recepies/lxc/lxc-watchdog.sh

#! /bin/sh
### BEGIN INIT INFO
# Provides:          lxc-watchdog
# Required-Start:    $remote_fs $named $network $time
# Required-Stop:     $remote_fs $named $network
# Required-Start:    
# Required-Stop:     
# Default-Start:     2 3 4 5
# Default-Stop:      0 1 6
# Short-Description: Manage Linux Containers startup/shutdown
# Description:       Uses clever inotify hack to monitor container's
#                    halt/reboot events watching /var/run/utmp
### END INIT INFO

# Author: Dobrica Pavlinusic <dpavlin@rot13.org>
#
# based on Tony Risinger post to lxc-users mailing list
# http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00074.html
#
# Install with:
# ln -sf /srv/sysadmin-cookbook/recepies/lxc/lxc-watchdog.sh /etc/init.d/lxc-watchdog
# update-rc.d lxc-watchdog defaults


which inotifywait >/dev/null || apt-get install inotify-tools


lxc_exists() {
        name=$1

        if [ ! -e /var/lib/lxc/$name/config ] ; then
                echo "Usage: $0 name"
                lxc_status
                exit 1
        fi
}


lxc_rootfs() {
        grep '^ *lxc\.rootfs *=' "/var/lib/lxc/$1/config" | cut -d= -f2 | sed 's/^ *//'
}


lxc_status() {
        ( 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
                boot="-"
                hostname=`cat $(lxc_rootfs $name)/etc/hostname`
                ip=`grep address $(lxc_rootfs $name)/etc/network/interfaces | sed 's/.*address //'`
                test -s /var/lib/lxc/$name/on_boot && boot="boot"
                echo "$name $status $boot $(lxc_rootfs $name) $ip $hostname"
        done ) | column -t
}


cleanup_init_scripts() {
        rootfs=$(lxc_rootfs $1)

        ls \
                $rootfs/etc/rc?.d/*umountfs \
                $rootfs/etc/rc?.d/*umountroot \
                $rootfs/etc/rc?.d/*hwclock* \
                $rootfs/etc/rc?.d/*udev* \
        2>/dev/null | xargs -i rm -v {}

        echo $1 > $rootfs/etc/hostname
}


setup_inittab() {
        rootfs=$(lxc_rootfs $1)
        remove=$2
        add=$3

        # let container respond to kill -SIGPWR
        inittab=$rootfs/etc/inittab
        if ! grep "$add" ${inittab} >/dev/null ; then
                grep -v "$remove" ${inittab} > ${inittab}.new
                echo $add >> ${inittab}.new
                mv ${inittab}.new ${inittab}
                echo "$inittab modified with $add"
        fi
}


lxc_log() {
        echo `date +%Y-%m-%dT%H:%M:%S` $*
}


lxc_kill() {
        name=$1
        sig=$2

        init_pid=`lxc-ps -C init -o pid | grep "^$name" | cut -d" " -f2-`
        if [ -z "$init_pid" ] ; then
                lxc-info -n $name
                exit 1
        fi
        lxc_log "$name kill $sig $init_pid"
        /bin/kill $sig $init_pid
}

lxc_stop() {
        lxc_log "$name stop"
        lxc_kill $name -SIGPWR
        lxc-wait -n $name -s STOPPED
        lxc_log "$name stoped"
#       rm -f /var/lib/lxc/${name}/on_boot
}


lxc_start() {
        name=$1

        if ! lxc-info -n $name | grep RUNNING ; then
                lxc_log "$name start"
                lxc-start -n $name -o /tmp/${name}.log -d
                lxc-wait  -n $name -s RUNNING
                lxc-info  -n $name
                test -f /var/lib/lxc/${name}/on_boot || echo $name > /var/lib/lxc/${name}/on_boot
        fi
}

lxc_watchdog() {
name=$1
rootfs=$(lxc_rootfs $1)

while true; do
        vps_utmp=${rootfs}/var/run/utmp
        tasks=`wc -l < /cgroup/${name}/tasks`
        test -z "$tasks" && exit 1
        if [ "$tasks" -eq 1 ]; then

                runlevel="$(runlevel ${vps_utmp})"
                lxc_log "$name runlevel $runlevel"

                case $runlevel in
                N*)
                        # nothing for new boot state
                ;;
                ??0)
                        lxc_log "$name halt"
                        lxc-stop -n "${name}"
                        lxc-wait -n ${name} -s STOPPED
                        break
                ;;
                ??6)
                        lxc_log "$name reboot";
                        lxc-stop -n ${name}
                        lxc-wait -n ${name} -s STOPPED
                        lxc-start -d -n ${name} -o /tmp/${name}.log
                ;;
                *)
                        # make sure vps is still running
                        state="$(lxc-info -n "${name}" | sed -e 's/.* is //')"
                        [ "$state" = "RUNNING" ] || break
                ;;
                esac
        else
                lxc_log "$name $tasks tasks"
        fi

        # time of 5 minutes on it JUST IN CASE...
        inotifywait -qqt 300 ${vps_utmp}
done

lxc_log "$name watchdog exited"

