X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=6c309d127ba6b5e3491758238ff056aedeaa802a;hp=57d4f24dbf28b56ce702ca2b9415a23ac7d676ea;hb=7d6250a7ebb1d4c803c3f7b31994bca911b09a4c;hpb=072d6b3156bf3b5d1738b43aadcba5c378c59ee9

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 57d4f24..6c309d1 100755
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -817,7 +817,7 @@ facet_type() {
 facet_number() {
 	local facet=$1
 
-	if [ $facet == mgs ]; then
+	if [ $facet == mgs ] || [ $facet == client ]; then
 		return 1
 	fi
 
@@ -1116,7 +1116,7 @@ get_osd_param() {
 	local name=$3
 
 	do_nodes $nodes "$LCTL get_param -n obdfilter.$device.$name \
-		osd-*.$device.$name 2>&1" | grep -v 'Found no match'
+		osd-*.$device.$name 2>&1" | grep -v 'error:'
 }
 
 set_osd_param() {
@@ -1126,7 +1126,7 @@ set_osd_param() {
 	local value=$4
 
 	do_nodes $nodes "$LCTL set_param -n obdfilter.$device.$name=$value \
-		osd-*.$device.$name=$value 2>&1" | grep -v 'Found no match'
+		osd-*.$device.$name=$value 2>&1" | grep -v 'error:'
 }
 
 set_debug_size () {
@@ -1223,6 +1223,8 @@ mount_facet() {
 	local opt=${facet}_opt
 	local mntpt=$(facet_mntpt $facet)
 	local opts="${!opt} $@"
+	local fstype=$(facet_fstype $facet)
+	local devicelabel
 
 	module_loaded lustre || load_modules
 
@@ -1236,6 +1238,16 @@ mount_facet() {
 		import_zpool $facet || return ${PIPESTATUS[0]}
 	fi
 
+	case $fstype in
+	ldiskfs)
+		devicelabel=$(do_facet ${facet} "$E2LABEL ${!dev}");;
+	zfs)
+		devicelabel=$(do_facet ${facet} "$ZFS get -H -o value \
+						lustre:svname ${!dev}");;
+	*)
+		error "unknown fstype!";;
+	esac
+
 	echo "Starting ${facet}: $opts ${!dev} $mntpt"
 	# for testing LU-482 error handling in mount_facets() and test_0a()
 	if [ -f $TMP/test-lu482-trigger ]; then
@@ -1261,7 +1273,6 @@ mount_facet() {
 	if [[ $opts =~ .*nosvc.* ]]; then
 		echo "Start ${!dev} without service"
 	else
-		local fstype=$(facet_fstype $facet)
 
 		case $fstype in
 		ldiskfs)
@@ -1279,6 +1290,12 @@ mount_facet() {
 		esac
 	fi
 
+	# commit the device label change to disk
+	if [[ $devicelabel =~ (:[a-zA-Z]{3}[0-9]{4}) ]]; then
+		do_facet $facet "sync; sync; sync"
+	fi
+
+
 	label=$(devicelabel ${facet} ${!dev})
 	[ -z "$label" ] && echo no label for ${!dev} && exit 1
 	eval export ${facet}_svc=${label}
@@ -1482,25 +1499,37 @@ setup_quota(){
 }
 
 zconf_mount() {
-    local client=$1
-    local mnt=$2
-    local opts=${3:-$MOUNT_OPTS}
-    opts=${opts:+-o $opts}
-    local flags=${4:-$MOUNT_FLAGS}
-
-    local device=$MGSNID:/$FSNAME
-    if [ -z "$mnt" -o -z "$FSNAME" ]; then
-        echo Bad zconf mount command: opt=$flags $opts dev=$device mnt=$mnt
-        exit 1
-    fi
-
-    echo "Starting client: $client: $flags $opts $device $mnt"
-    do_node $client mkdir -p $mnt
-    do_node $client $MOUNT_CMD $flags $opts $device $mnt || return 1
+	local client=$1
+	local mnt=$2
+	local opts=${3:-$MOUNT_OPTS}
+	opts=${opts:+-o $opts}
+	local flags=${4:-$MOUNT_FLAGS}
+
+	local device=$MGSNID:/$FSNAME$FILESET
+	if [ -z "$mnt" -o -z "$FSNAME" ]; then
+		echo "Bad mount command: opt=$flags $opts dev=$device " \
+		     "mnt=$mnt"
+		exit 1
+	fi
+
+	echo "Starting client: $client: $flags $opts $device $mnt"
+	do_node $client mkdir -p $mnt
+	if [ -n "$FILESET" -a -z "$SKIP_FILESET" ];then
+		do_node $client $MOUNT_CMD $flags $opts $MGSNID:/$FSNAME \
+			$mnt || return 1
+		#disable FILESET if not supported
+		do_nodes $client lctl get_param -n \
+			mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
+				device=$MGSNID:/$FSNAME
+		do_node $client mkdir -p $mnt/$FILESET
+		do_node $client "! grep -q $mnt' ' /proc/mounts ||
+			umount $mnt"
+	fi
+	do_node $client $MOUNT_CMD $flags $opts $device $mnt || return 1
+
+	set_default_debug_nodes $client
 
