X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=6ce8a57330bf5c1310e096f244a2d07763cb956c;hb=814bb394843434883a94fe6432cd8c656035a3e1;hp=231ad43bd9b46621be6c09188edfed5cd11e0401;hpb=3341c8c31871ad5bcea914260643bf164194ee9a;p=fs%2Flustre-release.git

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 231ad43..6ce8a57 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,25 +2,23 @@
 
 set -e
 
-#         bug  5493  LU2034
-ALWAYS_EXCEPT="52 $RECOVERY_SMALL_EXCEPT"
-
-export MULTIOP=${MULTIOP:-multiop}
 PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
 . $LUSTRE/tests/test-framework.sh
 init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
 
-require_dsh_mds || exit 0
-
-# also long tests: 19, 21a, 21e, 21f, 23, 27
-
-[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
+ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
+if $SHARED_KEY; then
+	# bug number for skipped test: LU-12896
+	ALWAYS_EXCEPT+="               110k"
+	# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+fi
 
 build_test_filter
 
+require_dsh_mds || exit 0
+
 # Allow us to override the setup if we already have a mounted system by
 # setting SETUP=" " and CLEANUP=" "
 SETUP=${SETUP:-""}
@@ -171,8 +169,8 @@ test_10b() {
 	local before=$(date +%s)
 	local evict
 
-	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
-		skip "Need MDS version at least 2.6.53" && return
+	[[ "$MDS1_VERSION" -lt $(version_code 2.6.53) ]] &&
+		skip "Need MDS version at least 2.6.53"
 	do_facet client "stat $DIR > /dev/null"  ||
 		error "failed to stat $DIR: $?"
 	drop_bl_callback_once "chmod 0777 $DIR" ||
@@ -243,8 +241,8 @@ test_10d() {
 	local before=$(date +%s)
 	local evict
 
-	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
-		skip "Need MDS version at least 2.6.90" && return
+	[[ "$MDS1_VERSION" -lt $(version_code 2.6.90) ]] &&
+		skip "Need MDS version at least 2.6.90"
 
 	# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
 	sleep 1
@@ -284,11 +282,11 @@ run_test 10d "test failed blocking ast"
 
 test_10e()
 {
-	[[ $(lustre_version_code ost1) -le $(version_code 2.8.58) ]] &&
-		skip "Need OST version at least 2.8.59" && return 0
-	[ $CLIENTCOUNT -lt 2 ] && skip "need two clients" && return 0
+	[[ "$OST1_VERSION" -le $(version_code 2.8.58) ]] &&
+		skip "Need OST version at least 2.8.59"
+	[ $CLIENTCOUNT -lt 2 ] && skip "need two clients"
 	[ $(facet_host client) == $(facet_host ost1) ] &&
-		skip "need ost1 and client on different nodes" && return 0
+		skip "need ost1 and client on different nodes"
 	local -a clients=(${CLIENTS//,/ })
 	local client1=${clients[0]}
 	local client2=${clients[1]}
@@ -1231,7 +1229,7 @@ test_51() {
         for i in $SEQ
         do
 		#echo failover in $i sec
-		log "test_$testnum: failover in $i sec"
+		log "$TESTNAME: failover in $i sec"
 		sleep $i
 		facet_failover $SINGLEMDS
         done
@@ -1294,15 +1292,16 @@ test_53() {
 run_test 53 "touch: drop rep"
 
 test_54() {
-	zconf_mount `hostname` $MOUNT2
-        touch $DIR/$tfile
-        touch $DIR2/$tfile.1
-        sleep 10
-        cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
-        fail $SINGLEMDS
-        umount $MOUNT2
-        ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
-        [ x"$ERROR" == x ] || error "back in time occured"
+	zconf_mount $(hostname) $MOUNT2
+	touch $DIR/$tfile
+	touch $DIR2/$tfile.1
+	sleep 10
+	cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
+	fail $SINGLEMDS
+	umount $MOUNT2
+	ERROR=$(dmesg | egrep "(test 54|went back in time)" | tail -n1 |
+		grep "went back in time")
+	[ x"$ERROR" == x ] || error "back in time occured"
 }
 run_test 54 "back in time"
 
@@ -1322,7 +1321,7 @@ test_55() {
 	# Minimum pass speed is 2MBps
 	local ddtimeout=64
 	# LU-2887/LU-3089 - set min pass speed to 500KBps
-	[ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
+	[ "$ost1_FSTYPE" = zfs ] && ddtimeout=256
 
 	# first dd should be finished quickly
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
@@ -1531,8 +1530,8 @@ test_65() {
 	mount_client $DIR2
 
