X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-scrub.sh;h=5ca94e26b6a5da39669431236ba9ed0138efe761;hb=25374e3b35f5776a9fd6e5b4cac273f7000511e9;hp=4db0e258b7736478a392aa1b94554d672eedb940;hpb=dc2e5bc3d9277ecf10e645350797a742c4342a44;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-scrub.sh b/lustre/tests/sanity-scrub.sh
index 4db0e25..5ca94e2 100644
--- a/lustre/tests/sanity-scrub.sh
+++ b/lustre/tests/sanity-scrub.sh
@@ -21,13 +21,19 @@ require_dsh_mds || exit 0
 
 SAVED_MDSSIZE=${MDSSIZE}
 SAVED_OSTSIZE=${OSTSIZE}
+SAVED_OSTCOUNT=${OSTCOUNT}
 # use small MDS + OST size to speed formatting time
 # do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
 MDSSIZE=100000
 OSTSIZE=100000
+# no need too much OSTs, to reduce the format/start/stop overhead
+[ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
 
 MOUNT_2=""
-check_and_setup_lustre
+
+# build up a clean test environment.
+formatall
+setupall
 
 [ $(facet_fstype $SINGLEMDS) != "ldiskfs" ] &&
 	skip "test OI scrub only for ldiskfs" && check_and_cleanup_lustre &&
@@ -42,6 +48,13 @@ check_and_setup_lustre
 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.90) ]] &&
 	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 1a"
 
+[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.1) ]] &&
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 15"
+
+[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.90) ]] &&
+[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.4.50) ]] &&
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 15"
+
 [[ $(lustre_version_code ost1) -lt $(version_code 2.4.50) ]] &&
 	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14"
 
@@ -92,16 +105,13 @@ scrub_prep() {
 	local nfiles=$1
 	local n
 
-	echo "formatall"
-	formatall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
+	check_mount_and_prep
 
-	echo "preparing..."
+	echo "preparing... $(date)"
 	for n in $(seq $MDSCOUNT); do
 		echo "creating $nfiles files on mds$n"
 		if [ $n -eq 1 ]; then
-			mkdir -p $DIR/$tdir/mds$n ||
+			mkdir $DIR/$tdir/mds$n ||
 				error "Failed to create directory mds$n"
 		else
 			$LFS mkdir -i $((n - 1)) $DIR/$tdir/mds$n ||
@@ -110,11 +120,11 @@ scrub_prep() {
 		cp $LUSTRE/tests/*.sh $DIR/$tdir/mds$n ||
 			error "Failed to copy files to mds$n"
 		if [[ $nfiles -gt 0 ]]; then
-			createmany -o $DIR/$tdir/mds$n/$tfile $nfiles ||
-				error "createmany failed on mds$n"
+			createmany -m $DIR/$tdir/mds$n/$tfile $nfiles > \
+				/dev/null || error "createmany failed on mds$n"
 		fi
 	done
-	echo "prepared."
+	echo "prepared $(date)."
 	cleanup_mount $MOUNT > /dev/null || error "Fail to stop client!"
 	for n in $(seq $MDSCOUNT); do
 		echo "stop mds$n"
@@ -147,13 +157,26 @@ scrub_stop_mds() {
 scrub_check_status() {
 	local error_id=$1
 	local expected=$2
+	local n
+
+	for n in $(seq $MDSCOUNT); do
+		wait_update_facet mds$n "$LCTL get_param -n \
+			osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
+			awk '/^status/ { print \\\$2 }'" "$expected" 6 ||
+			error "($error_id) Expected '$expected' on mds$n"
+	done
+}
+
+scrub_check_flags() {
+	local error_id=$1
+	local expected=$2
 	local actual
 	local n
 
 	for n in $(seq $MDSCOUNT); do
 		actual=$(do_facet mds$n $LCTL get_param -n \
 			osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
-			awk '/^status/ { print $2 }')
+			awk '/^flags/ { print $2 }')
 		if [ "$actual" != "$expected" ]; then
 			error "($error_id) Expected '$expected' on mds$n, but" \
 			       "got '$actual'"
@@ -161,7 +184,7 @@ scrub_check_status() {
 	done
 }
 
