X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-single.sh;h=b8b77314d56c88a7b91c6f409d3d4b9fa94e63a9;hp=fd6901fa83b03804aa2c266dd9e4bf6afbb4a69b;hb=f44fe5abbc74ca79790c100a30193ded1ef1e6c9;hpb=0444a40d9838b868092c78d3bdd4c7c3a00199e6

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index fd6901f..b8b7731 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -21,15 +21,17 @@ GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
 require_dsh_mds || exit 0
 
 # Skip these tests
-# bug number:  17466 18857      LU-1473
-ALWAYS_EXCEPT="61d   33a 33b    62	$REPLAY_SINGLE_EXCEPT"
+# bug number for skipped tests:
+#                                    LU-472
+ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT 61d"
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
-#                                                  63 min  7 min  AT AT AT AT"
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a      44b    65 66 67 68"
+#                                  7.5  (min)"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="44b"
 
 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
 # bug number for skipped test:        LU-1867	LU-3127
-        ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89	73b"
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89	73b"
 
 build_test_filter
 
@@ -1177,6 +1179,10 @@ run_test 52 "time out lock replay (3764)"
 
 # bug 3462 - simultaneous MDC requests
 test_53a() {
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
 	cancel_lru_locks mdc    # cleanup locks from former test cases
 	mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
 	mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
@@ -1281,6 +1287,10 @@ test_53c() {
 run_test 53c "|X| open request and close request while two MDC requests in flight"
 
 test_53d() {
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
 	cancel_lru_locks mdc    # cleanup locks from former test cases
 
 	mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
@@ -1527,8 +1537,11 @@ test_58b() {
 
 	large_xattr_enabled &&
 		orig="$(generate_string $(max_xattr_size))" || orig="bar"
+	# Original extended attribute can be long. Print a small version of
+	# attribute if an error occurs
+	local sm_msg=$(printf "%.9s" $orig)
 
-	mount_client $MOUNT2
+	mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
 	replay_barrier $SINGLEMDS
@@ -1536,43 +1549,52 @@ test_58b() {
 	fail $SINGLEMDS
 	new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
 	[[ "$new" = "$orig" ]] ||
-		error "xattr set ($orig) is not what was returned ($new)"
+		error "xattr set ($sm_msg...) differs from xattr get ($new)"
 	rm -f $DIR/$tdir/$tfile
 	rmdir $DIR/$tdir
 	cleanup_58
+	wait_clients_import_state ${CLIENTS:-$HOSTNAME} "mgs" FULL
 }
 run_test 58b "test replay of setxattr op"
 
 test_58c() { # bug 16570
-    local orig
-    local orig1
-    local new
-
-    trap cleanup_58 EXIT
-
-    if large_xattr_enabled; then
-        local xattr_size=$(max_xattr_size)
-        orig="$(generate_string $((xattr_size / 2)))"
-        orig1="$(generate_string $xattr_size)"
-    else
-        orig="bar"
-        orig1="bar1"
-    fi
+	local orig
+	local orig1
+	local new
+
+	trap cleanup_58 EXIT
+
+	if large_xattr_enabled; then
+		local xattr_size=$(max_xattr_size)
+		orig="$(generate_string $((xattr_size / 2)))"
+		orig1="$(generate_string $xattr_size)"
+	else
+		orig="bar"
+		orig1="bar1"
+	fi
+
+	# PING_INTERVAL max(obd_timeout / 4, 1U)
+	sleep $((TIMEOUT / 4))
+
+	# Original extended attribute can be long. Print a small version of
+	# attribute if an error occurs
+	local sm_msg=$(printf "%.9s" $orig)
+	local sm_msg1=$(printf "%.9s" $orig1)
 
-	mount_client $MOUNT2
+	mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
 	drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
 		error "drop_request for setfattr failed"
 	new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
 	[[ "$new" = "$orig" ]] ||
-		error "xattr set ($orig) is not what was returned ($new)"
+		error "xattr set ($sm_msg...) differs from xattr get ($new)"
 	drop_reint_reply "setfattr -n trusted.foo1 \
 			  -v $orig1 $DIR/$tdir/$tfile" ||
-		error "drop_request for setfattr failed"
+		error "drop_reint_reply for setfattr failed"
 	new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
 	[[ "$new" = "$orig1" ]] ||
-		error "second xattr set ($orig1) not what was returned ($new)"
+		error "second xattr set ($sm_msg1...) differs xattr get ($new)"
 	rm -f $DIR/$tdir/$tfile
 	rmdir $DIR/$tdir
 	cleanup_58
@@ -2006,15 +2028,6 @@ check_for_process () {
 	killall_process $clients "$prog" -0
 }
 