 	#grant lock1, export2
-	$SETSTRIPE -i -0 $DIR2/$tfile || return 1
-	$MULTIOP $DIR2/$tfile Ow  || return 2
+	$LFS setstripe -i -0 $DIR2/$tfile || error "setstripe failed"
+	$MULTIOP $DIR2/$tfile Ow  || error "multiop failed"
 
 #define OBD_FAIL_LDLM_BL_EVICT            0x31e
 	do_facet ost $LCTL set_param fail_loc=0x31e
@@ -1562,8 +1561,8 @@ run_test 65 "lock enqueue for destroyed export"
 
 test_66()
 {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
-		{ skip "Need MDS version at least 2.7.51"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.51) ]] ||
+		skip "Need MDS version at least 2.7.51"
 
 	local list=$(comma_list $(osts_nodes))
 
@@ -1580,7 +1579,8 @@ test_66()
 	do_nodes $list $LCTL set_param fail_loc=0x80000136
 
 	#initiate the re-connect & re-send
-	local mdccli=$($LCTL dl | awk '/-MDT0000-mdc-/ {print $4;}')
+	local mdtname="MDT0000"
+	local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
 	local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
 	$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
 	sleep 2
@@ -1928,7 +1928,7 @@ test_105()
 	# Since the client just mounted, its last_rcvd entry is not on disk.
 	# Send an RPC so exp_need_sync forces last_rcvd to commit this export
 	# so the client can reconnect during OST recovery (LU-924, LU-1582)
-	$SETSTRIPE -i 0 $DIR/$tfile
+	$LFS setstripe -i 0 $DIR/$tfile
 	dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
 
         # make sure MGS's state is Partial
@@ -1964,8 +1964,8 @@ cleanup_106() {
 }
 
 test_106() { # LU-1789
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
-		{ skip "Need MDS version at least 2.3.50"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.3.50) ]] ||
+		skip "Need MDS version at least 2.3.50"
 
 #define OBD_FAIL_MDC_LIGHTWEIGHT         0x805
 	$LCTL set_param fail_loc=0x805
@@ -1991,9 +1991,8 @@ test_106() { # LU-1789
 	# lightweight goes through LUSTRE_IMP_RECOVER during failover
 	touch -c $DIR2/$tfile || true
 	$LCTL dk $TMP/lustre-log-$TESTNAME.log
-	recovered=`awk '/MDT0000-mdc-[0-9a-f]*: lwp recover/ {
-				      print;
-		      }' $TMP/lustre-log-$TESTNAME.log`
+	recovered=$(awk '/MDT0000-mdc-[0-9a-f]*. lwp recover/ { print }' \
+		    $TMP/lustre-log-$TESTNAME.log)
 	[ -z "$recovered" ] && error "lightweight client was not recovered"
 
 	# and all operations performed by lightweight client should be
@@ -2031,7 +2030,7 @@ run_test 107 "drop reint reply, then restart MDT"
 
 test_108() {
 	mkdir -p $DIR/$tdir
-	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
+	$LFS setstripe -c 1 -i 0 $DIR/$tdir
 
 	dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
 	local dd_pid=$!
@@ -2094,7 +2093,7 @@ test_110c () {
 	drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
 		error "lfs mkdir failed"
 
-	diridx=$($GETSTRIPE -m $remote_dir)
+	diridx=$($LFS getstripe -m $remote_dir)
 	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
@@ -2147,10 +2146,10 @@ test_110f () {
 run_test 110f "remove remote directory: drop slave rep"
 
 test_110g () {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.11.0) ]] ||
-		{ skip "Need MDS version at least 2.11.0"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
+		skip "Need MDS version at least 2.11.0"
 
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
 
 	mkdir -p $DIR/$tdir
 	touch $DIR/$tdir/$tfile
@@ -2174,10 +2173,9 @@ test_110g () {
 run_test 110g "drop reply during migration"
 
 test_110h () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; return 0; }
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
 
 	local src_dir=$DIR/$tdir/source_dir
 	local tgt_dir=$DIR/$tdir/target_dir
@@ -2202,10 +2200,9 @@ test_110h () {
 run_test 110h "drop update reply during cross-MDT file rename"
 
 test_110i () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; return 0; }
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
 
 	local src_dir=$DIR/$tdir/source_dir
 	local tgt_dir=$DIR/$tdir/target_dir
@@ -2233,10 +2230,9 @@ test_110i () {
 run_test 110i "drop update reply during cross-MDT dir rename"
 
 test_110j () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; return 0; }
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
 
 	local remote_dir=$DIR/$tdir/remote_dir
 	local local_dir=$DIR/$tdir/local_dir
@@ -2256,19 +2252,43 @@ test_110j () {
 }
 run_test 110j "drop update reply during cross-MDT ln"
 