-scrub_check_flags() {
+scrub_check_params() {
 	local error_id=$1
 	local expected=$2
 	local actual
@@ -170,7 +193,7 @@ scrub_check_flags() {
 	for n in $(seq $MDSCOUNT); do
 		actual=$(do_facet mds$n $LCTL get_param -n \
 			osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
-			awk '/^flags/ { print $2 }')
+			awk '/^param/ { print $2 }')
 		if [ "$actual" != "$expected" ]; then
 			error "($error_id) Expected '$expected' on mds$n, but" \
 			       "got '$actual'"
@@ -178,6 +201,29 @@ scrub_check_flags() {
 	done
 }
 
+scrub_check_repaired() {
+	local error_id=$1
+	local expected=$2
+	local actual
+	local n
+
+	for n in $(seq $MDSCOUNT); do
+		actual=$(do_facet mds$n $LCTL get_param -n \
+			osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
+			awk '/^updated/ { print $2 }')
+
+		if [ $expected -eq 0 -a $actual -ne 0 ]; then
+			error "($error_id) Expected no repaired on mds$n, but" \
+			       "got '$actual'"
+		fi
+
+		if [ $expected -ne 0 -a $actual -lt $expected ]; then
+			error "($error_id) Expected '$expected' on mds$n, but" \
+			       "got '$actual'"
+		fi
+	done
+}
+
 scrub_check_data() {
 	local error_id=$1
 	local n
@@ -237,21 +283,15 @@ test_1a() {
 	start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
 		error "(1) Fail to start MDS!"
 
-	local STATUS=$($SHOW_SCRUB | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "init" ] ||
-		error "(2) Expect 'init', but got '$STATUS'"
-
 	local FLAGS=$($SHOW_SCRUB | awk '/^flags/ { print $2 }')
 	[ -z "$FLAGS" ] || error "(3) Expect empty flags, but got '$FLAGS'"
 
 	mount_client $MOUNT || error "(4) Fail to start client!"
-
 	#define OBD_FAIL_OSD_FID_MAPPING			0x193
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x193
 	# update .lustre OI mapping
 	touch $MOUNT/.lustre
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
-
 	umount_client $MOUNT || error "(5) Fail to stop client!"
 
 	echo "stop $SINGLEMDS"
@@ -261,10 +301,6 @@ test_1a() {
 	start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
 		error "(7) Fail to start MDS!"
 
-	local STATUS=$($SHOW_SCRUB | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "init" ] ||
-		error "(8) Expect 'init', but got '$STATUS'"
-
 	local FLAGS=$($SHOW_SCRUB | awk '/^flags/ { print $2 }')
 	[ "$FLAGS" == "inconsistent" ] ||
 		error "(9) Expect 'inconsistent', but got '$FLAGS'"
@@ -276,7 +312,6 @@ test_1b() {
 	scrub_remove_ois 1
 	echo "start MDTs without disabling OI scrub"
 	scrub_start_mds 2 "$MOUNT_OPTS_SCRUB"
-	sleep 3
 	scrub_check_status 3 completed
 	mount_client $MOUNT || error "(4) Fail to start client!"
 	scrub_check_data 5
@@ -288,20 +323,15 @@ test_1c() {
 
