recepies/lxc/lxc-watchdog.sh

#!/bin/sh

# lxc-watchdog.sh
#
# Dobrica Pavlinusic <dpavlin@rot13.org> 2010-03-15
#
# this script can be used to start/stop Linux containers 
# using clever inotify hack to monitor halt/reboot from
# Tony Risinger posted 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
                on_boot="       "
                test -s /var/lib/lxc/$name/on_boot && on_boot="on_boot"
                echo "$name $status $on_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_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 $name
                exit 1
        fi
        echo "$name kill $sig $init_pid"
        /bin/kill $sig $init_pid
}

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


lxc_start() {
        name=$1

        if ! lxc-info -n $name | grep RUNNING ; then
                echo "$name start"
                lxc-start -n $name -o /tmp/${name}.log -d
                lxc-wait  -n $name -s RUNNING
                lxc-info  -n $name
                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 ${vps_utmp})"
                echo `date +%Y-%m-%dT%H:%M:%S` "$name runlevel $runlevel"

                case $runlevel in
                N*)
                        # nothing for new boot state
                ;;
                ??0)
                        echo "$name halt"
                        lxc-stop -n "${name}"
                        lxc-wait -n ${name} -s STOPPED
                        break
                ;;
                ??6)
                        echo "$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
                echo "# $name $tasks tasks"
        fi

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

echo "${name} 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
        ( 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
        ;;
*)
        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`
                command_on_lxc $command $name
        done
else
        while [ ! -z "$1" ] ; do
                command_on_lxc $command $1
                shift
        done
fi

