc-info -n $name | grep RUNNING ; then
 100                 echo "$name start"
 101                 lxc-start -n $name -o /tmp/${name}.log -d
 102                 lxc-wait  -n $name -s RUNNING
 103                 lxc-info  -n $name
 104                 test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
 105         fi
 106 }
 107
 108 lxc_log() {
 109         echo `date +%Y-%m-%dT%H:%M:%S` $*
 110 }
 111
 112 lxc_watchdog() {
 113 name=$1
 114 rootfs=$(lxc_rootfs $1)
 115
 116 while true; do
 117         vps_utmp=${rootfs}/var/run/utmp
 118         tasks=`wc -l < /cgroup/${name}/tasks`
 119         test -z "$tasks" && exit 1
 120         if [ "$tasks" -eq 1 ]; then
 121
 122                 runlevel="$(runlevel ${vps_utmp})"
 123                 lxc_log "$name runlevel $runlevel"
 124
 125                 case $runlevel in
 126                 N*)
 127                         # nothing for new boot state
 128                 ;;
 129                 ??0)
 130                         lxc_log "$name halt"
 131                         lxc-stop -n "${name}"
 132                         lxc-wait -n ${name} -s STOPPED
 133                         break
 134                 ;;
 135                 ??6)
 136                         lxc_log "$name reboot";
 137                         lxc-stop -n ${name}
 138                         lxc-wait -n ${name} -s STOPPED
 139                         lxc-start -d -n ${name} -o /tmp/${name}.log
 140                 ;;
 141                 *)
 142                         # make sure vps is still running
 143                         state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
 144                         [ "$state" = "RUNNING" ] || break
 145                 ;;
 146                 esac
 147         else
 148                 lxc_log "$name $tasks tasks"
 149         fi
 150
 151         # time of 5 minutes on it JUST IN CASE...
 152         inotifywait -qqt 300 ${vps_utmp}
 153 done
 154
 155 lxc_log "$name exited"
 156
 157 }
 158
 159
 160 command_on_lxc() {
 161 command=$1
 162 shift
 163
 164 echo "# $command $1"
 165
 166 case "$command" in
 167
 168 start)
 169         lxc_exists $1
 170         cleanup_init_scripts $1
 171         setup_inittab $1 ::power      "p0::powerfail:/sbin/init 0"
 172         setup_inittab $1 ::ctrlaltdel "p6::ctrlaltdel:/sbin/init 6"
 173         lxc_start $1
 174         # give container 5 seconds to start more than one process
 175         ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
 176         ;;
 177 stop|halt)
 178         lxc_exists $1
 179         lxc_stop $1
 180         ;;
 181 reload|force-reload|restart|reboot)
 182         lxc_kill $1 -SIGINT
 183         ;;
 184 watchdog)
 185         lxc_watchdog $1
 186         ;;
 187 boot)
 188         echo $1 > /var/lib/lxc/$1/on_boot
 189         ;;
 190 disable)
 191         echo -n > /var/lib/lxc/$1/on_boot
 192         ;;
 193 *)
 194         echo "Usage: $0 {start|stop|restart|status}" >&2
 195         exit 3
 196         ;;
 197
 198 esac
 199
 200 }
 201
 202 command=$1
 203
 204 test "$command" = "status" && lxc_status && exit
 205
 206 if [ -z "$1" ] ; then
 207         ls /var/lib/lxc/*/on_boot | while read path ; do
 208                 name=`echo $path | cut -d/ -f5`
 209                 command_on_lxc $command $name
 210         done
 211 else
 212         shift
 213         while [ ! -z "$1" ] ; do
 214                 command_on_lxc $command $1
 215                 shift
 216         done
 217 fi
 218