+test_110k() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTS"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
+		skip "Need MDS version at least 2.12.55"
+
+	stop mds2 || error "stop mds2 failed"
+	umount $MOUNT
+
+#define OBD_FAIL_FLD_QUERY_REQ 0x1103
+	do_facet mds2 lctl set_param fail_loc=0x1103
+	local OPTS="$MDS_MOUNT_OPTS -o abort_recovery"
+	start mds2 $(mdsdevname 2) $OPTS ||
+		error "start MDS with abort_recovery should succeed"
+	do_facet mds2 lctl set_param fail_loc=0
+
+	# cleanup
+	stop mds2 || error "cleanup: stop mds2 failed"
+	start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS ||
+		error "cleanup: start mds2 failed"
+	zconf_mount $(hostname) $MOUNT || error "cleanup: mount failed"
+	client_up || error "post-failover df failed"
+}
+run_test 110k "FID_QUERY failed during recovery"
+
 # LU-2844 mdt prepare fail should not cause umount oops
 test_111 ()
 {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
-		{ skip "Need MDS version at least 2.3.62"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.3.62) ]] ||
+		skip "Need MDS version at least 2.3.62"
 
 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x151
 	stop $SINGLEMDS || error "stop MDS failed"
-	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) &&
+	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS &&
 		error "start MDS should fail"
 	do_facet $SINGLEMDS lctl set_param fail_loc=0
-	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
+	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
 		error "start MDS failed"
 }
 run_test 111 "mdd setup fail should not cause umount oops"
@@ -2339,8 +2359,8 @@ test_115_write() {
 }
 
 test_115a() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
 	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
@@ -2349,8 +2369,8 @@ test_115a() {
 run_test 115a "read: late REQ MDunlink and no bulk"
 
 test_115b() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
 	#define OBD_FAIL_OST_ENOSPC              0x215
@@ -2362,8 +2382,8 @@ test_115b() {
 run_test 115b "write: late REQ MDunlink and no bulk"
 
 test_115c() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
 	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
@@ -2372,8 +2392,8 @@ test_115c() {
 run_test 115c "read: late Reply MDunlink and no bulk"
 
 test_115d() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
 	#define OBD_FAIL_OST_ENOSPC              0x215
@@ -2382,8 +2402,8 @@ test_115d() {
 run_test 115d "write: late Reply MDunlink and no bulk"
 
 test_115e() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
 	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
@@ -2392,8 +2412,8 @@ test_115e() {
 run_test 115e "read: late Bulk MDunlink and no reply"
 
 test_115f() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
 	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
@@ -2402,8 +2422,8 @@ test_115f() {
 run_test 115f "read: late REQ MDunlink and no reply"
 
 test_115g() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
 
 	#define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
 	test_115_read 0x8000051c 0
@@ -2576,10 +2596,9 @@ test_130_base() {
 }
 
 test_130a() {
-	remote_mds_nodsh && skip "remote MDS with nodsh" && return
-	local server_version=$(lustre_version_code $SINGLEMDS)
-	[[ $server_version -ge $(version_code 2.7.2) ]] ||
-		{ skip "Need server version newer than 2.7.1"; return 0; }
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+		skip "Need server version newer than 2.7.1"
 
 	test_130_base
 
@@ -2589,10 +2608,9 @@ test_130a() {
 run_test 130a "enqueue resend on not existing file"
 
 test_130b() {
-	remote_mds_nodsh && skip "remote MDS with nodsh" && return
-	local server_version=$(lustre_version_code $SINGLEMDS)
-	[[ $server_version -ge $(version_code 2.7.2) ]] ||
-		{ skip "Need server version newer than 2.7.1"; return 0; }
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+		skip "Need server version newer than 2.7.1"
 
 	test_130_base
 	# let the reply to be dropped
@@ -2652,7 +2670,7 @@ test_132() {
 
 	rm -f $DIR/$tfile
 	# get a lock on client so that export would reach the stale list
-	$SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+	$LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
 	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
 		error "dd failed"
 
@@ -2684,7 +2702,7 @@ test_131() {
 
 	rm -f $DIR/$tfile
 	# get a lock on client so that export would reach the stale list
-	$SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+	$LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
 	dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
 
 	# another IO under the same lock
@@ -2759,7 +2777,7 @@ test_134() {
 run_test 134 "race between failover and search for reply data free slot"
 
 test_135() {
-	[ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
+	[ "$MDS1_VERSION" -lt $(version_code 2.12.51) ] &&
 		skip "Need MDS version at least 2.12.51"
 
 	mkdir -p $DIR/$tdir
@@ -2773,6 +2791,251 @@ test_135() {
 }
 run_test 135 "DOM: open/create resend to return size"
 