1	dpavlin	181	#!/bin/sh
2	root	175
3	dpavlin	181	# lxc-watchdog.sh
4			#
5			# Dobrica Pavlinusic <dpavlin@rot13.org> 2010-03-15
6			#
7			# this script can be used to start/stop Linux containers
8			# using clever inotify hack to monitor halt/reboot from
9			# Tony Risinger posted to lxc-users mailing list
10			#
11	root	175	# http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00074.html
12
13	dpavlin	181
14	dpavlin	177	which inotifywait >/dev/null \|\| apt-get install inotify-tools
15	root	175
16
17	dpavlin	181	lxc_exists() {
18			name=$1
19	root	175
20	dpavlin	181	if [ ! -e /var/lib/lxc/$name/config ] ; then
21			echo "Usage: $0 name"
22			lxc_status
23			exit 1
24			fi
25			}
26	root	175
27	dpavlin	179
28	dpavlin	181	lxc_rootfs() {
29			grep lxc.rootfs "/var/lib/lxc/$1/config" \| cut -d= -f2 \| sed 's/^ *//'
30			}
31	dpavlin	178
32	root	175
33	dpavlin	183	lxc_status() {
34			lxc-ls -1 \| sort -u \| xargs -i lxc-info -n {} \| sed "s/'//g" \| while read name is status ; do
35			on_boot=" "
36			test -s /var/lib/lxc/$name/on_boot && on_boot="on_boot"
37			echo "$name $status $on_boot $(lxc_rootfs $name)"
38			done
39			}
40
41
42	dpavlin	181	cleanup_init_scripts() {
43			rootfs=$(lxc_rootfs $1)
44	dpavlin	179
45	dpavlin	181	ls \
46			$rootfs/etc/rc?.d/*umountfs \
47			$rootfs/etc/rc?.d/*umountroot \
48			$rootfs/etc/rc?.d/hwclock \
49			2>/dev/null \| xargs -i rm -v {}
50			}
51	dpavlin	179
52	dpavlin	181
53			setup_inittab() {
54			rootfs=$(lxc_rootfs $1)
55	dpavlin	183	remove=$2
56			add=$3
57	dpavlin	181
58			# let container respond to kill -SIGPWR
59			inittab=$rootfs/etc/inittab
60	dpavlin	183	if ! grep "$add" ${inittab} >/dev/null ; then
61			grep -v "$remove" ${inittab} > ${inittab}.new
62			echo $add >> ${inittab}.new
63	dpavlin	181	mv ${inittab}.new ${inittab}
64	dpavlin	183	echo "$inittab modified with $add"
65	dpavlin	181	fi
66			}
67
68
69	dpavlin	183	lxc_kill() {
70	dpavlin	181	name=$1
71	dpavlin	183	sig=$2
72	dpavlin	181
73			init_pid=`lxc-ps -C init -o pid \| grep "^$name" \| cut -d" " -f2-`
74			if [ -z "$init_pid" ] ; then
75			lxc-info -n $name
76			exit 1
77			fi
78	dpavlin	183	echo "$name kill $sig $init_pid"
79			/bin/kill $sig $init_pid
80			}
81
82			lxc_stop() {
83			lxc_kill $name -SIGPWR
84	dpavlin	179	lxc-wait -n $name -s STOPPED
85	dpavlin	183	# rm -f /var/lib/lxc/${name}/on_boot
86	dpavlin	181	}
87	dpavlin	179
88
89	dpavlin	181	lxc_start() {
90			name=$1
91
92			if ! lxc-info -n $name \| grep RUNNING ; then
93			echo "$name start"
94			lxc-start -n $name -o /tmp/${name}.log -d
95			lxc-wait -n $name -s RUNNING
96			lxc-info -n $name
97	dpavlin	183	echo $name > /var/lib/lxc/${name}/on_boot
98	dpavlin	181	fi
99			}
100
101			lxc_watchdog() {
102			name=$1
103			rootfs=$(lxc_rootfs $1)
104
105	root	175	while true; do
106			vps_utmp=${rootfs}/var/run/utmp
107	dpavlin	181	tasks=`wc -l < /cgroup/${name}/tasks`
108			test -z "$tasks" && exit 1
109			if [ "$tasks" -eq 1 ]; then
110	root	175
111			runlevel="$(runlevel ${vps_utmp})"
112	dpavlin	183	echo `date +%Y-%m-%dT%H:%M:%S` "$name runlevel $runlevel"
113	root	175
114			case $runlevel in
115			N*)
116			# nothing for new boot state
117			;;
118			??0)
119			echo "$name halt"
120			lxc-stop -n "${name}"
121	dpavlin	181	lxc-wait -n ${name} -s STOPPED
122	root	175	break
123			;;
124			??6)
125			echo "$name reboot";
126			lxc-stop -n ${name}
127			lxc-wait -n ${name} -s STOPPED
128			lxc-start -d -n ${name} -o /tmp/${name}.log
129			;;
130			*)
131			# make sure vps is still running
132			state="$(lxc-info -n "${name}" \| sed -e 's/.* is //')"
133			[ "$state" = "RUNNING" ] \|\| break
134			;;
135			esac
136	dpavlin	181	else
137			echo "# $name $tasks tasks"
138	root	175	fi
139	dpavlin	182
140			# time of 5 minutes on it JUST IN CASE...
141			inotifywait -qqt 300 ${vps_utmp}
142	root	175	done
143
144	dpavlin	181	echo "${name} exited"
145
146			}
147
148
149	dpavlin	183	command_on_lxc() {
150			command=$1
151			shift
152	dpavlin	181
153	dpavlin	183	echo "# $command $1"
154
155			case "$command" in
156
157	dpavlin	181	start)
158	dpavlin	183	lxc_exists $1
159			cleanup_init_scripts $1
160			setup_inittab $1 ::power "p0::powerfail:/sbin/init 0"
161			setup_inittab $1 ::ctrlaltdel "p6::ctrlaltdel:/sbin/init 6"
162			lxc_start $1
163			( nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
164	dpavlin	181	;;
165	dpavlin	183	stop\|halt)
166			lxc_exists $1
167			lxc_stop $1
168	dpavlin	181	;;
169	dpavlin	183	reload\|force-reload\|restart\|reboot)
170			lxc_kill $1 -SIGINT
171	dpavlin	181	;;
172			watchdog)
173	dpavlin	183	lxc_watchdog $1
174	dpavlin	181	;;
175			*)
176			echo "Usage: $0 {start\|stop\|restart\|status}" >&2
177			exit 3
178			;;
179
180			esac
181
182	dpavlin	183	}
183
184			command=$1
185			shift
186
187			test "$command" = "status" && lxc_status && exit
188
189			if [ -z "$1" ] ; then
190			ls /var/lib/lxc/*/on_boot \| while read path ; do
191			name=`echo $path \| cut -d/ -f5`
192			command_on_lxc $command $name
193			done
194			else
195			while [ ! -z "$1" ] ; do
196			command_on_lxc $command $1
197			shift
198			done
199			fi
200