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_hostname() {
        inside=`cat $(lxc_rootfs $1)/etc/hostname`
        config=`grep lxc.utsname /var/lib/lxc/$name/config | cut -d= -f2`
        echo "$config [$inside]";
}

lxc_ip() {
        ( grep lxc.network.ipv4 /var/lib/lxc/$name/config | grep -v '^#' | cut -d= -f2 || \
        grep address $(lxc_rootfs $name)/etc/network/interfaces | grep -v '^#' | sed 's/.*address //' ) | \
        head -1 | \
        sed -e 's/ *//g' -e 's/\/.*$//'
}

lxc_status() {
        ( find /var/lib/lxc/ -name "config" | cut -d/ -f5 | sort -u | while read name ; do
                status=`lxc-info -n $name 2>/dev/null | grep state: | cut -d: -f2`
                boot="-"
                test -s /var/lib/lxc/$name/on_boot && boot="boot"
                echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
        done ) | column -t
}


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

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

        echo $1 > $rootfs/etc/hostname
        grep $1 $rootfs/etc/hosts || echo "$(lxc_ip $1) $1" >> $rootfs/etc/hosts
}


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

        # let container respond to kill -SIGPWR
        inittab=$rootfs/etc/inittab
        if test -e $inittab && ! grep "$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=$1
        rootfs=$(lxc_rootfs $1)

        if [ ! -e $rootfs ] ; then
                echo "ERROR $name rootfs $rootfs not found"
                return
        fi

        if ! lxc-info -n $name | grep RUNNING ; then
                lxc_log "$name start"
                dev=`df -P $rootfs | tail -1 | cut -d" " -f1`
                mount $dev -o remount,rw # fix debian upgrade which remounts dir ro
                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 $1)
run=$rootfs/var/run
test -L $run && run=$rootfs/`readlink $run` # recent Debian have symlink to /run
cd $run || echo "can't cd watchdog into $run"

while true; do
        tasks=`wc -l < /cgroup/${name}/tasks`
        stop_on=1 # init
        sulogins=`lxc-ps --name $name | grep sulogin | wc -l`
        if [ "$sulogins" -gt 0 ] ; then
                stop_on=`expr $stop_on + $sulogins`
        fi

        test -z "$tasks" && exit 1
        if [ "$tasks" -eq $stop_on ]; then

                runlevel="$(runlevel utmp)"
                lxc_log "$name runlevel $runlevel"

                case $runlevel in
                N*)
                        # nothing for new boot state
                ;;
                ??0|unknown)
                        lxc_log "$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 $sulogins console"
        fi