}


usage() {
        echo "Usage: $0 {start|stop|restart|status|boot|disable} [name name ... ]" >&2
        exit 3
}

command_on_lxc() {
command=$1
shift

echo "# $command $1"

case "$command" in

start)
        lxc_exists $1
        cleanup_init_scripts $1
        setup_inittab $1 ":respawn:/sbin/getty.*tty1"   "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
        setup_inittab $1 "::power"                      "p0::powerfail:/sbin/init 0"
        setup_inittab $1 "::ctrlaltdel"                 "p6::ctrlaltdel:/sbin/init 6"
        lxc_start $1
        # give container 5 seconds to start more than one process
        ( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
        ;;
stop|halt)
        lxc_exists $1
        lxc_stop $1
        ;;
reload|force-reload|restart|reboot)
        lxc_kill $1 -SIGINT
        ;;
watchdog)
        lxc_watchdog $1
        ;;
boot)
        echo $1 > /var/lib/lxc/$1/on_boot
        ;;
disable)
        echo -n > /var/lib/lxc/$1/on_boot
        ;;
*)
        usage
        ;;

esac

}

command=$1
test -z "$command" && usage
test "$command" = "status" && lxc_status && exit
shift

if [ -z "$1" ] ; then
        ls /var/lib/lxc/*/on_boot | while read path ; do
                name=`echo $path | cut -d/ -f5`
                if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
                        command_on_lxc $command $name
                else
                        echo "# skip $command $name"
                fi
        done
else
        while [ ! -z "$1" ] ; do
                command_on_lxc $command $1
                shift
        done
fi