-killall_process () {
-	local clients=${1:-$(hostname)}
-	local name=$2
-	local signal=$3
-	local rc=0
-
-	do_nodes $clients "killall $signal $name"
-}
-
 test_70b () {
 	local clients=${CLIENTS:-$HOSTNAME}
 
@@ -2029,6 +2042,10 @@ test_70b () {
 	local start_ts=$(date +%s)
 	local cmd="rundbench 1 -t $duration"
 	local pid=""
+	if [ $MDSCOUNT -ge 2 ]; then
+		test_mkdir -p -c$MDSCOUNT $DIR/$tdir
+		$LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir
+	fi
 	do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
 		PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
 		DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
@@ -2049,6 +2066,7 @@ test_70b () {
 
 	elapsed=$(($(date +%s) - start_ts))
 	local num_failovers=0
+	local fail_index=1
 	while [ $elapsed -lt $duration ]; do
 		if ! check_for_process $clients dbench; then
 			error_noexit "dbench stopped on some of $clients!"
@@ -2056,20 +2074,418 @@ test_70b () {
 			break
 		fi
 		sleep 1
-		replay_barrier $SINGLEMDS
+		replay_barrier mds$fail_index
 		sleep 1 # give clients a time to do operations
 		# Increment the number of failovers
 		num_failovers=$((num_failovers+1))
-		log "$TESTNAME fail $SINGLEMDS $num_failovers times"
-		fail $SINGLEMDS
+		log "$TESTNAME fail mds$fail_index $num_failovers times"
+		fail mds$fail_index
 		elapsed=$(($(date +%s) - start_ts))
+		if [ $fail_index -ge $MDSCOUNT ]; then
+			fail_index=1
+		else
+			fail_index=$((fail_index+1))
+		fi
 	done
 
 	wait $pid || error "rundbench load on $clients failed!"
 }
-run_test 70b "mds recovery; $CLIENTCOUNT clients"
+run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
 # end multi-client tests
 