        # time of 5 minutes on it JUST IN CASE...
        inotifywait -qqt 300 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	dpavlin	232	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	root	175
49	dpavlin	232	lxc_ip() {
50	dpavlin	235	( grep lxc.network.ipv4 /var/lib/lxc/$name/config \| grep -v '^#' \| cut -d= -f2 \|\| \
51			grep address $(lxc_rootfs $name)/etc/network/interfaces \| grep -v '^#' \| sed 's/.*address //' ) \| \
52			head -1 \| \
53	dpavlin	233	sed -e 's/ //g' -e 's/\/.$//'
54	dpavlin	232	}
55
56	dpavlin	183	lxc_status() {
57	dpavlin	271	( find /var/lib/lxc/ -name "config" \| cut -d/ -f5 \| sort -u \| while read name ; do
58			status=`lxc-info -n $name 2>/dev/null \| grep state: \| cut -d: -f2`
59	dpavlin	196	boot="-"
60	dpavlin	189	test -s /var/lib/lxc/$name/on_boot && boot="boot"
61	dpavlin	232	echo "$name $status $boot $(lxc_rootfs $name) $(lxc_ip $name) $(lxc_hostname $name)"
62	dpavlin	196	done ) \| column -t
63	dpavlin	183	}
64
65
66	dpavlin	181	cleanup_init_scripts() {
67			rootfs=$(lxc_rootfs $1)
68	dpavlin	179
69	dpavlin	181	ls \
70	dpavlin	234	$rootfs/etc/rc?.d/checkroot \
71	dpavlin	181	$rootfs/etc/rc?.d/*umountfs \
72			$rootfs/etc/rc?.d/*umountroot \
73			$rootfs/etc/rc?.d/hwclock \
74	root	218	$rootfs/etc/rc?.d/udev \
75	dpavlin	237	$rootfs/etc/rc?.d/checkfs \
76	dpavlin	181	2>/dev/null \| xargs -i rm -v {}
77	dpavlin	226
78			echo $1 > $rootfs/etc/hostname
79	dpavlin	233	grep $1 $rootfs/etc/hosts \|\| echo "$(lxc_ip $1) $1" >> $rootfs/etc/hosts
80	dpavlin	181	}
81	dpavlin	179
82	dpavlin	181
83			setup_inittab() {
84			rootfs=$(lxc_rootfs $1)
85	dpavlin	183	remove=$2
86			add=$3
87	dpavlin	181
88			# let container respond to kill -SIGPWR
89			inittab=$rootfs/etc/inittab
90	dpavlin	227	if test -e $inittab && ! grep "$add" ${inittab} >/dev/null ; then
91	dpavlin	183	grep -v "$remove" ${inittab} > ${inittab}.new
92			echo $add >> ${inittab}.new
93	dpavlin	181	mv ${inittab}.new ${inittab}
94	dpavlin	183	echo "$inittab modified with $add"
95	dpavlin	181	fi
96			}
97
98
99	dpavlin	189	lxc_log() {
100			echo `date +%Y-%m-%dT%H:%M:%S` $*
101			}
102
103
104	dpavlin	183	lxc_kill() {
105	dpavlin	181	name=$1
106	dpavlin	183	sig=$2
107	dpavlin	181
108			init_pid=`lxc-ps -C init -o pid \| grep "^$name" \| cut -d" " -f2-`
109			if [ -z "$init_pid" ] ; then
110			lxc-info -n $name
111			exit 1
112			fi
113	dpavlin	189	lxc_log "$name kill $sig $init_pid"
114	dpavlin	183	/bin/kill $sig $init_pid
115			}
116
117			lxc_stop() {
118	dpavlin	189	lxc_log "$name stop"
119	dpavlin	183	lxc_kill $name -SIGPWR
120	dpavlin	179	lxc-wait -n $name -s STOPPED
121	dpavlin	189	lxc_log "$name stoped"
122	dpavlin	183	# rm -f /var/lib/lxc/${name}/on_boot
123	dpavlin	181	}
124	dpavlin	179
125
126	dpavlin	181	lxc_start() {
127			name=$1
128	dpavlin	227	rootfs=$(lxc_rootfs $1)
129	dpavlin	181
130	dpavlin	227	if [ ! -e $rootfs ] ; then
131			echo "ERROR $name rootfs $rootfs not found"
132			return
133			fi
134
135	dpavlin	181	if ! lxc-info -n $name \| grep RUNNING ; then
136	dpavlin	189	lxc_log "$name start"
137	dpavlin	244	dev=`df -P $rootfs \| tail -1 \| cut -d" " -f1`
138			mount $dev -o remount,rw # fix debian upgrade which remounts dir ro
139	dpavlin	181	lxc-start -n $name -o /tmp/${name}.log -d
140			lxc-wait -n $name -s RUNNING
141			lxc-info -n $name
142	dpavlin	185	test -f /var/lib/lxc/${name}/on_boot \|\| echo $name > /var/lib/lxc/${name}/on_boot
143	dpavlin	181	fi
144			}
145
146			lxc_watchdog() {
147			name=$1