1	dpavlin	184	#! /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	root	175
15	dpavlin	184	# Author: Dobrica Pavlinusic <dpavlin@rot13.org>
16	dpavlin	181	#
17	dpavlin	184	# based on Tony Risinger post to lxc-users mailing list
18	root	175	# http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00074.html
19	root	190	#
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	root	175
24	dpavlin	181
25	dpavlin	177	which inotifywait >/dev/null \|\| apt-get install inotify-tools
26	root	175
27
28	dpavlin	181	lxc_exists() {
29			name=$1
30	root	175
31	dpavlin	181	if [ ! -e /var/lib/lxc/$name/config ] ; then
32			echo "Usage: $0 name"
33			lxc_status
34			exit 1
35			fi
36			}
37	root	175
38	dpavlin	179
39	dpavlin	181	lxc_rootfs() {
40	dpavlin	194	grep '^ lxc\.rootfs =' "/var/lib/lxc/$1/config" \| cut -d= -f2 \| sed 's/^ *//'
41	dpavlin	181	}
42	dpavlin	178
43	root	175
44	dpavlin	183	lxc_status() {
45	dpavlin	196	( 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
46			boot="-"
47	dpavlin	216	hostname=`cat $(lxc_rootfs $name)/etc/hostname`
48			ip=`grep address $(lxc_rootfs $name)/etc/network/interfaces \| sed 's/.*address //'`
49	dpavlin	189	test -s /var/lib/lxc/$name/on_boot && boot="boot"
50	dpavlin	216	echo "$name $status $boot $(lxc_rootfs $name) $ip $hostname"
51	dpavlin	196	done ) \| column -t
52	dpavlin	183	}
53
54
55	dpavlin	181	cleanup_init_scripts() {
56			rootfs=$(lxc_rootfs $1)
57	dpavlin	179
58	dpavlin	181	ls \
59			$rootfs/etc/rc?.d/*umountfs \
60			$rootfs/etc/rc?.d/*umountroot \
61			$rootfs/etc/rc?.d/hwclock \
62	root	218	$rootfs/etc/rc?.d/udev \
63	dpavlin	181	2>/dev/null \| xargs -i rm -v {}
64	dpavlin	226
65			echo $1 > $rootfs/etc/hostname
66	dpavlin	181	}
67	dpavlin	179
68	dpavlin	181
69			setup_inittab() {
70			rootfs=$(lxc_rootfs $1)
71	dpavlin	183	remove=$2
72			add=$3
73	dpavlin	181
74			# let container respond to kill -SIGPWR
75			inittab=$rootfs/etc/inittab
76	dpavlin	183	if ! grep "$add" ${inittab} >/dev/null ; then
77			grep -v "$remove" ${inittab} > ${inittab}.new
78			echo $add >> ${inittab}.new
79	dpavlin	181	mv ${inittab}.new ${inittab}
80	dpavlin	183	echo "$inittab modified with $add"
81	dpavlin	181	fi
82			}
83
84
85	dpavlin	189	lxc_log() {
86			echo `date +%Y-%m-%dT%H:%M:%S` $*
87			}
88
89
90	dpavlin	183	lxc_kill() {
91	dpavlin	181	name=$1
92	dpavlin	183	sig=$2
93	dpavlin	181
94			init_pid=`lxc-ps -C init -o pid \| grep "^$name" \| cut -d" " -f2-`
95			if [ -z "$init_pid" ] ; then
96			lxc-info -n $name
97			exit 1
98			fi
99	dpavlin	189	lxc_log "$name kill $sig $init_pid"
100	dpavlin	183	/bin/kill $sig $init_pid
101			}
102
103			lxc_stop() {
104	dpavlin	189	lxc_log "$name stop"
105	dpavlin	183	lxc_kill $name -SIGPWR
106	dpavlin	179	lxc-wait -n $name -s STOPPED
107	dpavlin	189	lxc_log "$name stoped"
108	dpavlin	183	# rm -f /var/lib/lxc/${name}/on_boot
109	dpavlin	181	}
110	dpavlin	179
111
112	dpavlin	181	lxc_start() {
113			name=$1
114
115			if ! lxc-info -n $name \| grep RUNNING ; then
116	dpavlin	189	lxc_log "$name start"
117	dpavlin	181	lxc-start -n $name -o /tmp/${name}.log -d
118			lxc-wait -n $name -s RUNNING
119			lxc-info -n $name
120	dpavlin	185	test -f /var/lib/lxc/${name}/on_boot \|\| echo $name > /var/lib/lxc/${name}/on_boot
121	dpavlin	181	fi
122			}
123
124			lxc_watchdog() {
125			name=$1
126			rootfs=$(lxc_rootfs $1)
127
128	root	175	while true; do
129			vps_utmp=${rootfs}/var/run/utmp
130	dpavlin	181	tasks=`wc -l < /cgroup/${name}/tasks`
131			test -z "$tasks" && exit 1
132			if [ "$tasks" -eq 1 ]; then
133	root	175
134			runlevel="$(runlevel ${vps_utmp})"
135	dpavlin	187	lxc_log "$name runlevel $runlevel"
136	root	175
137			case $runlevel in
138			N*)
139			# nothing for new boot state
140			;;
141			??0)
142	dpavlin	187	lxc_log "$name halt"
143	root	175	lxc-stop -n "${name}"
144	dpavlin	181	lxc-wait -n ${name} -s STOPPED
145	root	175	break
146			;;
147			??6)
148	dpavlin	187	lxc_log "$name reboot";
149	root	175	lxc-stop -n ${name}
150			lxc-wait -n ${name} -s STOPPED
151			lxc-start -d -n ${name} -o /tmp/${name}.log
152			;;
153			*)
154			# make sure vps is still running
155			state="$(lxc-info -n "${name}" \| sed -e 's/.* is //')"
156			[ "$state" = "RUNNING" ] \|\| break
157			;;
158			esac
159	dpavlin	181	else
160	dpavlin	187	lxc_log "$name $tasks tasks"
161	root	175	fi
162	dpavlin	182
163			# time of 5 minutes on it JUST IN CASE...
164			inotifywait -qqt 300 ${vps_utmp}
165	root	175	done
166
167	dpavlin	189	lxc_log "$name watchdog exited"
168	dpavlin	181
169			}
170
171
172	root	190	usage() {
173			echo "Usage: $0 {start\|stop\|restart\|status\|boot\|disable} [name name ... ]" >&2
174			exit 3
175			}
176
177	dpavlin	183	command_on_lxc() {
178			command=$1
179			shift
180	dpavlin	181
181	dpavlin	183	echo "# $command $1"
182
183			case "$command" in
184
185	dpavlin	181	start)
186	dpavlin	183	lxc_exists $1
187			cleanup_init_scripts $1
188	dpavlin	193	setup_inittab $1 ":respawn:/sbin/getty.*tty1" "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
189			setup_inittab $1 "::power" "p0::powerfail:/sbin/init 0"
190			setup_inittab $1 "::ctrlaltdel" "p6::ctrlaltdel:/sbin/init 6"
191	dpavlin	183	lxc_start $1
192	dpavlin	186	# give container 5 seconds to start more than one process
193			( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
194	dpavlin	181	;;
195	dpavlin	183	stop\|halt)
196			lxc_exists $1
197			lxc_stop $1
198	dpavlin	181	;;
199	dpavlin	183	reload\|force-reload\|restart\|reboot)
200			lxc_kill $1 -SIGINT
201	dpavlin	181	;;
202			watchdog)
203	dpavlin	183	lxc_watchdog $1
204	dpavlin	181	;;
205	dpavlin	188	boot)
206			echo $1 > /var/lib/lxc/$1/on_boot
207			;;
208			disable)
209			echo -n > /var/lib/lxc/$1/on_boot
210			;;
211	dpavlin	181	*)
212	root	190	usage
213	dpavlin	181	;;
214
215			esac
216
217	dpavlin	183	}
218
219			command=$1
220	root	190	test -z "$command" && usage
221			test "$command" = "status" && lxc_status && exit
222	dpavlin	189	shift
223	dpavlin	183
224			if [ -z "$1" ] ; then
225			ls /var/lib/lxc/*/on_boot \| while read path ; do
226			name=`echo $path \| cut -d/ -f5`
227	dpavlin	189	if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
228			command_on_lxc $command $name
229			else
230			echo "# skip $command $name"
231			fi
232	dpavlin	183	done
233			else
234			while [ ! -z "$1" ] ; do
235			command_on_lxc $command $1
236			shift
237			done
238			fi
239