eventscripts: Clean up 20.multipathd

Reduce the complexity, including the depth of background processes.

Signed-off-by: Martin Schwenke <martin@meltin.net>

(This used to be ctdb commit 49f077c475b078889ff0492fe7d567a64d6cb87c)

1 files changed, 44 insertions, 57 deletions

diff --git a/ctdb/config/events.d/20.multipathd b/ctdb/config/events.d/20.multipathd
index c2956d24a1..64748dae61 100755
--- a/ctdb/config/events.d/20.multipathd
+++ b/ctdb/config/events.d/20.multipathd
@@ -15,78 +15,65 @@ service_name="multipathd"
 
 loadconfig 
 
-[ -z "$CTDB_MONITOR_MPDEVICES" ] && {
-	exit 0
-}
+[ -n "$CTDB_MONITOR_MPDEVICES" ] || exit 0
 
 ctdb_setup_service_state_dir
 
-MPFAILURE="$service_state_dir/failure"
+multipath_fail="${service_state_dir}/fail"
 
 multipathd_check_background()
 {
-	for DEVICE in $CTDB_MONITOR_MPDEVICES; do
-		# check that we can see all devices
-		if [ -z "`multipath -ll $DEVICE`" ]; then
-			echo Device $DEVICE not known to multipathd
-			exit 1
-		fi
-
-		# check that all devices are active
-		if [ -z "`multipath -ll $DEVICE|grep prio|grep active`" ]; then
-			echo Device $DEVICE has no active paths
-			exit 1
-		fi
-	done
-	exit 0
+    for _device in $CTDB_MONITOR_MPDEVICES; do
+	# Check multipath knows about the device
+	_out=$(multipath -ll "$_device")
+	if [ -z "$_out" ] ; then
+	    echo "device \"${_device}\" not known to multipathd" >"$multipath_fail"
+	    exit 1
+	fi
+
+	# Check for at least 1 active path
+	if ! echo "$_out" | grep 'prio=.* status=active' >/dev/null 2>&1 ; then
+	    echo "multipath device \"${_device}\" has no active paths" >"$multipath_fail"
+	    exit 1
+	fi
+    done
+    exit 0
 }
 
 multipathd_check()
 {
-	# run the actual check in the background since the call to 
-	# multipath may block.
-	(
-	multipathd_check_background &
-	pid="$!"
-	timeleft=10
-
-	while [ $timeleft -gt 0 ]; do
-		timeleft=$(($timeleft - 1))
-
-		# see if the process still exists
-		kill -0 $pid > /dev/null 2>&1 || {
-			# it doesn't exist, grab its exit status
-			wait $pid
-			[ $? = 0 ] || {
-				echo "20.multipathd: multipath background update exited with status $?"
-				touch $MPFAILURE
-				exit 1
-			}
-			rm $MPFAILURE 2>/dev/null
-			exit 0
-		}
-		sleep 1
-	done
-	echo "20.multipathd: Callout to multipath checks hung."
-	touch $MPFAILURE
-	exit 1
-	) &
-
-	if [ -f $MPFAILURE ]; then
-		return 1
-	else
+    # Run the actual check in the background since the call to
+    # multipath may block
+    multipathd_check_background </dev/null >/dev/null 2>&1 &
+    _pid="$!"
+    _timeleft=10
+
+    while [ $_timeleft -gt 0 ]; do
+	_timeleft=$(($_timeleft - 1))
+
+	# see if the process still exists
+	kill -0 $_pid >/dev/null 2>&1 || {
+	    if wait $_pid ; then
 		return 0
-	fi
+	    else
+		echo -n "ERROR: "
+		cat "$multipath_fail"
+		rm -f "$multipath_fail"
+		return 1
+	    fi
+	}
+	sleep 1
+    done
+
+    echo "ERROR: callout to multipath checks hung"
+    # If hung then this probably won't work, but worth trying...
+    kill -9 $_pid >/dev/null 2>&1
+    return 1
 }
 
 case "$1" in
     monitor)
-	multipathd_check
-	[ "$?" = "0" ] || {
-		echo 20.multipathd: monitoring of multipathing failed
-		exit 1
-	}
-	exit 0
+	multipathd_check || die "multipath monitoring failed"
 	;;
 
     *)