This is the mail archive of the cluster-cvs@sourceware.org mailing list for the cluster.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

cluster: STABLE3 - rgmanager: Optimize fork/clone during status checks


Gitweb:        http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=969408c96d1a4c3fae1435ca96a837d957b05cb9
Commit:        969408c96d1a4c3fae1435ca96a837d957b05cb9
Parent:        7378622847b57be25e19eb06f55a567e6be7a00b
Author:        Lon Hohberger <lhh@redhat.com>
AuthorDate:    Tue Mar 3 11:45:50 2009 -0500
Committer:     Lon Hohberger <lhh@redhat.com>
CommitterDate: Tue Mar 31 15:07:37 2009 -0400

rgmanager: Optimize fork/clone during status checks

* New option: quick_status trades off logging and verbosity
  for vastly improved performance.  This reduces or eliminates
  load spikes on machines with lots of file system resources
  mounted, but should only be used in such cases (because logging
  is disabled when using quick_status)

rhbz250718 (lots of analysis here)
rhbz487599 (RHEL4 bug)
rhbz250718 (RHEL5 bug)

Signed-off-by: Lon Hohberger <lhh@redhat.com>
---
 rgmanager/src/resources/fs.sh.in |  267 +++++++++++++++++---------------------
 1 files changed, 120 insertions(+), 147 deletions(-)

diff --git a/rgmanager/src/resources/fs.sh.in b/rgmanager/src/resources/fs.sh.in
index b00cf46..aa2f398 100644
--- a/rgmanager/src/resources/fs.sh.in
+++ b/rgmanager/src/resources/fs.sh.in
@@ -19,14 +19,58 @@ NO=1
 YES_STR="yes"
 INVALIDATEBUFFERS="/bin/true"
 
-# Grab nfs lock tricks if available
+#
+# Using a global to contain the return value saves
+# clone() operations.  This is important to reduce
+# resource consumption during status checks.
+#
+# There is no way to return a string from a function
+# in bash without cloning the process, which is exactly
+# what we are trying to avoid.  So, we have to resort
+# to using a dedicated global variable.  This one is
+# for the real_device() function below.
+#
+declare REAL_DEVICE
+
+#
+# Stub ocf_log function for when we are using 
+# quick_status, since ocf_log generally forks (and
+# sourcing ocf-shellfuncs forks -a lot-).
+#
+ocf_log()
+{
+	echo $*
+}
+
+#
+# Assume NFS_TRICKS are not available until we are
+# proved otherwise.
+#
 export NFS_TRICKS=1
-if [ -f "$(dirname $0)/svclib_nfslock" ]; then
-	. $(dirname $0)/svclib_nfslock
-	NFS_TRICKS=0
-fi
 
-. $(dirname $0)/ocf-shellfuncs
+#
+# Quick status doesn't fork() or clone() when using
+# device files directly.  (i.e. not symlinks, LABEL= or
+# UUID=
+#
+if [ "$1" = "status" -o "$1" = "monitor" ] &&
+   [ "$OCF_RESKEY_quick_status" = "1" ]; then
+	echo Using Quick Status
+
+	# XXX maybe we can make ocf-shellfuncs have a 'quick' mode too?
+	export OCF_SUCCESS=0
+	export OCF_ERR_GENERIC=1
+else
+	#
+	# Grab nfs lock tricks if available
+	#
+	if [ -f "$(dirname $0)/svclib_nfslock" ]; then
+		. $(dirname $0)/svclib_nfslock
+		NFS_TRICKS=0
+	fi
+
+	. $(dirname $0)/ocf-shellfuncs
+fi
 
 meta_data()
 {
@@ -98,20 +142,21 @@ meta_data()
 	    <content type="boolean"/>
         </parameter>
 