148			rootfs=$(lxc_rootfs $1)
149	root	264	run=$rootfs/var/run
150			test -L $run && run=$rootfs/`readlink $run` # recent Debian have symlink to /run
151			cd $run \|\| echo "can't cd watchdog into $run"
152	dpavlin	181
153	root	175	while true; do
154	dpavlin	181	tasks=`wc -l < /cgroup/${name}/tasks`
155	dpavlin	236	stop_on=1 # init
156	dpavlin	245	sulogins=`lxc-ps --name $name \| grep sulogin \| wc -l`
157	dpavlin	236	if [ "$sulogins" -gt 0 ] ; then
158			stop_on=`expr $stop_on + $sulogins`
159			fi
160
161	dpavlin	181	test -z "$tasks" && exit 1
162	dpavlin	236	if [ "$tasks" -eq $stop_on ]; then
163	root	175
164	root	264	runlevel="$(runlevel utmp)"
165	dpavlin	187	lxc_log "$name runlevel $runlevel"
166	root	175
167			case $runlevel in
168			N*)
169			# nothing for new boot state
170			;;
171	root	264	??0\|unknown)
172	dpavlin	187	lxc_log "$name halt"
173	root	175	lxc-stop -n "${name}"
174	dpavlin	181	lxc-wait -n ${name} -s STOPPED
175	root	175	break
176			;;
177			??6)
178	dpavlin	187	lxc_log "$name reboot";
179	root	175	lxc-stop -n ${name}
180			lxc-wait -n ${name} -s STOPPED
181			lxc-start -d -n ${name} -o /tmp/${name}.log
182			;;
183			*)
184			# make sure vps is still running
185			state="$(lxc-info -n "${name}" \| sed -e 's/.* is //')"
186			[ "$state" = "RUNNING" ] \|\| break
187			;;
188			esac
189	dpavlin	181	else
190	dpavlin	236	lxc_log "$name $tasks tasks $sulogins console"
191	root	175	fi
192	dpavlin	182
193			# time of 5 minutes on it JUST IN CASE...
194	root	264	inotifywait -qqt 300 utmp
195	root	175	done
196
197	dpavlin	189	lxc_log "$name watchdog exited"
198	dpavlin	181
199			}
200
201
202	root	190	usage() {
203			echo "Usage: $0 {start\|stop\|restart\|status\|boot\|disable} [name name ... ]" >&2
204			exit 3
205			}
206
207	dpavlin	183	command_on_lxc() {
208			command=$1
209			shift
210	dpavlin	181
211	dpavlin	183	echo "# $command $1"
212
213			case "$command" in
214
215	dpavlin	181	start)
216	dpavlin	183	lxc_exists $1
217			cleanup_init_scripts $1
218	dpavlin	193	setup_inittab $1 ":respawn:/sbin/getty.*tty1" "c1:12345:respawn:/sbin/getty 38400 tty1 linux"
219			setup_inittab $1 "::power" "p0::powerfail:/sbin/init 0"
220			setup_inittab $1 "::ctrlaltdel" "p6::ctrlaltdel:/sbin/init 6"
221	dpavlin	183	lxc_start $1
222	dpavlin	186	# give container 5 seconds to start more than one process
223			( sleep 5 ; nohup $0 watchdog $1 >> /tmp/$1.log 2>/dev/null ) &
224	dpavlin	181	;;
225	dpavlin	183	stop\|halt)
226			lxc_exists $1
227			lxc_stop $1
228	dpavlin	181	;;
229	dpavlin	183	reload\|force-reload\|restart\|reboot)
230			lxc_kill $1 -SIGINT
231	dpavlin	181	;;
232			watchdog)
233	dpavlin	183	lxc_watchdog $1
234	dpavlin	181	;;
235	dpavlin	188	boot)
236			echo $1 > /var/lib/lxc/$1/on_boot
237			;;
238			disable)
239			echo -n > /var/lib/lxc/$1/on_boot
240			;;
241	dpavlin	181	*)
242	root	190	usage
243	dpavlin	181	;;
244
245			esac
246
247	dpavlin	183	}
248
249			command=$1
250	root	190	test -z "$command" && usage
251			test "$command" = "status" && lxc_status && exit
252	dpavlin	189	shift
253	dpavlin	183
254			if [ -z "$1" ] ; then
255			ls /var/lib/lxc/*/on_boot \| while read path ; do
256			name=`echo $path \| cut -d/ -f5`
257	dpavlin	189	if [ "$command" != "start" -o "$command" = "start" -a -s $path ] ; then
258			command_on_lxc $command $name
259			else
260			echo "# skip $command $name"
261			fi
262	dpavlin	183	done
263			else
264			while [ ! -z "$1" ] ; do
265			command_on_lxc $command $1
266			shift
267			done
268			fi
269