-    set_default_debug_nodes $client
-
-    return 0
+	return 0
 }
 
 zconf_umount() {
@@ -1590,21 +1619,35 @@ sanity_mount_check () {
 
 # mount clients if not mouted
 zconf_mount_clients() {
-    local clients=$1
-    local mnt=$2
-    local opts=${3:-$MOUNT_OPTS}
-    opts=${opts:+-o $opts}
-    local flags=${4:-$MOUNT_FLAGS}
-
-    local device=$MGSNID:/$FSNAME
-    if [ -z "$mnt" -o -z "$FSNAME" ]; then
-        echo Bad zconf mount command: opt=$flags $opts dev=$device mnt=$mnt
-        exit 1
-    fi
-
-    echo "Starting client $clients: $flags $opts $device $mnt"
-
-    do_nodes $clients "
+	local clients=$1
+	local mnt=$2
+	local opts=${3:-$MOUNT_OPTS}
+	opts=${opts:+-o $opts}
+	local flags=${4:-$MOUNT_FLAGS}
+
+	local device=$MGSNID:/$FSNAME$FILESET
+	if [ -z "$mnt" -o -z "$FSNAME" ]; then
+		echo "Bad conf mount command: opt=$flags $opts dev=$device " \
+		     "mnt=$mnt"
+		exit 1
+	fi
+
+	echo "Starting client $clients: $flags $opts $device $mnt"
+	if [ -n "$FILESET" -a ! -n "$SKIP_FILESET" ]; then
+		do_nodes $clients "! grep -q $mnt' ' /proc/mounts ||
+			umount $mnt"
+		do_nodes $clients $MOUNT_CMD $flags $opts $MGSNID:/$FSNAME \
+			$mnt || return 1
+		#disable FILESET if not supported
+		do_nodes $clients lctl get_param -n \
+			mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
+				device=$MGSNID:/$FSNAME
+		do_nodes $clients mkdir -p $mnt/$FILESET
+		do_nodes $clients "! grep -q $mnt' ' /proc/mounts ||
+			umount $mnt"
+	fi
+
+	do_nodes $clients "
 running=\\\$(mount | grep -c $mnt' ');
 rc=0;
 if [ \\\$running -eq 0 ] ; then
@@ -1614,12 +1657,12 @@ if [ \\\$running -eq 0 ] ; then
 fi;
 exit \\\$rc" || return ${PIPESTATUS[0]}
 
-    echo "Started clients $clients: "
-    do_nodes $clients "mount | grep $mnt' '"
+	echo "Started clients $clients: "
+	do_nodes $clients "mount | grep $mnt' '"
 
-    set_default_debug_nodes $clients
+	set_default_debug_nodes $clients
 
-    return 0
+	return 0
 }
 
 zconf_umount_clients() {
@@ -4642,20 +4685,17 @@ stop_full_debug_logging() {
 
 # prints bash call stack
 print_stack_trace() {
+	local skip=${1:-1}
 	echo "  Trace dump:"
-	for (( i=1; i < ${#BASH_LINENO[*]} ; i++ )) ; do
-		local s=${BASH_SOURCE[$i]}
-		local l=${BASH_LINENO[$i-1]}
-		local f=${FUNCNAME[$i]}
-		echo "  = $s:$l:$f()"
+	for (( i=$skip; i < ${#BASH_LINENO[*]} ; i++ )) ; do
+		local src=${BASH_SOURCE[$i]}
+		local lineno=${BASH_LINENO[$i-1]}
+		local funcname=${FUNCNAME[$i]}
+		echo "  = $src:$lineno:$funcname()"
 	done
 }
 