+random_fail_mdt() {
+	local max_index=$1
+	local duration=$2
+	local monitor_pid=$3
+	local elapsed
+	local start_ts=$(date +%s)
+	local num_failovers=0
+	local fail_index
+
+	elapsed=$(($(date +%s) - start_ts))
+	while [ $elapsed -lt $duration ]; do
+		fail_index=$((RANDOM%max_index+1))
+		kill -0 $monitor_pid ||
+			error "$monitor_pid stopped"
+		sleep 120
+		replay_barrier mds$fail_index
+		sleep 10
+		# Increment the number of failovers
+		num_failovers=$((num_failovers+1))
+		log "$TESTNAME fail mds$fail_index $num_failovers times"
+		fail mds$fail_index
+		elapsed=$(($(date +%s) - start_ts))
+	done
+}
+
+cleanup_70c() {
+	trap 0
+	rm -f $DIR/replay-single.70c.lck
+	rm -rf /$DIR/$tdir
+}
+
+test_70c () {
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=600
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_70c EXIT
+	(
+		while [ ! -e $DIR/replay-single.70c.lck ]; do
+			test_mkdir -p -c$MDSCOUNT $DIR/$tdir || break
+			if [ $MDSCOUNT -ge 2 ]; then
+				$LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
+				error "set default dirstripe failed"
+			fi
+			cd $DIR/$tdir || break
+			tar cf - /etc | tar xf - || error "tar failed in loop"
+		done
+	)&
+	tar_70c_pid=$!
+	echo "Started tar $tar_70c_pid"
+
+	random_fail_mdt $MDSCOUNT $duration $tar_70c_pid
+	kill -0 $tar_70c_pid || error "tar $tar_70c_pid stopped"
+
+	touch $DIR/replay-single.70c.lck
+	wait $tar_70c_pid || error "$?: tar failed"
+
+	cleanup_70c
+	true
+}
+run_test 70c "tar ${MDSCOUNT}mdts recovery"
+
+cleanup_70d() {
+	trap 0
+	kill -9 $mkdir_70d_pid
+}
+
+test_70d () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_70d EXIT
+	(
+		while true; do
+			$LFS mkdir -i0 -c2 $DIR/$tdir/test || {
+				echo "mkdir fails"
+				break
+			}
+			$LFS mkdir -i1 -c2 $DIR/$tdir/test1 || {
+				echo "mkdir fails"
+				break
+			}
+
+			touch $DIR/$tdir/test/a || {
+				echo "touch fails"
+				break;
+			}
+			mkdir $DIR/$tdir/test/b || {
+				echo "mkdir fails"
+				break;
+			}
+			rm -rf $DIR/$tdir/test || {
+				echo "rmdir fails"
+				break
+			}
+
+			touch $DIR/$tdir/test1/a || {
+				echo "touch fails"
+				break;
+			}
+			mkdir $DIR/$tdir/test1/b || {
+				echo "mkdir fails"
+				break;
+			}
+
+			rm -rf $DIR/$tdir/test1 || {
+				echo "rmdir fails"
+				break
+			}
+		done
+	)&
+	mkdir_70d_pid=$!
+	echo "Started  $mkdir_70d_pid"
+
+	random_fail_mdt $MDSCOUNT $duration $mkdir_70d_pid
+	kill -0 $mkdir_70d_pid || error "mkdir/rmdir $mkdir_70d_pid stopped"
+
+	cleanup_70d
+	true
+}
+run_test 70d "mkdir/rmdir striped dir ${MDSCOUNT}mdts recovery"
+
+cleanup_70e() {
+	trap 0
+	kill -9 $rename_70e_pid
+}
+
+test_70e () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	echo ha > /proc/sys/lnet/debug
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i0 $DIR/$tdir/test_0
+	$LFS mkdir -i0 $DIR/$tdir/test_1
+	touch $DIR/$tdir/test_0/a
+	touch $DIR/$tdir/test_1/b
+	trap cleanup_70e EXIT
+	(
+		while true; do
+			mrename $DIR/$tdir/test_0/a $DIR/$tdir/test_1/b > \
+						/dev/null || {
+				echo "a->b fails"
+				break;
+			}
+
+			checkstat $DIR/$tdir/test_0/a && {
+				echo "a still exists"
+				break
+			}
+
+			checkstat $DIR/$tdir/test_1/b || {
+				echo "b still  exists"
+				break
+			}
+
+			touch $DIR/$tdir/test_0/a || {
+				echo "touch a fails"
+				break
+			}
+
+			mrename $DIR/$tdir/test_1/b $DIR/$tdir/test_0/a > \
+						/dev/null || {
+				echo "a->a fails"
+				break;
+			}
+		done
+	)&
+	rename_70e_pid=$!
+	echo "Started  $rename_70e_pid"
+
+	random_fail_mdt 2 $duration $rename_70e_pid
+	kill -0 $rename_70e_pid || error "rename $rename_70e_pid stopped"
+
+	cleanup_70e
+	true
+}
+run_test 70e "rename cross-MDT with random fails"
+
+test_70f_write_and_read(){
+	local srcfile=$1
+	local stopflag=$2
+	local client
+
+	echo "Write/read files in: '$DIR/$tdir', clients: '$CLIENTS' ..."
+	for client in ${CLIENTS//,/ }; do
+		[ -f $stopflag ] || return
+
+		local tgtfile=$DIR/$tdir/$tfile.$client
+		do_node $client dd $DD_OPTS bs=1M count=10 if=$srcfile \
+			of=$tgtfile 2>/dev/null ||
+			error "dd $DD_OPTS bs=1M count=10 if=$srcfile " \
+			      "of=$tgtfile failed on $client, rc=$?"
