X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=4ef21eb60461cb05e31635d8acf7aee9361510cd;hb=3f5fc6412d3ae2d3a57f8fdb8a457f35e9d9576e;hp=37a66ce30d69191ee2ce836a51beb190e18ea9fe;hpb=57118830eb55ab43b4b6fc096ad40b2bd83c6de3;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index 37a66ce..4ef21eb 100644
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,12 +11,11 @@ SRCDIR=$(dirname $0)
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test:
+# bug number for skipped test:    3815
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
+# bug number for skipped test:4178         4176
+ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# skip test cases failed before landing - Jinshan
-
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 31a 34 35 36"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 201 221 223a 223b 225"
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 
@@ -31,9 +30,9 @@ MCREATE=${MCREATE:-mcreate}
 MOUNT_2=${MOUNT_2:-"yes"}
 FAIL_ON_ERROR=false
 
-if [[ $MDSCOUNT -ge 2 ]]; then
-	skip_env "Only run with single MDT for now" && exit
-fi
+# script only handles up to 10 MDTs (because of MDT_PREFIX)
+[ $MDSCOUNT -gt 9 ] &&
+	error "script cannot handle more than 9 MDTs, please fix" && exit
 
 check_and_setup_lustre
 
@@ -95,8 +94,9 @@ init_agt_vars() {
 	HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
 	HSM_ARCHIVE_NUMBER=2
 
-	MDT_PARAM="mdt.$FSNAME-MDT0000"
-	HSM_PARAM="$MDT_PARAM.hsm"
+	# The test only support up to 10 MDTs
+	MDT_PREFIX="mdt.$FSNAME-MDT000"
+	HSM_PARAM="${MDT_PREFIX}0.hsm"
 
 	# archive is purged at copytool setup
 	HSM_ARCHIVE_PURGE=true
@@ -117,6 +117,22 @@ cleanup() {
 	cdt_set_sanity_policy
 }
 
+get_mdt_devices() {
+	local mdtno
+	# get MDT device for each mdc
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		MDT[$idx]=$($LCTL get_param -n \
+			mdc.$FSNAME-MDT000${idx}-mdc-*.mds_server_uuid |
+			awk '{gsub(/_UUID/,""); print $1}' | head -1)
+	done
+}
+
+search_copytools() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+}
+
 search_and_kill_copytool() {
 	local agents=${1:-$(facet_active_host $SINGLEAGT)}
 
@@ -128,7 +144,7 @@ copytool_setup() {
 	local facet=${1:-$SINGLEAGT}
 	local lustre_mntpnt=${2:-$MOUNT}
 	local arc_id=$3
-	local hsm_root=$(copytool_device $facet)
+	local hsm_root=${4:-$(copytool_device $facet)}
 	local agent=$(facet_active_host $facet)
 
 	if [[ -z "$arc_id" ]] &&
@@ -164,10 +180,34 @@ copytool_setup() {
 copytool_cleanup() {
 	trap - EXIT
 	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	local mdtno
+	local idx
+	local oldstate
+	local mdt_hsmctrl
 
 	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
 	sleep 1
 	echo "Copytool is stopped on $agents"
+
+	# clean all CDTs orphans requests from previous tests
+	# that would otherwise need to timeout to clear.
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		idx=$(($mdtno - 1))
+		mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
+		oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
+				   "$mdt_hsmctrl")
+		# skip already stop[ed,ing] CDTs
+		echo $oldstate | grep stop && continue
+
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"stopped" 20 ||
+			error "mds${mdtno} cdt state is not stopped"
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"$oldstate" 20 ||
+			error "mds${mdtno} cdt state is not $oldstate"
+	done
 }
 
 copytool_suspend() {
@@ -179,7 +219,7 @@ copytool_suspend() {
 
 copytool_remove_backend() {
 	local fid=$1
-	local be=$(find $HSM_ARCHIVE -name $fid)
+	local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
 	echo "Remove from backend: $fid = $be"
 	do_facet $SINGLEAGT rm -f $be
 }
@@ -204,20 +244,66 @@ copy2archive() {
 	do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
 }
 
+mdts_set_param() {
+	local arg=$1
+	local key=$2
+	local value=$3
+	local mdtno
+	local rc=0
+	if [[ "$value" != "" ]]; then
+		value="=$value"
+	fi
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		local facet=mds${mdtno}
+		# if $arg include -P option, run 1 set_param per MDT on the MGS
+		# else, run set_param on each MDT
+		[[ $arg = *"-P"* ]] && facet=mgs
+		do_facet $facet $LCTL set_param $arg mdt.${MDT[$idx]}.$key$value
+		[[ $? != 0 ]] && rc=1
+	done
+	return $rc
+}
+
+mdts_check_param() {
+	local key="$1"
+	local target="$2"
+	local timeout="$3"
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		wait_result mds${mdtno} \
+			"$LCTL get_param -n $MDT_PREFIX${idx}.$key" "$target" \
+			$timeout ||
+			error "$key state is not '$target' on mds${mdtno}"
+	done
+}
+
 changelog_setup() {
-	CL_USER=$(do_facet $SINGLEMDS $LCTL --device $MDT0\
-		  changelog_register -n)
-	do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog_mask="+hsm"
-	$LFS changelog_clear $MDT0 $CL_USER 0
+	CL_USERS=()
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		local cl_user=$(do_facet mds${mdtno} $LCTL \
+			     --device ${MDT[$idx]} \
+			     changelog_register -n)
+		CL_USERS+=($cl_user)
+		do_facet mds${mdtno} lctl set_param \
+			mdd.${MDT[$idx]}.changelog_mask="+hsm"
+		$LFS changelog_clear ${MDT[$idx]} $cl_user 0
+	done
 }
 
