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