+test_136() {
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.12.52) ]] ||
+		skip "Need MDS version at least 2.12.52"
+
+	local mdts=$(comma_list $(mdts_nodes))
+	local MDT0=$(facet_svc $SINGLEMDS)
+
+	local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
+	[ -n "$clog" ] || error "changelog_register failed"
+	cl_mask=$(do_facet mds1 $LCTL get_param \
+				mdd.$MDT0.changelog_mask -n)
+	changelog_chmask "ALL"
+
+	# generate some changelog records to accumulate
+	test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
+	createmany -m $DIR/$tdir/$tfile 10000 ||
+		error "create $DIR/$tdir/$tfile failed"
+
+	local size1=$(do_facet $SINGLEMDS \
+		      $LCTL get_param -n mdd.$MDT0.changelog_size)
+	echo "Changelog size $size1"
+
+	#define OBD_FAIL_LLOG_PURGE_DELAY		    0x1318
+	do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
+
+	# launch changelog_deregister in background on MDS
+	do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
+			> foo.out 2> foo.err < /dev/null &"
+	# give time to reach fail_loc
+	sleep 15
+
+	# fail_loc will make MDS sleep in the middle of changelog_deregister
+	# take this opportunity to abruptly kill MDS
+	FAILURE_MODE_save=$FAILURE_MODE
+	FAILURE_MODE=HARD
+	fail mds1
+	FAILURE_MODE=$FAILURE_MODE_save
+
+	do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
+
+	local size2=$(do_facet $SINGLEMDS \
+		      $LCTL get_param -n mdd.$MDT0.changelog_size)
+	echo "Changelog size $size2"
+	local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+			mdd.$MDT0.changelog_users | grep $clog")
+	echo "After crash, changelog user $clog2"
+
+	[ -n "$clog2" -o $size2 -lt $size1 ] ||
+		error "changelog record count unchanged"
+
+	do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
+}
+run_test 136 "changelog_deregister leaving pending records"
+
+test_137() {
+	df $DIR
+	mkdir -p $DIR/d1
+	mkdir -p $DIR/d2
+	dd if=/dev/zero of=$DIR/d1/$tfile bs=4096 count=1
+	dd if=/dev/zero of=$DIR/d2/$tfile bs=4096 count=1
+	cancel_lru_locks osc
+
+	#define OBD_FAIL_PTLRPC_RESEND_RACE	 0x525
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000525"
+
+	# RPC1: any reply is to be delayed to disable last_xid logic
+	ln $DIR/d1/$tfile $DIR/d1/f2 &
+	sleep 1
+
+	# RPC2: setattr1 reply is delayed & resent
+	# original reply comes to client; the resend get asleep
+	chmod 666 $DIR/d2/$tfile
+
+	# RPC3: setattr2 on the same file; run ahead of RPC2 resend
+	chmod 777 $DIR/d2/$tfile
+
+	# RPC2 resend wakes up
+	sleep 5
+	[ $(stat -c "%a" $DIR/d2/$tfile) == 777 ] || error "resend got applied"
+}
+run_test 137 "late resend must be skipped if already applied"
+
+test_138() {
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ "$MDS1_VERSION" -ge $(version_code 2.12.59) ]] ||
+		skip "Need server version newer than 2.12.59"
+
+	zconf_umount_clients $CLIENTS $MOUNT
+
+#define OBD_FAIL_TGT_RECOVERY_CONNECT 0x724
+	#delay a first step of recovey when MDS waiting clients
+	#and failing to get osp logs
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x724 fail_val=5
+
+	facet_failover $SINGLEMDS
+
+	#waiting failover and recovery timer
+	#the valuse is based on target_recovery_overseer() wait_event timeout
+	sleep 55
+	stop $SINGLEMDS || error "stop MDS failed"
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
+		error "start MDS failed"
+	zconf_mount_clients $CLIENTS $MOUNT
+}
+run_test 138 "Umount MDT during recovery"
+
+test_139() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $MDS1_VERSION -lt $(version_code 2.13.50) ] &&
+		skip "Need MDS version at least 2.13.50"
+
+	mdt_dev=$(mdsdevname 1)
+
+	stop $SINGLEMDS || error "stop $SINGLEMDS failed"
+
+#define OBD_FAIL_OSP_INVALID_LOGID		0x2106
+	do_facet $SINGLEMDS $LCTL set_param fail_val=0x68 fail_loc=0x80002106
+	start $SINGLEMDS $mdt_dev $MDS_MOUNT_OPTS || error "Fail to start MDT"
+}
+run_test 139 "corrupted catid won't cause crash"
+
+test_140a() {
+	[ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
+		skip "Need MDS version at least 2.13.50"