-	<!-- 
-        <parameter name="active_monitor">
+        <parameter name="quick_status">
             <longdesc lang="en">
-	    	If set, the cluster will spawn an active monitoring 
-		daemon which watches the ability to issue I/Os to the
-		file system.  Requires a file system with O_DIRECT
-		support.
+		Use quick status checks.  When set to 0 (the default), this
+		agent behaves normally.  When set to 1, this agent will not
+		log errors incurred or perform the file system accessibility
+		check (e.g. it will not try to read from/write to the file
+		system).  You should only set this to 1 if you have lots of
+		file systems on your cluster or you are seeing very high load
+		spikes as a direct result of this agent.
             </longdesc>
             <shortdesc lang="en">
-	    	Active Monitoring
+	    	Quick/brief status checks.
             </shortdesc>
 	    <content type="boolean"/>
         </parameter>
-	-->
 
 	<parameter name="self_fence">
 	    <longdesc lang="en">
@@ -240,30 +285,48 @@ verify_mountpoint()
 }
 
 
+#
+# This used to be called using $(...), but doing this causes bash
+# to set up a pipe and clone().  So, the output of this function is
+# stored in the global variable REAL_DEVICE, declared previously.
+#
 real_device()
 {
 	declare dev=$1
 	declare realdev
 
+	REAL_DEVICE=""
+
 	[ -z "$dev" ] && return $OCF_ERR_ARGS
 
+	# If our provided blockdev is a device, we are done
+	if [ -b "$dev" ]; then
+		REAL_DEVICE="$dev"
+	       	return $OCF_SUCCESS
+	fi
+
+	# Oops, we have a link.  Sorry, this is going to fork.
 	if [ -h "$dev" ]; then 
 		realdev=$(readlink -f $dev)
 		if [ $? -ne 0 ]; then
 			return $OCF_ERR_ARGS
 		fi
-		echo $realdev
+		REAL_DEVICE="$realdev"
 		return $OCF_SUCCESS
 	fi
 
-	if [ -b "$dev" ]; then
-		echo $dev
-	       	return $OCF_SUCCESS
+	# It's not a link, it's not a block device.  If it also
+	# does not match UUID= or LABEL=, then findfs is not 
+	# going to find anything useful, so we should quit now.
+	if [ "${dev/UUID=/}" = "$dev" ] &&
+	   [ "${dev/LABEL=/}" = "$dev" ]; then
+		return $OCF_ERR_GENERIC
 	fi
-		
+	
+	# When using LABEL= or UUID=, we can't save a fork.
 	realdev=$(findfs $dev 2> /dev/null)
 	if [ -n "$realdev" ] && [ -b "$realdev" ]; then
-		echo $realdev
+		REAL_DEVICE="$realdev"
 		return $OCF_SUCCESS
 	fi
 
@@ -280,7 +343,8 @@ verify_device()
 	       return $OCF_ERR_ARGS
 	fi
 
-	realdev=$(real_device $OCF_RESKEY_device)
+	real_device $OCF_RESKEY_device
+	realdev=$REAL_DEVICE
 	if [ -n "$realdev" ]; then
 		if [ "$realdev" != "$OCF_RESKEY_device" ]; then
 			ocf_log info "Specified $OCF_RESKEY_device maps to $realdev"
@@ -451,7 +515,7 @@ verify_all()
 mountInUse () {
 	typeset mp tmp_mp
 	typeset dev tmp_dev
-	typeset junk
+	typeset junka junkb junkc junkd
 
 	if [ $# -ne 2 ]; then
 		ocf_log err "Usage: mountInUse device mount_point".
@@ -461,7 +525,7 @@ mountInUse () {
 	dev=$1
 	mp=$2
 
-	while read tmp_dev tmp_mp junk; do
+	while read tmp_dev tmp_mp junka junkb junkc junkd; do
 		if [ -n "$tmp_dev" -a "$tmp_dev" = "$dev" ]; then
 			return $YES
 		fi
@@ -469,25 +533,13 @@ mountInUse () {
 		if [ -n "$tmp_mp" -a "$tmp_mp" = "$mp" ]; then
 			return $YES
 		fi
-	done < <(mount | awk '{print $1,$3}')
+	done < /proc/mounts
 
 	return $NO
 }
 
 
 #
-# trim_trailing_slash path
-#
-# Trim trailing slash from given path.
-#
-trim_trailing_slash() {
-	declare mpath=$1
-
-	echo $mpath | sed -e 's/\/*$//'
-}
-
-
-#
 # isMounted device mount_point
 #
 # Check to see if the device is mounted.  Print a warning if its not
@@ -497,26 +549,38 @@ isMounted () {
 
 	typeset mp tmp_mp
 	typeset dev tmp_dev
+	typeset ret=$FAIL
 
 	if [ $# -ne 2 ]; then
 		ocf_log err "Usage: isMounted device mount_point"
 		return $FAIL
 	fi
 