 changelog_cleanup() {
-#	$LFS changelog $MDT0
-	[[ -n "$CL_USER" ]] || return 0
-
-	$LFS changelog_clear $MDT0 $CL_USER 0
-	do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $CL_USER
-	CL_USER=
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		[[ -z  ${CL_USERS[$idx]} ]] && continue
+		$LFS changelog_clear ${MDT[$idx]} ${CL_USERS[$idx]} 0
+		do_facet mds${mdtno} lctl --device ${MDT[$idx]} \
+			changelog_deregister ${CL_USERS[$idx]}
+	done
+	CL_USERS=()
 }
 
 changelog_get_flags() {
@@ -238,64 +324,57 @@ set_hsm_param() {
 	local param=$1
 	local value=$2
 	local opt=$3
-	if [[ "$value" != "" ]]; then
-		value="=$value"
-	fi
-	do_facet $SINGLEMDS $LCTL set_param $opt -n $HSM_PARAM.$param$value
+	mdts_set_param "$opt -n" "hsm.$param" "$value"
 	return $?
 }
 
 set_test_state() {
 	local cmd=$1
 	local target=$2
-	do_facet $SINGLEMDS $LCTL set_param $MDT_PARAM.hsm_control=$cmd
-	wait_result $SINGLEMDS "$LCTL get_param -n $MDT_PARAM.hsm_control"\
-		$target 10 || error "cdt state is not $target"
+	mdts_set_param "" hsm_control "$cmd"
+	mdts_check_param hsm_control "$target" 10
 }
 
 cdt_set_sanity_policy() {
 	if [[ "$CDT_POLICY_HAD_CHANGED" ]]
 	then
 		# clear all
-		do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NRA
-		do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
+		mdts_set_param "" hsm.policy "+NRA"
+		mdts_set_param "" hsm.policy "-NBR"
 		CDT_POLICY_HAD_CHANGED=
 	fi
 }
 
 cdt_set_no_retry() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NRA
+	mdts_set_param "" hsm.policy "+NRA"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_no_retry() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NRA
+	mdts_set_param "" hsm.policy "-NRA"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_set_non_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NBR
+	mdts_set_param "" hsm.policy "+NBR"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_non_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
+	mdts_set_param "" hsm.policy "-NBR"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_mount_state() {
-	do_facet $SINGLEMDS $LCTL set_param -d -P $MDT_PARAM.hsm_control
+	mdts_set_param "-P -d" hsm_control ""
 }
 
 cdt_set_mount_state() {
-	do_facet $SINGLEMDS $LCTL set_param -P $MDT_PARAM.hsm_control=$1
+	mdts_set_param "-P" hsm_control "$1"
 }
 
 cdt_check_state() {
-	local target=$1
-	wait_result $SINGLEMDS\
-		"$LCTL get_param -n $MDT_PARAM.hsm_control" "$target" 20 ||
-			error "cdt state is not $target"
+	mdts_check_param hsm_control "$1" 20
 }
 
 cdt_disable() {
@@ -400,8 +479,15 @@ make_small() {
         path2fid $1 || error "cannot get fid on $1"
 }
 
+make_small_sync() {
+	dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
+		error "cannot create $1"
+	path2fid $1 || error "cannot get fid on $1"
+}
+
 cleanup_large_files() {
-	local ratio=$(df $MOUNT |awk '{print $5}' |sed 's/%//g' |grep -v Use)
+	local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
+		      sed 's/%//g')
 	[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
 }
 
@@ -465,19 +551,22 @@ wait_request_state() {
 	local fid=$1
 	local request=$2
 	local state=$3
+	# 4th arg (mdt index) is optional
+	local mdtidx=${4:-0}
+	local mds=mds$(($mdtidx + 1))
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
 	cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 
-	wait_result $SINGLEMDS "$cmd" $state 100 ||
-		error "request on $fid is not $state"
+	wait_result $mds "$cmd" $state 100 ||
+		error "request on $fid is not $state on $mds"
 }
 
 get_request_state() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 }
 
@@ -485,14 +574,14 @@ get_request_count() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
 wait_all_done() {
 	local timeout=$1
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
 	cmd+=" | egrep 'WAITING|STARTED'"
 
 	wait_result $SINGLEMDS "$cmd" "" $timeout ||
@@ -504,8 +593,8 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-	awk '{gsub(/_UUID/,""); print $1}' | head -1)
+# populate MDT device array
+get_mdt_devices
 