 	# OI files to be removed:
 	# idx 0: oi.16.0
-	# idx 1: oi.16.1
 	# idx 2: oi.16.{2,4,8,16,32}
 	# idx 3: oi.16.{3,9,27}
-	# idx 5: oi.16.{5,25}
-	# idx 7: oi.16.{7,49}
-	for index in 0 1 2 3 5 7; do
+	for index in 0 2 3; do
 		scrub_prep 0
 		scrub_remove_ois 1 $index
-
 		echo "start MDTs with OI scrub disabled"
 		scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
 		scrub_check_flags 3 recreated
 		scrub_start 4
-		sleep 3
 		scrub_check_status 5 completed
 		scrub_check_flags 6 ""
 	done
@@ -313,43 +343,45 @@ test_2() {
 	scrub_backup_restore 1
 	echo "starting MDTs without disabling OI scrub"
 	scrub_start_mds 2 "$MOUNT_OPTS_SCRUB"
-	sleep 3
 	scrub_check_status 3 completed
 	mount_client $MOUNT || error "(4) Fail to start client!"
 	scrub_check_data 5
 }
 run_test 2 "Trigger OI scrub when MDT mounts for backup/restore case"
 
+# test_3 is obsolete, it will be covered by test_5.
 test_3() {
+	formatall > /dev/null
+	setupall > /dev/null
+
 	scrub_prep 0
 	scrub_backup_restore 1
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-	sleep 3
 	scrub_check_status 3 init
 	scrub_check_flags 4 inconsistent
-	echo "stopall"
-	stopall > /dev/null
 }
-run_test 3 "Do not trigger OI scrub when MDT mounts if 'noscrub' specified"
+#run_test 3 "Do not trigger OI scrub when MDT mounts if 'noscrub' specified"
 
 test_4() {
 	scrub_prep 0
 	scrub_backup_restore 1
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-	scrub_check_status 3 init
 	scrub_check_flags 4 inconsistent
 	mount_client $MOUNT || error "(5) Fail to start client!"
 	scrub_enable_auto
 	scrub_check_data 6
-	sleep 3
 	scrub_check_status 7 completed
+	scrub_check_flags 8 ""
 }
 run_test 4 "Trigger OI scrub automatically if inconsistent OI mapping was found"
 
 test_5() {
-	scrub_prep 1500
+	formatall > /dev/null
+	setupall > /dev/null
+
+	scrub_prep 1000
 	scrub_backup_restore 1
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
@@ -358,73 +390,58 @@ test_5() {
 	mount_client $MOUNT || error "(5) Fail to start client!"
 	scrub_enable_auto
 
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
-	scrub_check_data 6
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
 
+	scrub_check_data 6
 	umount_client $MOUNT || error "(7) Fail to stop client!"
-
 	scrub_check_status 8 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_CRASH	 0x191
-		do_facet mds$n $LCTL set_param fail_loc=0x191
-	done
+	#define OBD_FAIL_OSD_SCRUB_CRASH	 0x191
+	do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x191
+
 	sleep 4
 	scrub_stop_mds 9
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 10 "$MOUNT_OPTS_NOSCRUB"
-
 	scrub_check_status 11 crashed
-
 	scrub_stop_mds 12
 
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
 	echo "starting MDTs without disabling OI scrub"
 	scrub_start_mds 13 "$MOUNT_OPTS_SCRUB"
-
 	scrub_check_status 14 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_FATAL	 0x192
-		do_facet mds$n $LCTL set_param fail_loc=0x192
-	done
-	sleep 4
-	scrub_check_status 15 failed
+	#define OBD_FAIL_OSD_SCRUB_FATAL	 0x192
+	do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x192
 
+	scrub_check_status 15 failed
 	mount_client $MOUNT || error "(16) Fail to start client!"
 