-##################################
-# Test interface
-##################################
-
-error_noexit() {
+report_error() {
 	local TYPE=${TYPE:-"FAIL"}
 
 	local dump=true
@@ -4665,10 +4705,8 @@ error_noexit() {
 		dump=false
 	fi
 
-
 	log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
-	print_stack_trace >&2
-
+	(print_stack_trace 2) >&2
 	mkdir -p $LOGDIR
 	# We need to dump the logs on all nodes
 	if $dump; then
@@ -4687,6 +4725,17 @@ error_noexit() {
 			echo "$@" > $LOGDIR/err
 		fi
 	fi
+
+	# cleanup the env for failed tests
+	reset_fail_loc
+}
+
+##################################
+# Test interface
+##################################
+
+error_noexit() {
+	report_error "$@"
 }
 
 exit_status () {
@@ -4698,12 +4747,13 @@ exit_status () {
 }
 
 error() {
-	error_noexit "$@"
+	report_error "$@"
 	exit 1
 }
 
 error_exit() {
-	error "$@"
+	report_error "$@"
+	exit 1
 }
 
 # use only if we are ignoring failures for this test, bugno required.
@@ -4713,11 +4763,11 @@ error_exit() {
 error_ignore() {
 	local TYPE="IGNORE ($1)"
 	shift
-	error_noexit "$@"
+	report_error "$@"
 }
 
 error_and_remount() {
-	error_noexit "$@"
+	report_error "$@"
 	remount_client $MOUNT
 	exit 1
 }
@@ -4929,6 +4979,18 @@ banner() {
     log "$msg== $(date +"%H:%M:%S (%s)")"
 }
 
+check_dmesg_for_errors() {
+	local res
+	local errors="VFS: Busy inodes after unmount of\|\
+ldiskfs_check_descriptors: Checksum for group 0 failed\|\
+group descriptors corrupted"
+
+	res=$(do_nodes $(comma_list $(nodes_list)) "dmesg" | grep "$errors")
+	[ -z "$res" ] && return 0
+	echo "Kernel error detected: $res"
+	return 1
+}
+
 #
 # Run a single test function and cleanup after it.
 #
@@ -4944,12 +5006,17 @@ run_one() {
 	local SAVE_UMASK=`umask`
 	umask 0022
 
+	if ! grep -q $DIR /proc/mounts; then
+		$SETUP
+	fi
+
 	banner "test $testnum: $message"
 	test_${testnum} || error "test_$testnum failed with $?"
 	cd $SAVE_PWD
 	reset_fail_loc
 	check_grant ${testnum} || error "check_grant $testnum failed with $?"
 	check_catastrophe || error "LBUG/LASSERT detected"
+	check_dmesg_for_errors || error "Error in dmesg detected"
 	if [ "$PARALLEL" != "yes" ]; then
 		ps auxww | grep -v grep | grep -q multiop &&
 					error "multiop still running"
@@ -4958,6 +5025,7 @@ run_one() {
 	unset tdir
 	unset tfile
 	umask $SAVE_UMASK
+	$CLEANUP
 	return 0
 }
 
@@ -6070,19 +6138,21 @@ create_pool() {
 }
 
 add_pool_to_list () {
-    local fsname=${1%%.*}
-    local poolname=${1##$fsname.}
+	local fsname=${1%%.*}
+	local poolname=${1##$fsname.}
 
-    local listvar=${fsname}_CREATED_POOLS
-    eval export ${listvar}=$(expand_list ${!listvar} $poolname)
+	local listvar=${fsname}_CREATED_POOLS
+	local temp=${listvar}=$(expand_list ${!listvar} $poolname)
+	eval export $temp
 }
 
 remove_pool_from_list () {
-    local fsname=${1%%.*}
-    local poolname=${1##$fsname.}
+	local fsname=${1%%.*}
+	local poolname=${1##$fsname.}
 
-    local listvar=${fsname}_CREATED_POOLS
-    eval export ${listvar}=$(exclude_items_from_list ${!listvar} $poolname)
+	local listvar=${fsname}_CREATED_POOLS
+	local temp=${listvar}=$(exclude_items_from_list ${!listvar} $poolname)
+	eval export $temp
 }
 
 destroy_pool_int() {
@@ -6219,6 +6289,31 @@ max_recovery_time() {
 	echo -n $service_time
 }
 
+recovery_time_min() {
+	local connection_switch_min=5
+	local connection_switch_inc=5
+	local connection_switch_max
+	local reconnect_delay_max
+	local initial_connect_timeout
+	local max
+	local timout_20
+
+	#connection_switch_max=min(50, max($connection_switch_min,$TIMEOUT)
+	(($connection_switch_min > $TIMEOUT)) &&
+		max=$connection_switch_min || max=$TIMEOUT
+	(($max < 50)) && connection_switch_max=$max || connection_switch_max=50
+
+	#initial_connect_timeout = max(connection_switch_min, obd_timeout/20)
+	timeout_20=$((TIMEOUT/20))
+	(($connection_switch_min > $timeout_20)) &&
+		initial_connect_timeout=$connection_switch_min ||
+		initial_connect_timeout=$timeout_20
+
+	reconnect_delay_max=$((connection_switch_max + connection_switch_inc + \
+			       initial_connect_timeout))
+	echo $((2 * reconnect_delay_max))
+}
+
 get_clients_mount_count () {
     local clients=${CLIENTS:-`hostname`}