-	dev=$(real_device $1)
+	real_device $1
+	dev=$REAL_DEVICE
 	if [ -z "$dev" ]; then
 		ocf_log err \
 			"fs (isMounted): Could not match $1 with a real device"
 		return $OCF_ERR_ARGS
 	fi
-	mp=$(readlink -f $2)
-	
-	while read tmp_dev tmp_mp
+
+	if [ -h "$2" ]; then
+		mp=$(readlink -f $2)
+	else
+		mp=$2
+	fi
+
+	ret=$NO
+
+	while read tmp_dev tmp_mp junk_a junk_b junk_c junk_d
 	do
-		#echo "spec=$1 dev=$dev  tmp_dev=$tmp_dev"
-		tmp_dev=$(real_device $tmp_dev)
-		tmp_mp=${tmp_mp/%\//}  #$(trim_trailing_slash $tmp_mp)
-		mp=${mp/%\//}		       #$(trim_trailing_slash $mp)
+		real_device $tmp_dev
+		tmp_dev=$REAL_DEVICE
+
+		# This bash glyph simply removes a trailing slash
+		# if one exists.  /a/b/ -> /a/b; /a/b -> /a/b.
+		tmp_mp=${tmp_mp%/}
+		mp=${mp%/}
 
 		if [ -n "$tmp_dev" -a "$tmp_dev" = "$dev" ]; then
 			#
@@ -527,11 +591,11 @@ isMounted () {
 				ocf_log warn \
 "Device $dev is mounted on $tmp_mp instead of $mp"
 			fi
-			return $YES
+			ret=$YES
 		fi
-	done < <(mount | awk '{print $1,$3}')
+	done < /proc/mounts
 
-	return $NO
+	return $ret
 }
 
 
@@ -702,80 +766,6 @@ killMountProcesses()
 }
 
 
-activeMonitor() {
-	declare monpath=$OCF_RESKEY_mountpoint/.clumanager
-	declare p
-	declare pid
-
-	if [ -z "$OCF_RESKEY_mountpoint" ]; then
-		ocf_log err "activeMonitor: No mount point specified"
-		return $OCF_ERR_ARGS
-	fi
-
-	if [ "$OCF_RESKEY_active_monitor" != "1" ] &&
-	   [ "$OCF_RESKEY_active_monitor" != "yes" ]; then
-		# Nothing bad happened; but no active monitoring specified.
-		return $OCF_SUCCESS
-	fi
-
-	if [ "$OCF_RESKEY_self_fence" = "1" ] ||
-	   [ "$OCF_RESKEY_self_fence" = "yes" ]; then
-		args="-i 2 -a reboot"
-	else
-		args="-i 2"
-	fi
-
-	case $1 in
-	start)
-		ocf_log info "Starting active monitoring of $OCF_RESKEY_mountpoint"
-		mkdir -p $(dirname $monpath) || return $OCF_ERR_GENERIC
-		devmon $args -p $monpath/devmon.data -P $monpath/devmon.pid
-		;;
-	stop)
-		ocf_log info "Stopping active monitoring of $OCF_RESKEY_mountpoint"
-		if ! [ -f $monpath/devmon.pid ]; then
-			# Someone removed the file or it wasn't there for
-			# some reason... Force unmount will kill us
-			return 0
-		fi
-
-		pid=$(cat $monpath/devmon.pid)
-		if [ -z "$pid" ]; then
-			# Someone emptied the file?
-			return 0
-		fi
-
-		for p in $(pidof devmon); do
-			if [ "$pid" = "$p" ]; then
-				ocf_log debug "Killing devmon $p for $OCF_RESKEY_mountpoint"
-				kill -TERM $p
-				return 0
-			fi
-		done
-		# none matching
-
-		return 0
-		;;
-	status)
-		pid=$(cat $monpath/devmon.pid)
-		for p in $(pidof devmon); do
-			if [ "$pid" = "$p" ]; then
-				return 0
-			fi
-		done
-
-		# none matching
-		ocf_log err "Active Monitor for $OCF_RESKEY_mountpoint has exited"
-		return $OCF_ERR_GENERIC
-		;;
-	*)
-		ocf_log err "usage: activeMonitor <start|stop|status>"
-		return $OCF_ERR_ARGS
-		;;
-	esac
-}
-
-
 #
 # Decide which quota options are enabled and return a string 
 # which we can pass to quotaon
@@ -892,7 +882,8 @@ startFilesystem() {
 	#
 	# Get the device
 	#
-	dev=$(real_device $OCF_RESKEY_device)
+	real_device $OCF_RESKEY_device
+	dev=$REAL_DEVICE
 	if [ -z "$dev" ]; then
 			ocf_log err "\
 startFilesystem: Could not match $OCF_RESKEY_device with a real device"
@@ -1067,7 +1058,6 @@ Unknown file system type '$fstype' for device $dev.  Assuming fsck is required."
 	fi
 
