kill child processes of addon-update on stop

cluster/saltbase/salt/kube-addons/initd

@@ -16,9 +16,9 @@
 PATH=/sbin:/usr/sbin:/bin:/usr/bin
 DESC="Kubernetes Addon Object Manager"
 NAME=kube-addons
-DAEMON_LOG_FILE=/var/log/$NAME.log
-PIDFILE=/var/run/$NAME.pid
-SCRIPTNAME=/etc/init.d/$NAME
+DAEMON_LOG_FILE=/var/log/${NAME}.log
+PIDFILE=/var/run/${NAME}.pid
+SCRIPTNAME=/etc/init.d/${NAME}
 KUBE_ADDONS_SH=/etc/kubernetes/kube-addons.sh
 
 # Define LSB log_* functions.
@@ -34,9 +34,10 @@ KUBE_ADDONS_SH=/etc/kubernetes/kube-addons.sh
 #
 do_start()
 {
-    ${KUBE_ADDONS_SH} </dev/null >>${DAEMON_LOG_FILE} 2>&1 &
-    echo $! > ${PIDFILE}
-    disown
+    # use setsid to make sure the new daemon has its own group (I suppose
+    # start-stop-daemon does create a process group, but let's stay on the
+    # safe side).
+    setsid start-stop-daemon --start --verbose --background --no-close --make-pidfile --pidfile "${PIDFILE}" --startas "${KUBE_ADDONS_SH}" </dev/null >> ${DAEMON_LOG_FILE} 2>&1
 }
 
 #
@@ -44,14 +45,38 @@ do_start()
 #
 do_stop()
 {
-    kill $(cat ${PIDFILE})
-    rm ${PIDFILE}
+    # start-stop-daemon is not used because we have to stop all children
+    # limitations:
+    #  - stop does not work if the pid file is missing
+    #  - stop does not work if the daemon process is missing (children will not
+    #    be killed)
+    # This is sufficient - remaining processes will end after a while.
+
+    local pid
+    pid=$(cat "${PIDFILE}" 2> /dev/null)
+    if [[ $? != 0 ]]; then
+        return 1
+    fi
+    local pgrp
+    # find the process group for the service and kill entire group
+    #  o - output format: pgpg - process group
+    pgrp=$(ps --no-headers --pid "${pid}" -o pgrp 2>/dev/null)
+    if [[ $? != 0 ]] || [[ "${pgrp}" == "" ]]; then
+        return 1
+    fi
+    pgrp=$(echo -e ${pgrp})   # strip whitespaces (that's why there are no quotes around pgrp)
+    # negative pid is for killing entire group
+    kill -- -${pgrp} 2> /dev/null
+    if [[ $? != 0 ]]; then
+        return 2
+    fi
+    rm -f "${PIDFILE}"
     return
 }
 
 case "$1" in
   start)
-        log_daemon_msg "Starting $DESC" "$NAME"
+        log_daemon_msg "Starting ${DESC}" "${NAME}"
         do_start
         case "$?" in
                 0|1) log_end_msg 0 || exit 0 ;;
@@ -59,19 +84,19 @@ case "$1" in
         esac
         ;;
   stop)
-        log_daemon_msg "Stopping $DESC" "$NAME"
+        log_daemon_msg "Stopping ${DESC}" "${NAME}"
         do_stop
         case "$?" in
-                0|1) log_end_msg 0 ;;
-                2) exit 1 ;;
+                0|1) log_end_msg 0 || exit 0 ;;
+                2) log_end_msg 1 || exit 1 ;;
         esac
         ;;
   status)
-        status_of_proc -p $PIDFILE $KUBE_ADDONS_SH $NAME
+        status_of_proc -p "${PIDFILE}" "${KUBE_ADDONS_SH}" "${NAME}"
         ;;
 
   restart|force-reload)
-        log_daemon_msg "Restarting $DESC" "$NAME"
+        log_daemon_msg "Restarting ${DESC}" "${NAME}"
         do_stop
         case "$?" in
           0|1)
@@ -89,7 +114,7 @@ case "$1" in
         esac
         ;;
   *)
-        echo "Usage: $SCRIPTNAME {start|stop|status|restart|force-reload}" >&2
+        echo "Usage: ${SCRIPTNAME} {start|stop|status|restart|force-reload}" >&2
         exit 3
         ;;
 esac

cluster/saltbase/salt/kube-addons/kube-addon-update.sh

@@ -162,7 +162,11 @@ function wait-for-jobs() {
     local rv=0
     local pid
     for pid in $(jobs -p); do
-        wait ${pid} || (rv=1; log ERR "error in pid ${pid}")
+        wait ${pid}
+        if [[ $? -ne 0 ]]; then
+            rv=1;
+            log ERR "error in pid ${pid}"
+        fi
         log DB2 "pid ${pid} completed, current error code: ${rv}"
     done
     return ${rv}