+	done
+
+	local prev_client=$(echo ${CLIENTS//,/ } | awk '{ print $NF }')
+	local index=0
+
+	for client in ${CLIENTS//,/ }; do
+		[ -f $stopflag ] || return
+
+		# flush client cache in case test is running on only one client
+		# do_node $client cancel_lru_locks osc
+		do_node $client $LCTL set_param ldlm.namespaces.*.lru_size=clear
+
+		tgtfile=$DIR/$tdir/$tfile.$client
+		local md5=$(do_node $prev_client "md5sum $tgtfile")
+		[ ${checksum[$index]// */} = ${md5// */} ] ||
+			error "$tgtfile: checksum doesn't match on $prev_client"
+		index=$((index + 1))
+		prev_client=$client
+	done
+}
+
+test_70f_loop(){
+	local srcfile=$1
+	local stopflag=$2
+	DD_OPTS=
+
+	mkdir -p $DIR/$tdir || error "cannot create $DIR/$tdir directory"
+	$SETSTRIPE -c -1 $DIR/$tdir || error "cannot $SETSTRIPE $DIR/$tdir"
+
+	touch $stopflag
+	while [ -f $stopflag ]; do
+		test_70f_write_and_read $srcfile $stopflag
+		# use direct IO and buffer cache in turns if loop
+		[ -n "$DD_OPTS" ] && DD_OPTS="" || DD_OPTS="oflag=direct"
+	done
+}
+
+test_70f_cleanup() {
+	trap 0
+	rm -f $TMP/$tfile.stop
+	do_nodes $CLIENTS rm -f $TMP/$tfile
+	rm -f $DIR/$tdir/$tfile.*
+}
+
+test_70f() {
+#	[ x$ost1failover_HOST = x$ost_HOST ] &&
+#		{ skip "Failover host not defined" && return; }
+#	[ -z "$CLIENTS" ] &&
+#		{ skip "CLIENTS are not specified." && return; }
+#	[ $CLIENTCOUNT -lt 2 ] &&
+#		{ skip "Need 2 or more clients, have $CLIENTCOUNT" && return; }
+
+	echo "mount clients $CLIENTS ..."
+	zconf_mount_clients $CLIENTS $MOUNT
+
+	local srcfile=$TMP/$tfile
+	local client
+	local index=0
+
+	trap test_70f_cleanup EXIT
+	# create a different source file local to each client node so we can
+	# detect if the file wasn't written out properly after failover
+	do_nodes $CLIENTS dd bs=1M count=10 if=/dev/urandom of=$srcfile \
+		2>/dev/null || error "can't create $srcfile on $CLIENTS"
+	for client in ${CLIENTS//,/ }; do
+		checksum[$index]=$(do_node $client "md5sum $srcfile")
+		index=$((index + 1))
+	done
+
+	local duration=120
+	[ "$SLOW" = "no" ] && duration=60
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	local stopflag=$TMP/$tfile.stop
+	test_70f_loop $srcfile $stopflag &
+	local pid=$!
+
+	local elapsed=0
+	local num_failovers=0
+	local start_ts=$SECONDS
+	while [ $elapsed -lt $duration ]; do
+		sleep 3
+		replay_barrier ost1
+		sleep 1
+		num_failovers=$((num_failovers + 1))
+		log "$TESTNAME failing OST $num_failovers times"
+		fail ost1
+		sleep 2
+		elapsed=$((SECONDS - start_ts))
+	done
+
+	rm -f $stopflag
+	wait $pid
+	test_70f_cleanup
+}
+run_test 70f "OSS O_DIRECT recovery with $CLIENTCOUNT clients"
+
+cleanup_71a() {
+	trap 0
+	kill -9 $mkdir_71a_pid
+}
+
+random_double_fail_mdt() {
+	local max_index=$1
+	local duration=$2
+	local monitor_pid=$3
+	local elapsed
+	local start_ts=$(date +%s)
+	local num_failovers=0
+	local fail_index
+	local second_index
+
+	elapsed=$(($(date +%s) - start_ts))
+	while [ $elapsed -lt $duration ]; do
+		fail_index=$((RANDOM%max_index + 1))
+		if [ $fail_index -eq $max_index ]; then
+			second_index=1
+		else
+			second_index=$((fail_index + 1))
+		fi
+		kill -0 $monitor_pid ||
+			error "$monitor_pid stopped"
+		sleep 120
+		replay_barrier mds$fail_index
+		replay_barrier mds$second_index
+		sleep 10
+		# Increment the number of failovers
+		num_failovers=$((num_failovers+1))
+		log "fail mds$fail_index mds$second_index $num_failovers times"
+		fail mds${fail_index},mds${second_index}
+		elapsed=$(($(date +%s) - start_ts))
+	done
+}
+
+test_71a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_71a EXIT
+	(
+		while true; do
+			$LFS mkdir -i0 -c2 $DIR/$tdir/test
+			rmdir $DIR/$tdir/test
+		done
+	)&
+	mkdir_71a_pid=$!
+	echo "Started  $mkdir_71a_pid"
+
+	random_double_fail_mdt 2 $duration $mkdir_71a_pid
+	kill -0 $mkdir_71a_pid || error "mkdir/rmdir $mkdir_71a_pid stopped"
+
+	cleanup_71a
+	true
+}
+run_test 71a "mkdir/rmdir striped dir with 2 mdts recovery"
+
 test_73a() {
 	multiop_bg_pause $DIR/$tfile O_tSc ||
 		error "multiop_bg_pause $DIR/$tfile failed"
@@ -2122,11 +2538,16 @@ run_test 74 "Ensure applications don't fail waiting for OST recovery"
 