 	enable_fs_quotas $opts $mp
-	activeMonitor start || return $OCF_ERR_GENERIC
 	
 	return $SUCCESS
 }
@@ -1113,7 +1103,8 @@ stopFilesystem() {
 	#
 	# Get the device
 	#
-	dev=$(real_device $OCF_RESKEY_device)
+	real_device $OCF_RESKEY_device
+	dev=$REAL_DEVICE
 	if [ -z "$dev" ]; then
 			ocf_log err "\
 stop: Could not match $OCF_RESKEY_device with a real device"
@@ -1161,7 +1152,6 @@ stop: Could not match $OCF_RESKEY_device with a real device"
 		       		quotaoff -$quotaopts $mp &> /dev/null
 			fi
 
-			activeMonitor stop || return $OCF_ERR_GENERIC
 
 			ocf_log info "unmounting $mp"
 			umount $mp
@@ -1204,7 +1194,7 @@ stop: Could not match $OCF_RESKEY_device with a real device"
 
 		if [ $try -ge $max_tries ]; then
 			done=$YES
-		else
+		elif [ "$done" -ne "$YES" ]; then
 			sleep $sleep_time
 			let try=try+1
 		fi
@@ -1241,33 +1231,16 @@ stop)
 	;;
 status|monitor)
   	isMounted ${OCF_RESKEY_device} ${OCF_RESKEY_mountpoint}
-	case $? in
-	$NO)
-		ocf_log info "fs:${OCF_RESKEY_name}: ${OCF_RESKEY_device} is not mounted on ${OCF_RESKEY_mountpoint}"
-		exit $OCF_NOT_RUNNING
-		;;
-	$YES)
-		;;
-	*)
-		ocf_log err "fs:${OCF_RESKEY_name}: ${OCF_RESKEY_device} is not mounted on ${OCF_RESKEY_mountpoint}"
-		exit $OCF_ERR_GENERIC
-		;;
-	esac
-		
+
  	if [ $? -ne $YES ]; then
 		ocf_log err "fs:${OCF_RESKEY_name}: ${OCF_RESKEY_device} is not mounted on ${OCF_RESKEY_mountpoint}"
 		exit $OCF_ERR_GENERIC
 	fi
 
-	if [ "$OCF_RESKEY_active_monitor" = "yes" ] ||
-	   [ "$OCF_RESKEY_active_monitor" = "1" ]; then
-
-	   	activeMonitor status
-		[ $? -eq 0 ] && exit 0
-		ocf_log err "fs:${OCF_RESKEY_name}: Active Monitoring reported a failure"
-		exit $OCF_ERR_GENERIC
+	if [ "$OCF_RESKEY_quick_status" = "1" ]; then
+		exit 0
 	fi
- 	
+
  	isAlive ${OCF_RESKEY_mountpoint}
  	[ $? -eq $YES ] && exit 0
 


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]