+
+	[ "$SHARED_KEY" = true ] &&
+		skip "server local client incompatible with SSK keys installed"
+
+	slr=$(do_facet mds1 \
+		$LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
+	stack_trap "do_facet mds1 $LCTL set_param \
+		mdt.*.local_recovery=$slr" EXIT
+
+	# disable recovery for local clients
+	# so local clients should be marked with no_recovery flag
+	do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
+	mount_mds_client
+
+	local cnt
+	cnt=$(do_facet mds1 $LCTL get_param "mdt.*.exports.*.export" |
+		grep export_flags.*no_recovery | wc -l)
+	echo "$cnt clients with recovery disabled"
+	umount_mds_client
+	[ $cnt -eq 0 ] && error "no clients with recovery disabled"
+
+	# enable recovery for local clients
+	# so no local clients should be marked with no_recovery flag
+	do_facet mds1 $LCTL set_param mdt.*.local_recovery=1
+	mount_mds_client
+
+	cnt=$(do_facet mds1 $LCTL get_param "mdt.*.exports.*.export" |
+		grep export_flags.*no_recovery | wc -l)
+	echo "$cnt clients with recovery disabled"
+	umount_mds_client
+	[ $cnt -eq 0 ] || error "$cnt clients with recovery disabled"
+}
+run_test 140a "local mount is flagged properly"
+
+test_140b() {
+	[ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
+		skip "Need MDS version at least 2.13.50"
+
+	[ "$SHARED_KEY" = true ] &&
+		skip "server local client incompatible with SSK keys installed"
+
+	slr=$(do_facet mds1 \
+		$LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
+	stack_trap "do_facet mds1 $LCTL set_param \
+		mdt.*.local_recovery=$slr" EXIT
+
+	# disable recovery for local clients
+	do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
+
+	mount_mds_client
+	replay_barrier mds1
+	umount_mds_client
+	fail mds1
+	local recovery=$(do_facet mds1 dmesg |
+			 awk -F: '/Recovery over after/ { print $4 }' |
+			 cut -d, -f1 | tail -1)
+	(( $recovery < $TIMEOUT*2 )) ||
+		error "recovery took too long $recovery > $((TIMEOUT * 2))"
+}
+run_test 140b "local mount is excluded from recovery"
+
+test_141() {
+	local oldc
+	local newc
+
+	[ $PARALLEL == "yes" ] && skip "skip parallel run"
+	combined_mgs_mds || skip "needs combined MGS/MDT"
+	( local_mode || from_build_tree ) &&
+		skip "cannot run in local mode or from build tree"
+
+	# some get_param have a bug to handle dot in param name
+	do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
+	oldc=$(do_facet $SINGLEMDS $LCTL get_param -n \
+		'ldlm.namespaces.MGC*.lock_count')
+	fail $SINGLEMDS
+	do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
+	newc=$(do_facet $SINGLEMDS $LCTL get_param -n \
+		'ldlm.namespaces.MGC*.lock_count')
+
+	[ $oldc -eq $newc ] || error "mgc lost locks ($oldc != $newc)"
+	return 0
+}
+run_test 141 "do not lose locks on MGS restart"
+
+test_142() {
+	[ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
+		skip "Need MDS version at least 2.11.56"
+
+	#define OBD_FAIL_MDS_ORPHAN_DELETE	0x165
+	do_facet mds1 $LCTL set_param fail_loc=0x165
+	$MULTIOP $DIR/$tfile Ouc || error "multiop failed"
+
+	stop mds1
+	start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
+
+	wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
+		error "MDD orphan cleanup thread not quit"
+}
+run_test 142 "orphan name stub can be cleaned up in startup"
+
+test_143() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.13.00) ] &&
+		skip "Need MDS version at least 2.13.00"
+	[ $PARALLEL == "yes" ] && skip "skip parallel run"
+
+	local mntpt=$(facet_mntpt $SINGLEMDS)
+	stop mds1
+	mount_fstype $SINGLEMDS || error "mount as fstype $SINGLEMDS failed"
+	do_facet $SINGLEMDS touch $mntpt/PENDING/$tfile
+	unmount_fstype $SINGLEMDS
+	start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
+
+	wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
+	wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
+		error "MDD orphan cleanup thread not quit"
+}
+run_test 143 "orphan cleanup thread shouldn't be blocked even delete failed"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status