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


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() {
        lxc-ls -1 | sort -u | xargs -i lxc-info -n {} | sed "s/'//g" | while read name is status ; do
                boot="    "
                test -s /var/lib/lxc/$name/on_boot && boot="boot"
                echo "$name $status $boot $(lxc_rootfs $name)"
        done
}


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

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


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"

}


command_on_lxc() {
command=$1
shift

echo "# $command $1"

case "$command" in

start)
        lxc_exists $1
        cleanup_init_scripts $1
        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
        ;;
*)
        echo "Usage: $0 {start|stop|restart|status}" >&2
        exit 3
        ;;

esac

}

command=$1
shift

test "$command" = "status" && lxc_status && exit

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
20	dpavlin	181
21	dpavlin	177	which inotifywait >/dev/null \|\| apt-get install inotify-tools
22	root	175
23
24	dpavlin	181	lxc_exists() {
25			name=$1
26	root	175
27	dpavlin	181	if [ ! -e /var/lib/lxc/$name/config ] ; then
28			echo "Usage: $0 name"
29			lxc_status
30			exit 1
31			fi
32			}
33	root	175
34	dpavlin	179
35	dpavlin	181	lxc_rootfs() {
36			grep lxc.rootfs "/var/lib/lxc/$1/config" \| cut -d= -f2 \| sed 's/^ *//'
37			}
38	dpavlin	178
39	root	175
40	dpavlin	183	lxc_status() {
41			lxc-ls -1 \| sort -u \| xargs -i lxc-info -n {} \| sed "s/'//g" \| while read name is status ; do
42	dpavlin	189	boot=" "
43			test -s /var/lib/lxc/$name/on_boot && boot="boot"
44			echo "$name $status $boot $(lxc_rootfs $name)"
45	dpavlin	183	done
46			}
47
48
49	dpavlin	181	cleanup_init_scripts() {
50			rootfs=$(lxc_rootfs $1)
51	dpavlin	179
52	dpavlin	181	ls \
53			$rootfs/etc/rc?.d/*umountfs \
54			$rootfs/etc/rc?.d/*umountroot \
55			$rootfs/etc/rc?.d/hwclock \
56			2>/dev/null \| xargs -i rm -v {}
57			}
58	dpavlin	179
59	dpavlin	181
60			setup_inittab() {
61			rootfs=$(lxc_rootfs $1)
62	dpavlin	183	remove=$2
63			add=$3
64	dpavlin	181
65			# let container respond to kill -SIGPWR
66			inittab=$rootfs/etc/inittab
67	dpavlin	183	if ! grep "$add" ${inittab} >/dev/null ; then
68			grep -v "$remove" ${inittab} > ${inittab}.new
69			echo $add >> ${inittab}.new
70	dpavlin	181	mv ${inittab}.new ${inittab}
71	dpavlin	183	echo "$inittab modified with $add"
72	dpavlin	181	fi
73			}
74
75
76	dpavlin	189	lxc_log() {
77			echo `date +%Y-%m-%dT%H:%M:%S` $*
78			}
79
80
81	dpavlin	183	lxc_kill() {
82	dpavlin	181	name=$1
83	dpavlin	183	sig=$2
84	dpavlin	181
85			init_pid=`lxc-ps -C init -o pid \| grep "^$name" \| cut -d" " -f2-`
86			if [ -z "$init_pid" ] ; then
87			lxc-info -n $name
88			exit 1
89			fi
90	dpavlin	189	lxc_log "$name kill $sig $init_pid"
91	dpavlin	183	/bin/kill $sig $init_pid
92			}
93
94			lxc_stop() {
95	dpavlin	189	lxc_log "$name stop"
96	dpavlin	183	lxc_kill $name -SIGPWR
97	dpavlin	179	lxc-wait -n $name -s STOPPED
98	dpavlin	189	lxc_log "$name stoped"
99	dpavlin	183	# rm -f /var/lib/lxc/${name}/on_boot
100	dpavlin	181	}
101	dpavlin	179
102
103	dpavlin	181	lxc_start() {