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
+	local n
 	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-		stat $DIR/$tdir/mds$n/${tfile}1000 ||
-			error "(17) Failed to stat mds$n/${tfile}1000"
+		stat $DIR/$tdir/mds$n/${tfile}800 ||
+			error "(17) Failed to stat mds$n/${tfile}800"
 	done
 
 	scrub_check_status 18 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 19 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 19 completed
 	scrub_check_flags 20 ""
 }
 run_test 5 "OI scrub state machine"
@@ -434,45 +451,41 @@ test_6() {
 	scrub_backup_restore 1
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-	scrub_check_status 3 init
 	scrub_check_flags 4 inconsistent
 	mount_client $MOUNT || error "(5) Fail to start client!"
 	scrub_enable_auto
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
+
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
 	scrub_check_data 6
 
 	# Sleep 5 sec to guarantee at least one object processed by OI scrub
 	sleep 5
 	# Fail the OI scrub to guarantee there is at least one checkpoint
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_FATAL	 0x192
-		do_facet mds$n $LCTL set_param fail_loc=0x192
-	done
-	sleep 4
+	#define OBD_FAIL_OSD_SCRUB_FATAL	 0x192
+	do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x192
+
 	scrub_check_status 7 failed
 
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
+	local n
 	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
 		# stat will re-trigger OI scrub
 		stat $DIR/$tdir/mds$n/${tfile}800 ||
 			error "(8) Failed to stat mds$n/${tfile}800"
 	done
 
 	umount_client $MOUNT || error "(9) Fail to stop client!"
-
 	scrub_check_status 10 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_CRASH	 0x191
-		do_facet mds$n $LCTL set_param fail_loc=0x191
-	done
+	#define OBD_FAIL_OSD_SCRUB_CRASH	 0x191
+	do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x191
+
 	sleep 4
 	local -a position0
 	for n in $(seq $MDSCOUNT); do
@@ -483,11 +496,10 @@ test_6() {
 
 	scrub_stop_mds 11
 
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
 	echo "starting MDTs without disabling OI scrub"
 	scrub_start_mds 12 "$MOUNT_OPTS_SCRUB"
 
@@ -495,7 +507,7 @@ test_6() {
 
 	local -a position1
 	for n in $(seq $MDSCOUNT); do
-		positions1[$n]=$(scrub_status $n |
+		position1[$n]=$(scrub_status $n |
 			awk '/^latest_start_position/ {print $2}')
 		if [ ${position0[$n]} -ne ${position1[$n]} ]; then
 			error "(14) Expected position ${position0[$n]}, but" \
@@ -503,13 +515,10 @@ test_6() {
 		fi
 	done
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 15 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 15 completed
 	scrub_check_flags 16 ""
 }
 run_test 6 "OI scrub resumes from last checkpoint"
@@ -517,39 +526,31 @@ run_test 6 "OI scrub resumes from last checkpoint"
 test_7() {
 	scrub_prep 500
 	scrub_backup_restore 1
-
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-	scrub_check_status 3 init
 	scrub_check_flags 4 inconsistent
-
 	mount_client $MOUNT || error "(5) Fail to start client!"
-
 	scrub_enable_auto
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
+
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
+
 	scrub_check_data 6
 
+	local n
 	for n in $(seq $MDSCOUNT); do
 		stat $DIR/$tdir/mds$n/${tfile}300 ||
 			error "(7) Failed to stat mds$n/${tfile}300!"
 	done
 
 	scrub_check_status 8 scanning
-
 	scrub_check_flags 9 inconsistent,auto
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 10 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 10 completed
 	scrub_check_flags ""
 }
 run_test 7 "System is available during OI scrub scanning"
@@ -557,39 +558,25 @@ run_test 7 "System is available during OI scrub scanning"
 test_8() {
 	scrub_prep 128
 	scrub_backup_restore 1
-
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-
-	scrub_check_status 3 init
-
 	scrub_check_flags 4 inconsistent
 
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=1
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
-	scrub_start 5
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=1 fail_loc=0x190
 
+	scrub_start 5
 	scrub_check_status 6 scanning
-
 	scrub_stop 7
-
 	scrub_check_status 8 stopped
-
 	scrub_start 9
-
 	scrub_check_status 10 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 11 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 11 completed
 	scrub_check_flags 12 ""
 }
 run_test 8 "Control OI scrub manually"
@@ -600,14 +587,11 @@ test_9() {
 		return 0
 	fi
 