 remote_dir_check_80() {
 	local MDTIDX=1
-	local diridx=$($GETSTRIPE -M $remote_dir)
+	local diridx
+	local fileidx
+
+	diridx=$($GETSTRIPE -M $remote_dir) ||
+		error "$GETSTRIPE -M $remote_dir failed"
 	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
 
 	createmany -o $remote_dir/f-%d 20 || error "creation failed"
-	local fileidx=$($GETSTRIPE -M $remote_dir/f-1)
+	fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
+		error "$GETSTRIPE -M $remote_dir/f-1 failed"
 	[ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
 
 	return 0
@@ -2148,6 +2569,7 @@ test_80a() {
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
 	fail mds${MDTIDX}
 
 	wait $CLIENT_PID || error "remote creation failed"
@@ -2175,6 +2597,8 @@ test_80b() {
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "remote creation failed"
@@ -2202,6 +2626,8 @@ test_80c() {
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX}
 	fail mds$((MDTIDX + 1))
 
@@ -2228,6 +2654,9 @@ test_80d() {
 	# sleep 3 seconds to make sure MDTs are failed after
 	# lfs mkdir -i has finished on all of MDTs.
 	sleep 3
+
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX},mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "remote creation failed"
@@ -2259,6 +2688,7 @@ test_80e() {
 	# lfs mkdir -i has finished on all of MDTs.
 	sleep 3
 
+	replay_barrier mds1
 	fail mds${MDTIDX}
 
 	wait $CLIENT_PID || error "remote creation failed"
@@ -2285,6 +2715,7 @@ test_80f() {
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds2
 	fail mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "remote creation failed"
@@ -2316,6 +2747,8 @@ test_80g() {
 	# lfs mkdir -i has finished on all of MDTs.
 	sleep 3
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX}
 	fail mds$((MDTIDX + 1))
 
@@ -2343,6 +2776,8 @@ test_80h() {
 	# lfs mkdir -i has finished on all of MDTs.
 	sleep 3
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX},mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "remote dir creation failed"
@@ -2373,6 +2808,7 @@ test_81a() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds2
 	fail mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2402,6 +2838,7 @@ test_81b() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
 	fail mds${MDTIDX}
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2432,6 +2869,8 @@ test_81c() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX}
 	fail mds$((MDTIDX + 1))
 
@@ -2458,6 +2897,8 @@ test_81d() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX},mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2489,6 +2930,7 @@ test_81e() {
 	local CLIENT_PID=$!
 	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
 
+	replay_barrier mds1
 	fail mds${MDTIDX}
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2519,6 +2961,7 @@ test_81f() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds2
 	fail mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2549,6 +2992,8 @@ test_81g() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX}
 	fail mds$((MDTIDX + 1))
 
@@ -2575,6 +3020,8 @@ test_81h() {
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
 
+	replay_barrier mds1
+	replay_barrier mds2
 	fail mds${MDTIDX},mds$((MDTIDX + 1))
 
 	wait $CLIENT_PID || error "rm remote dir failed"
@@ -2607,7 +3054,7 @@ test_85a() { #bug 16774
 	done
 
 	lov_id=$(lctl dl | grep "clilov")
-	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
+	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
 	count=$(lctl get_param -n \
 		ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
 	echo "before recovery: unused locks count = $count"
@@ -2627,7 +3074,7 @@ run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
 test_85b() { #bug 16774
 	lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
 
-	do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME ||
+	create_pool $FSNAME.$TESTNAME ||
 		error "unable to create pool $TESTNAME"
 	do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
 		error "unable to add pool $TESTNAME"
@@ -2647,11 +3094,11 @@ test_85b() { #bug 16774
 	done
 
 	lov_id=$(lctl dl | grep "clilov")
-	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
-	count=$(lctl get_param \
-		-n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
+	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
+	count=$(lctl get_param -n \
+			  ldlm.namespaces.*OST0000*$addr.lock_unused_count)
 	echo "before recovery: unused locks count = $count"
-	[ $count != 0 ] || error "unused locks ($count) should be zero"
+	[ $count -ne 0 ] || error "unused locks ($count) should be zero"
 
 	fail ost1
 
@@ -2680,7 +3127,7 @@ test_86() {
 }
 run_test 86 "umount server after clear nid_stats should not hit LBUG"
 
-test_87() {
+test_87a() {
 	do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
 
 	replay_barrier ost1
@@ -2696,7 +3143,7 @@ test_87() {
 		error "New checksum $cksum2 does not match original $cksum"
 	fi
 }
-run_test 87 "write replay"
+run_test 87a "write replay"
 
 test_87b() {
 	do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
@@ -2847,6 +3294,8 @@ test_90() { # bug 19494
             return 0
         fi
     fi
+	# ensure all OSTs are active to allow allocations
+	wait_osts_up
 
 	mkdir $dir || error "mkdir $dir failed"
 