104			name=$1
105
106			if ! lxc-info -n $name \| grep RUNNING ; then
107	dpavlin	189	lxc_log "$name start"
108	dpavlin	181	lxc-start -n $name -o /tmp/${name}.log -d
109			lxc-wait -n $name -s RUNNING
110			lxc-info -n $name
111	dpavlin	185	test -f /var/lib/lxc/${name}/on_boot \|\| echo $name > /var/lib/lxc/${name}/on_boot
112	dpavlin	181	fi
113			}
114
115			lxc_watchdog() {
116			name=$1
117			rootfs=$(lxc_rootfs $1)
118
119	root	175	while true; do
120			vps_utmp=${rootfs}/var/run/utmp
121	dpavlin	181	tasks=`wc -l < /cgroup/${name}/tasks`
122			test -z "$tasks" && exit 1
123			if [ "$tasks" -eq 1 ]; then
124	root	175
125			runlevel="$(runlevel ${vps_utmp})"
126	dpavlin	187	lxc_log "$name runlevel $runlevel"
127	root	175
128			case $runlevel in
129			N*)
130			# nothing for new boot state
131			;;
132			??0)
133	dpavlin	187	lxc_log "$name halt"
134	root	175	lxc-stop -n "${name}"
135	dpavlin	181	lxc-wait -n ${name} -s STOPPED
136	root	175	break
137			;;
138			??6)
139	dpavlin	187	lxc_log "$name reboot";
140	root	175	lxc-stop -n ${name}
141			lxc-wait -n ${name} -s STOPPED
142			lxc-start -d -n ${name} -o /tmp/${name}.log
143			;;
144			*)
145			# make sure vps is still running
146			state="$(lxc-info -n "${name}" \| sed -e 's/.* is //')"
147			[ "$state" = "RUNNING" ] \|\| break
148			;;
149			esac
150	dpavlin	181	else
151	dpavlin	187	lxc_log "$name $tasks tasks"
152	root	175	fi
153	dpavlin	182
154			# time of 5 minutes on it JUST IN CASE...
155			inotifywait -qqt 300 ${vps_utmp}
156	root	175	done
157
158	dpavlin	189	lxc_log "$name watchdog exited"
159	dpavlin	181
160			}
161
162
163	dpavlin	183	command_on_lxc() {
164			command=$1
165			shift
166	dpavlin	181
167	dpavlin	183	echo "# $command $1"
168
169			case "$command" in
170
171	dpavlin	181	start)
172	dpavlin	183	lxc_exists $1
173			cleanup_init_scripts $1
174			setup_inittab $1 ::power "p0::powerfail:/sbin/init 0"
175			setup_inittab $1 ::ctrlaltdel "p6::ctrlaltdel:/sbin/init 6"
176			lxc_start $1
177	dpavlin	186	# give container 5 seconds to start more than one process
178			( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
179	dpavlin	181	;;
180	dpavlin	183	stop\|halt)
181			lxc_exists $1
182			lxc_stop $1
183	dpavlin	181	;;
184	dpavlin	183	reload\|force-reload\|restart\|reboot)
185			lxc_kill $1 -SIGINT
186	dpavlin	181	;;
187			watchdog)
188	dpavlin	183	lxc_watchdog $1
189	dpavlin	181	;;
190	dpavlin	188	boot)
191			echo $1 > /var/lib/lxc/$1/on_boot
192			;;
193			disable)
194			echo -n > /var/lib/lxc/$1/on_boot
195			;;
196	dpavlin	181	*)
197			echo "Usage: $0 {start\|stop\|restart\|status}" >&2
198			exit 3
199			;;
200
201			esac
202
203	dpavlin	183	}
204
205			command=$1
206	dpavlin	189	shift
207	dpavlin	183
208			test "$command" = "status" && lxc_status && exit
209
210			if [ -z "$1" ] ; then
211			ls /var/lib/lxc/*/on_boot \| while read path ; do
212			name=`echo $path \| cut -d/ -f5`
213	dpavlin	189	if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
214			command_on_lxc $command $name
215			else
216			echo "# skip $command $name"
217			fi
218	dpavlin	183	done
219			else
220			while [ ! -z "$1" ] ; do
221			command_on_lxc $command $1
222			shift
223			done
224			fi
225