-	scrub_prep 8000
+	scrub_prep 6000
 	scrub_backup_restore 1
 
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-
-	scrub_check_status 3 init
-
 	scrub_check_flags 4 inconsistent
 
 	local BASE_SPEED1=100
@@ -617,9 +601,7 @@ test_9() {
 
 	sleep $RUN_TIME1
 	scrub_check_status 6 completed
-
 	scrub_check_flags 7 ""
-
 	# OI scrub should run with limited speed under non-inconsistent case
 	scrub_start 8 -s $BASE_SPEED1 -r
 
@@ -674,7 +656,7 @@ test_9() {
 		do_facet mds$n $LCTL set_param -n \
 				mdd.$(facet_svc mds$n).lfsck_speed_limit 0
 	done
-	sleep 6
+
 	scrub_check_status 13 completed
 }
 run_test 9 "OI scrub speed control"
@@ -682,50 +664,32 @@ run_test 9 "OI scrub speed control"
 test_10a() {
 	scrub_prep 0
 	scrub_backup_restore 1
-
 	echo "starting mds$n with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-
-	scrub_check_status 3 init
-
 	scrub_check_flags 4 inconsistent
-
 	mount_client $MOUNT || error "(5) Fail to start client!"
-
 	scrub_enable_auto
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=1
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
-	scrub_check_data 6
 
-	scrub_check_status 7 scanning
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=1 fail_loc=0x190
 
+	scrub_check_data 6
+	scrub_check_status 7 scanning
 	umount_client $MOUNT || error "(8) Fail to stop client!"
-
 	scrub_stop_mds 9
-
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 10 "$MOUNT_OPTS_NOSCRUB"
-
 	scrub_check_status 11 paused
-
 	scrub_stop_mds 12
-
 	echo "starting MDTs without disabling OI scrub"
 	scrub_start_mds 13 "$MOUNT_OPTS_SCRUB"
-
 	scrub_check_status 14 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 15 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 15 completed
 	scrub_check_flags 16 ""
 }
 run_test 10a "non-stopped OI scrub should auto restarts after MDS remount (1)"
@@ -734,79 +698,54 @@ run_test 10a "non-stopped OI scrub should auto restarts after MDS remount (1)"
 test_10b() {
 	scrub_prep 0
 	scrub_backup_restore 1
-
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
-
-	scrub_check_status 3 init
-
 	scrub_check_flags 4 inconsistent
 
-	local n
-	for n in $(seq $MDSCOUNT); do
-		#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
-		do_facet mds$n $LCTL set_param fail_val=3
-		do_facet mds$n $LCTL set_param fail_loc=0x190
-	done
+	#define OBD_FAIL_OSD_SCRUB_DELAY	 0x190
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_val=3 fail_loc=0x190
 
 	scrub_start 5
-
 	scrub_check_status 6 scanning
-
 	scrub_stop_mds 7
-
 	echo "starting MDTs with OI scrub disabled"
 	scrub_start_mds 8 "$MOUNT_OPTS_NOSCRUB"
-
 	scrub_check_status 9 paused
-
 	scrub_stop_mds 10
-
 	echo "starting MDTs without disabling OI scrub"
 	scrub_start_mds 11 "$MOUNT_OPTS_SCRUB"
-
 	scrub_check_status 12 scanning
 
-	for n in $(seq $MDSCOUNT); do
-		do_facet mds$n $LCTL set_param fail_loc=0
-		do_facet mds$n $LCTL set_param fail_val=0
-	done
-	sleep 5
-	scrub_check_status 13 completed
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param fail_loc=0 fail_val=0
 