@@ -2918,7 +3367,7 @@ test_90() { # bug 19494
 }
 run_test 90 "lfs find identifies the missing striped file segments"
 
-test_93() {
+test_93a() {
 	local server_version=$(lustre_version_code $SINGLEMDS)
 		[[ $server_version -ge $(version_code 2.6.90) ]] ||
 		[[ $server_version -ge $(version_code 2.5.4) &&
@@ -2940,7 +3389,28 @@ test_93() {
 	do_facet ost1 "$LCTL set_param fail_loc=0x715"
 	fail ost1
 }
-run_test 93 "replay + reconnect"
+run_test 93a "replay + reconnect"
+
+test_93b() {
+	local server_version=$(lustre_version_code $SINGLEMDS)
+		[[ $server_version -ge $(version_code 2.7.90) ]] ||
+		{ skip "Need MDS version 2.7.90+"; return; }
+
+	cancel_lru_locks mdc
+
+	createmany -o $DIR/$tfile 20 ||
+			error "createmany -o $DIR/$tfile failed"
+
+	#define OBD_FAIL_TGT_REPLAY_RECONNECT     0x715
+	# We need to emulate a state that MDT is waiting for other clients
+	# not completing the recovery. Final ping is queued, but reply will be
+	# sent on the recovery completion. It is done by sleep before
+	# processing final pings
+	do_facet mds1 "$LCTL set_param fail_val=80"
+	do_facet mds1 "$LCTL set_param fail_loc=0x715"
+	fail mds1
+}
+run_test 93b "replay + reconnect on mds"
 
 striped_dir_check_100() {
 	local striped_dir=$DIR/$tdir/striped_dir
@@ -3035,6 +3505,1095 @@ test_101() { #LU-5648
 }
 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
 