+	scrub_check_status 13 completed
 	scrub_check_flags 14 ""
 }
 #run_test 10b "non-stopped OI scrub should auto restarts after MDS remount (2)"
 
 test_11() {
-	echo "stopall"
-	stopall > /dev/null
-	echo "formatall"
-	formatall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
-
 	local CREATED=100
-	local tname=`date +%s`
-	rm -rf $MOUNT/$tname > /dev/null
-	mkdir -p $MOUNT/$tname || error "(0) Failed to create $MOUNT/$tname"
 	local n
+
+	check_mount_and_prep
+
 	for n in $(seq $MDSCOUNT); do
-		$LFS mkdir -i $((n - 1)) $MOUNT/$tname/mds$n ||
-			error "(1) Fail to mkdir $MOUNT/$tname/mds$n"
+		$LFS mkdir -i $((n - 1)) $DIR/$tdir/mds$n ||
+			error "(1) Fail to mkdir $DIR/$tdir/mds$n"
 
-		createmany -o $MOUNT/$tname/mds$n/f $CREATED ||
-			error "(2) Fail to create in $tname/mds$n"
+		createmany -o $DIR/$tdir/mds$n/f $CREATED ||
+			error "(2) Fail to create under $tdir/mds$n"
 	done
 
-	cleanup_mount $MOUNT
-	do_facet $SINGLEMDS $LCTL clear
-	start_full_debug_logging
 	# reset OI scrub start point by force
 	scrub_start 3 -r
-	sleep 3
 	scrub_check_status 4 completed
 
+	declare -a checked0
+	declare -a checked1
+
 	# OI scrub should skip the new created objects for the first accessing
 	# notice we're creating a new llog for every OST on every startup
 	# new features can make this even less stable, so we only check
@@ -818,149 +757,196 @@ test_11() {
 		[ $SKIPPED -ge $MAXIMUM -o $SKIPPED -lt $MINIMUM ] &&
 			error "(5) Expect [ $MINIMUM , $MAXIMUM ) objects" \
 				"skipped on mds$n, but got $SKIPPED"
+
+		checked0[$n]=$(scrub_status $n | awk '/^checked/ { print $2 }')
 	done
 
 	# reset OI scrub start point by force
-	scrub_start -r
-	sleep 3
+	scrub_start 6 -r
 	scrub_check_status 7 completed
 
 	# OI scrub should skip the new created object only once
 	for n in $(seq $MDSCOUNT); do
 		SKIPPED=$(scrub_status $n | awk '/^noscrub/ { print $2 }')
-		[ $SKIPPED -eq 0 ] ||
+		checked1[$n]=$(scrub_status $n | awk '/^checked/ { print $2 }')
+
+		[ ${checked0[$n]} -ne ${checked1[$n]} -o $SKIPPED -eq 0 ] ||
 			error "(8) Expect 0 objects skipped on mds$n, but" \
 				"got $SKIPPED"
 	done
-
-	stop_full_debug_logging
-	restore_mount $MOUNT || error "(9) Fail to start client!"
-	rm -rf $MOUNT/$tname > /dev/null
 }
 run_test 11 "OI scrub skips the new created objects only once"
 
 test_12() {
-	echo "stopall"
-	stopall > /dev/null
-	echo "formatall"
-	formatall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
-
-	mkdir -p $DIR/$tdir
+	check_mount_and_prep
 	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
 
+	local count=$(precreated_ost_obj_count 0 0)
+
 	#define OBD_FAIL_OSD_COMPAT_INVALID_ENTRY		0x195
 	do_facet ost1 $LCTL set_param fail_loc=0x195
-	createmany -o $DIR/$tdir/f 1000
+	createmany -o $DIR/$tdir/f $((count + 32))
 
-	echo "stopall"
-	stopall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
+	umount_client $MOUNT || error "(1) Fail to stop client!"
 