+test_102a() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop request on MDT to force resend
+	#define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
+	do_facet $facet "$LCTL set_param fail_loc=0x159"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
+
+test_102b() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop reply on MDT to force reconstruction
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
+
+test_102c() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	replay_barrier $facet
+
+	# drop reply on MDT
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+
+	# fail MDT
+	fail $facet
+
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
+
+test_102d() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop reply on MDT
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+
+	# write MDT transactions to disk
+	do_facet $facet "sync; sync; sync"
+
+	do_facet $facet "$LCTL set_param fail_loc=0"
+
+	# fail MDT
+	fail $facet
+
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
+
+test_103() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+#define OBD_FAIL_MDS_TRACK_OVERFLOW 0x162
+	do_facet mds1 $LCTL set_param fail_loc=0x80000162
+
+	mkdir -p $DIR/$tdir
+	createmany -o $DIR/$tdir/t- 30 ||
+		error "create files on remote directory failed"
+	sync
+	rm -rf $DIR/$tdir/t-*
+	sync
+#MDS should crash with tr->otr_next_id overflow
+	fail mds1
+}
+run_test 103 "Check otr_next_id overflow"
+
+
+check_striped_dir_110()
+{
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
+			error "create striped dir failed"
+	local stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
+	[ $stripe_count -eq $MDSCOUNT ] ||
+		error "$stripe_count != 2 after recovery"
+}
+
+test_110a() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds1
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	fail mds1
+
+	check_striped_dir_110 || error "check striped_dir failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110a "DNE: create striped dir, fail MDT1"
+
+test_110b() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds1
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	umount $MOUNT
+	fail mds1
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110b "DNE: create striped dir, fail MDT1 and client"
+
+test_110c() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds2
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	fail mds2
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110c "DNE: create striped dir, fail MDT2"
+
+test_110d() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds2
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	umount $MOUNT
+	fail mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110d "DNE: create striped dir, fail MDT2 and client"
+
+test_110e() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds2
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	umount $MOUNT
+	replay_barrier mds1
+	fail mds1,mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
+
+test_110f() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds1
+	replay_barrier mds2
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	fail mds1,mds2
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
+
+test_110g() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds1
+	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
+	umount $MOUNT
+	replay_barrier mds2
+	fail mds1,mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	check_striped_dir_110 || error "check striped_dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
+
+test_111a() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds1
+	rm -rf $DIR/$tdir/striped_dir
+	fail mds1
+
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111a "DNE: unlink striped dir, fail MDT1"
+
+test_111b() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds2
+	rm -rf $DIR/$tdir/striped_dir
+	umount $MOUNT
+	fail mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111b "DNE: unlink striped dir, fail MDT2"
+
+test_111c() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds1
+	rm -rf $DIR/$tdir/striped_dir
+	umount $MOUNT
+	replay_barrier mds2
+	fail mds1,mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
+
+test_111d() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds2
+	rm -rf $DIR/$tdir/striped_dir
+	umount $MOUNT
+	replay_barrier mds1
+	fail mds1,mds2
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+
+	return 0
+}
+run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
+
+test_111e() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds2
+	rm -rf $DIR/$tdir/striped_dir
+	replay_barrier mds1
+	fail mds1,mds2
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
+
+test_111f() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds1
+	rm -rf $DIR/$tdir/striped_dir
+	replay_barrier mds2
+	fail mds1,mds2
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
+
+test_111g() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
+	replay_barrier mds1
+	replay_barrier mds2
+	rm -rf $DIR/$tdir/striped_dir
+	fail mds1,mds2
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
+			error "striped dir still exists"
+	return 0
+}
+run_test 111g "DNE: unlink striped dir, fail MDT1/MDT2"
+
+test_112_rename_prepare() {
+	mkdir -p $DIR/$tdir/src_dir
+	$LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
+		error "create remote source failed"
+
+	touch $DIR/$tdir/src_dir/src_child/a
+
+	$LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
+		error "create remote target dir failed"
+
+	$LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
+		error "create remote target child failed"
+}
+
+test_112_check() {
+	find $DIR/$tdir/
+	$CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
+		error "src_child still exists after rename"
+
+	$CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
+		error "missing file(a) after rename"
+}
+
+test_112a() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+	fail mds1
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112a "DNE: cross MDT rename, fail MDT1"
+
+test_112b() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds2
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds2
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112b "DNE: cross MDT rename, fail MDT2"
+
+test_112c() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds3
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds3
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112c "DNE: cross MDT rename, fail MDT3"
+
+test_112d() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112d "DNE: cross MDT rename, fail MDT4"
+
+test_112e() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds2
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds2
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
+
+test_112f() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds3
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds3
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
+
+test_112g() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
+