-	do_facet ost1 $LCTL set_param fail_loc=0
-	local STATUS=$($SHOW_SCRUB_ON_OST | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "init" ] ||
-		error "(1) Expect 'init', but got '$STATUS'"
+	stop ost1 || error "(2) Fail to stop ost1"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(2) ls should fail"
+	#define OBD_FAIL_OST_NODESTROY		 0x233
+	do_facet ost1 $LCTL set_param fail_loc=0x233
 
-	sleep 3
-	local STATUS=$($SHOW_SCRUB_ON_OST | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "completed" ] ||
-		error "(3) Expect 'completed', but got '$STATUS'"
+	start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB ||
+		error "(3) Fail to start ost1"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 || error "(4) ls should succeed"
+	mount_client $MOUNT || error "(4) Fail to start client!"
+
+	ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(5) ls should fail"
+
+	$START_SCRUB_ON_OST -r || error "(6) Fail to start OI scrub on OST!"
+
+	do_facet ost1 $LCTL set_param fail_loc=0
+	wait_update_facet ost1 "$LCTL get_param -n \
+		osd-ldiskfs.$(facet_svc ost1).oi_scrub |
+		awk '/^status/ { print \\\$2 }'" "completed" 6 ||
+		error "(7) Expected '$expected' on ost1"
+
+	ls -ail $DIR/$tdir > /dev/null || {
+		$SHOW_SCRUB_ON_OST
+		error "(8) ls should succeed"
+	}
 }
 run_test 12 "OI scrub can rebuild invalid /O entries"
 
 test_13() {
-	echo "stopall"
-	stopall > /dev/null
-	echo "formatall"
-	formatall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
-
-	mkdir -p $DIR/$tdir
+	check_mount_and_prep
 	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
 
+	local count=$(precreated_ost_obj_count 0 0)
+
 	#define OBD_FAIL_OSD_COMPAT_NO_ENTRY		0x196
 	do_facet ost1 $LCTL set_param fail_loc=0x196
-	createmany -o $DIR/$tdir/f 1000
+	createmany -o $DIR/$tdir/f $((count + 32))
 	do_facet ost1 $LCTL set_param fail_loc=0
 
-	echo "stopall"
-	stopall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
+	umount_client $MOUNT || error "(1) Fail to stop client!"
+
+	stop ost1 || error "(2) Fail to stop ost1"
+
+	start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB ||
+		error "(3) Fail to start ost1"
 
-	local STATUS=$($SHOW_SCRUB_ON_OST | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "init" ] ||
-		error "(1) Expect 'init', but got '$STATUS'"
+	mount_client $MOUNT || error "(4) Fail to start client!"
+
+	ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(5) ls should fail"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(2) ls should fail"
+	$START_SCRUB_ON_OST -r || error "(6) Fail to start OI scrub on OST!"
 
-	$START_SCRUB_ON_OST || error "(3) Fail to start OI scrub on OST!"
-	sleep 3
-	local STATUS=$($SHOW_SCRUB_ON_OST | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "completed" ] ||
-		error "(4) Expect 'completed', but got '$STATUS'"
+	wait_update_facet ost1 "$LCTL get_param -n \
+		osd-ldiskfs.$(facet_svc ost1).oi_scrub |
+		awk '/^status/ { print \\\$2 }'" "completed" 6 ||
+		error "(7) Expected '$expected' on ost1"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 || error "(5) ls should succeed"
+	ls -ail $DIR/$tdir > /dev/null || error "(8) ls should succeed"
 }
 run_test 13 "OI scrub can rebuild missed /O entries"
 
 test_14() {
-	echo "stopall"
-	stopall > /dev/null
-	echo "formatall"
-	formatall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
-
-	mkdir -p $DIR/$tdir
+	check_mount_and_prep
 	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
 
+	local count=$(precreated_ost_obj_count 0 0)
+
 	#define OBD_FAIL_OSD_COMPAT_NO_ENTRY		0x196
 	do_facet ost1 $LCTL set_param fail_loc=0x196
-	createmany -o $DIR/$tdir/f 64
+	createmany -o $DIR/$tdir/f $((count + 32))
 	do_facet ost1 $LCTL set_param fail_loc=0
 