+test_112h() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds2
+	replay_barrier mds3
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds2,mds3
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
+
+test_112i() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds2
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds2,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
+
+test_112j() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds3
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds3,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
+
+test_112k() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds2
+	replay_barrier mds3
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds2,mds3
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
+
+test_112l() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds2
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds2,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
+
+test_112m() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds1
+	replay_barrier mds3
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds1,mds3,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
+
+test_112n() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	test_112_rename_prepare
+	replay_barrier mds2
+	replay_barrier mds3
+	replay_barrier mds4
+
+	mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
+		error "rename dir cross MDT failed!"
+
+	fail mds2,mds3,mds4
+
+	test_112_check
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
+
+test_115() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local fail_index=0
+	local index
+	local i
+	local j
+
+	mkdir -p $DIR/$tdir
+	for ((j=0;j<$((MDSCOUNT));j++)); do
+		fail_index=$((fail_index+1))
+		index=$((fail_index % MDSCOUNT))
+		replay_barrier mds$((index + 1))
+		for ((i=0;i<5;i++)); do
+			test_mkdir -i$index -c$MDSCOUNT $DIR/$tdir/test_$i ||
+				error "create striped dir $DIR/$tdir/test_$i"
+		done
+
+		fail mds$((index + 1))
+		for ((i=0;i<5;i++)); do
+			checkstat -t dir $DIR/$tdir/test_$i ||
+				error "$DIR/$tdir/test_$i does not exist!"
+		done
+		rm -rf $DIR/$tdir/test_* ||
+				error "rmdir fails"
+	done
+}
+run_test 115 "failover for create/unlink striped directory"
+
+test_116a() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
+		skip "Do not support large update log before 2.7.55" &&
+		return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local fail_index=0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds1
+
+	# OBD_FAIL_SPLIT_UPDATE_REC       0x1702
+	do_facet mds1 "lctl set_param fail_loc=0x80001702"
+	$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
+
+	fail mds1
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
+		error "stried_dir does not exists"
+}
+run_test 116a "large update log master MDT recovery"
+
+test_116b() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
+		skip "Do not support large update log before 2.7.55" &&
+		return 0
+
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local fail_index=0
+
+	mkdir -p $DIR/$tdir
+	replay_barrier mds2
+
+	# OBD_FAIL_SPLIT_UPDATE_REC       0x1702
+	do_facet mds2 "lctl set_param fail_loc=0x80001702"
+	$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
+
+	fail mds2
+	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
+		error "stried_dir does not exists"
+}
+run_test 116b "large update log slave MDT recovery"
+
+test_117() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local index
+	local mds_indexs
+
+	mkdir -p $DIR/$tdir
+	$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/remote_dir
+	$LFS setdirstripe -i1 -c$MDSCOUNT $DIR/$tdir/remote_dir_1
+	sleep 2
+
+	# Let's set rdonly on all MDTs, so client will send
+	# replay requests on all MDTs and replay these requests
+	# at the same time. This test will verify the recovery
+	# will not be deadlock in this case, LU-7531.
+	for ((index = 0; index < $((MDSCOUNT)); index++)); do
+		replay_barrier mds$((index + 1))
+		if [ -z $mds_indexs ]; then
+			mds_indexs="${mds_indexs}mds$((index+1))"
+		else
+			mds_indexs="${mds_indexs},mds$((index+1))"
+		fi
+	done
+
+	rm -rf $DIR/$tdir/remote_dir
+	rm -rf $DIR/$tdir/remote_dir_1
+
+	fail $mds_indexs
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 117 "DNE: cross MDT unlink, fail MDT1 and MDT2"
+
+test_118() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
+		error "setdirstripe fails"
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1 ||
+		error "setdirstripe fails 1"
+	rm -rf $DIR/$tdir/striped_dir* || error "rmdir fails"
+
+	# OBD_FAIL_INVALIDATE_UPDATE       0x1705
+	do_facet mds1 "lctl set_param fail_loc=0x1705"
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
+	do_facet mds1 "lctl set_param fail_loc=0x0"
+
+	replay_barrier mds1
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
+	fail mds1
+
+	true
+}
+run_test 118 "invalidate osp update will not cause update log corruption"
+
+test_119() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+	local stripe_count
+	local hard_timeout=$(do_facet mds1 \
+		"lctl get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard")
+
+	local clients=${CLIENTS:-$HOSTNAME}
+	local time_min=$(recovery_time_min)
+
+	mkdir -p $DIR/$tdir
+	mkdir $DIR/$tdir/tmp
+	rmdir $DIR/$tdir/tmp
+
+	replay_barrier mds1
+	mkdir $DIR/$tdir/dir_1
+	for ((i = 0; i < 20; i++)); do
+		$LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i
+	done
+
+	stop mds1
+	change_active mds1
+	wait_for_facet mds1
+
+	#define OBD_FAIL_TGT_REPLAY_DELAY  0x714
+	do_facet mds1 $LCTL set_param fail_loc=0x80000714
+	#sleep (timeout + 5), so mds will evict the client exports,
+	#but DNE update recovery will keep going.
+	do_facet mds1 $LCTL set_param fail_val=$((time_min + 5))
+
+	mount_facet mds1 "-o recovery_time_hard=$time_min"
+
+	wait_clients_import_state "$clients" mds1 FULL
+
+	clients_up || clients_up || error "failover df: $?"
+
+	#revert back the hard timeout
+	do_facet mds1 $LCTL set_param \
+		mdt.$FSNAME-MDT0000.recovery_time_hard=$hard_timeout
+
+	for ((i = 0; i < 20; i++)); do
+		stripe_count=$($LFS getdirstripe -c $DIR/$tdir/stripe_dir-$i)
+		[ $stripe_count == 2 ] || {
+			error "stripe_dir-$i creation replay fails"
+			break
+		}
+	done
+}
+run_test 119 "timeout of normal replay does not cause DNE replay fails  "
+
+test_120() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+
+	mkdir $DIR/$tdir
+	replay_barrier_nosync mds1
+	for ((i = 0; i < 20; i++)); do
+		mkdir $DIR/$tdir/dir-$i || {
+			error "create dir-$i fails"
+			break
+		}
+		$LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i || {
+			error "create stripe_dir-$i fails"
+			break
+		}
+	done
+
+	fail_abort mds1
+
+	for ((i = 0; i < 20; i++)); do
+		[ ! -e "$DIR/$tdir/dir-$i" ] || {
+			error "dir-$i still exists"
+			break
+		}
+		[ ! -e "$DIR/$tdir/stripe_dir-$i" ] || {
+			error "stripe_dir-$i still exists"
+			break
+		}
+	done
+}
+run_test 120 "DNE fail abort should stop both normal and DNE replay"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status