-	echo "stopall"
-	stopall > /dev/null
-	echo "setupall"
-	setupall > /dev/null
-
-	local STATUS=$($SHOW_SCRUB_ON_OST | awk '/^status/ { print $2 }')
-	[ "$STATUS" == "init" ] ||
-		error "(1) Expect 'init', but got '$STATUS'"
+	umount_client $MOUNT || error "(1) Fail to stop client!"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(2) ls should fail"
-
-	echo "stopall"
-	stopall > /dev/null
+	stop ost1 || error "(2) Fail to stop ost1"
 
 	echo "run e2fsck"
 	run_e2fsck $(facet_host ost1) $(ostdevname 1) "-y" ||
 		error "(3) Fail to run e2fsck error"
 
-	echo "setupall"
-	setupall > /dev/null
+	start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
+		error "(4) Fail to start ost1"
+
+	mount_client $MOUNT || error "(5) Fail to start client!"
 
 	local LF_REPAIRED=$($SHOW_SCRUB_ON_OST |
 			    awk '/^lf_reparied/ { print $2 }')
 	[ $LF_REPAIRED -gt 0 ] ||
-		error "(4) Some entry under /lost+found should be repaired"
+		error "(6) Some entry under /lost+found should be repaired"
 
-	ls -ail $DIR/$tdir > /dev/null 2>&1 || error "(5) ls should succeed"
+	ls -ail $DIR/$tdir > /dev/null || error "(7) ls should succeed"
 }
 run_test 14 "OI scrub can repair objects under lost+found"
 
+test_15() {
+	# skip test_15 for LU-4182
+	[ $MDSCOUNT -ge 2 ] && skip "skip now for >= 2 MDTs" && return
+	local server_version=$(lustre_version_code $SINGLEMDS)
+	scrub_prep 20
+	scrub_backup_restore 1
+	echo "starting MDTs with OI scrub disabled"
+	scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
+	scrub_check_status 3 init
+	scrub_check_flags 4 inconsistent
+
+	# run under dryrun mode
+	if [ $server_version -lt $(version_code 2.5.58) ]; then
+		scrub_start 5 --dryrun on
+	else
+		scrub_start 5 --dryrun
+	fi
+	scrub_check_status 6 completed
+	scrub_check_flags 7 inconsistent
+	scrub_check_params 8 dryrun
+	scrub_check_repaired 9 20
+
+	# run under dryrun mode again
+	if [ $server_version -lt $(version_code 2.5.58) ]; then
+		scrub_start 10 --dryrun on
+	else
+		scrub_start 10 --dryrun
+	fi
+	scrub_check_status 11 completed
+	scrub_check_flags 12 inconsistent
+	scrub_check_params 13 dryrun
+	scrub_check_repaired 14 20
+
+	# run under normal mode
+	#
+	# Lustre-2.x (x <= 5) used "-n off" to disable dryrun which does not
+	# work under Lustre-2.y (y >= 6), the test script should be fixed as
+	# "-noff" or "--dryrun=off" or nothing by default.
+	if [ $server_version -lt $(version_code 2.5.58) ]; then
+		scrub_start 15 --dryrun off
+	else
+		scrub_start 15
+	fi
+	scrub_check_status 16 completed
+	scrub_check_flags 17 ""
+	scrub_check_params 18 ""
+	scrub_check_repaired 19 20
+
+	# run under normal mode again
+	if [ $server_version -lt $(version_code 2.5.58) ]; then
+		scrub_start 20 --dryrun off
+	else
+		scrub_start 20
+	fi
+	scrub_check_status 21 completed
+	scrub_check_flags 22 ""
+	scrub_check_params 23 ""
+	scrub_check_repaired 24 0
+}
+run_test 15 "Dryrun mode OI scrub"
+
 # restore MDS/OST size
 MDSSIZE=${SAVED_MDSSIZE}
 OSTSIZE=${SAVED_OSTSIZE}
+OSTCOUNT=${SAVED_OSTCOUNT}
 
 # cleanup the system at last
 formatall