 # initiate variables
 init_agt_vars
@@ -1066,6 +1155,53 @@ test_12n() {
 }
 run_test 12n "Import/implicit restore/release"
 
+test_12o() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/hosts $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "release of $f failed"
+
+#define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS		0x152
+	do_facet $SINGLEMDS lctl set_param fail_loc=0x152
+
+	# set no retry action mode
+	cdt_set_no_retry
+
+	diff -q /etc/hosts $f
+	local st=$?
+
+	# we check we had a restore failure
+	wait_request_state $fid RESTORE FAILED
+
+	[[ $st -eq 0 ]] && error "Restore must fail"
+
+	# remove no retry action mode
+	cdt_clear_no_retry
+
+	# check file is still released
+	check_hsm_flags $f "0x0000000d"
+
+	# retry w/o failure injection
+	do_facet $SINGLEMDS lctl set_param fail_loc=0
+
+	diff -q /etc/hosts $f
+	st=$?
+
+	# we check we had a restore done
+	wait_request_state $fid RESTORE SUCCEED
+
+	[[ $st -eq 0 ]] || error "Restored file differs"
+
+	copytool_cleanup
+}
+run_test 12o "Layout-swap failure during Restore leaves file released"
+
 test_13() {
 	# test needs a running copytool
 	copytool_setup
@@ -1082,8 +1218,8 @@ test_13() {
 			CURR_FILE="$CURR_DIR/$tfile.$f"
 			# write file-specific data
 			do_facet $SINGLEAGT \
-				echo "d=$d, f=$f, dir=$CURR_DIR, "\
-				     "file=$CURR_FILE" > $CURR_FILE
+				"echo d=$d, f=$f, dir=$CURR_DIR, "\
+					"file=$CURR_FILE > $CURR_FILE"
 		done
 	done
 	# import to Lustre
@@ -1260,19 +1396,41 @@ test_21() {
 	local fid=$(make_small $f)
 	check_hsm_flags $f "0x00000000"
 
+	# LU-4388/LU-4389 - ZFS does not report full number of blocks
+	# used until file is flushed to disk
+	if [  $(facet_fstype ost1) == "zfs" ]; then
+	    # this causes an OST_SYNC rpc to be sent
+	    dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
+	    # clear locks to reread file data
+	    cancel_lru_locks osc
+	fi
+
+	local orig_size=$(stat -c "%s" $f)
+	local orig_blocks=$(stat -c "%b" $f)
+
+	start_full_debug_logging
+
 	$LFS hsm_archive $f || error "could not archive file"
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	[ $(stat -c "%b" $f) -ne "1" ] || error "wrong block number"
-	local sz=$(stat -c "%s" $f)
-	[ $sz -ne "0" ] || error "file size should not be zero"
+	local blocks=$(stat -c "%b" $f)
+	[ $blocks -eq $orig_blocks ] ||
+		error "$f: wrong block number after archive: " \
+		      "$blocks != $orig_blocks"
+	local size=$(stat -c "%s" $f)
+	[ $size -eq $orig_size ] ||
+		error "$f: wrong size after archive: $size != $orig_size"
 
 	# Release and check states
 	$LFS hsm_release $f || error "could not release file"
 	check_hsm_flags $f "0x0000000d"
 
-	[ $(stat -c "%b" $f) -eq "1" ] || error "wrong block number"
-	[ $(stat -c "%s" $f) -eq $sz ] || error "wrong file size"
+	blocks=$(stat -c "%b" $f)
+	[ $blocks -gt 5 ] &&
+		error "$f: too many blocks after release: $blocks > 5"
+	size=$(stat -c "%s" $f)
+	[ $size -ne $orig_size ] &&
+		error "$f: wrong size after release: $size != $orig_size"
 
 	# Check we can release an file without stripe info
 	f=$f.nolov
@@ -1290,6 +1448,8 @@ test_21() {
 	$LFS hsm_release $f || fail "second release should succeed"
 	check_hsm_flags $f "0x0000000d"
 
+	stop_full_debug_logging
+
 	copytool_cleanup
 }
 run_test 21 "Simple release tests"
@@ -1827,6 +1987,41 @@ test_30b() {
 }
 run_test 30b "Restore at exec (release case)"
 
+test_30c() {
+	needclients 2 || return 0
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/SLEEP
+	local fid=$(copy_file /bin/sleep $f)
+	chmod 755 $f
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+	check_hsm_flags $f "0x0000000d"
+	# set no retry action mode
+	cdt_set_no_retry
+	do_node $CLIENT2 "$f 10" &
+	local pid=$!
+	sleep 3
+	echo 'Hi!' > $f
+	[[ $? == 0 ]] && error "Update during exec of released file must fail"
+	wait $pid
+	[[ $? == 0 ]] || error "Execution failed during run"
+	cmp /bin/sleep $f
+	[[ $? == 0 ]] || error "Binary overwritten during exec"
+
+	# cleanup
+	# remove no try action mode
+	cdt_clear_no_retry
+	check_hsm_flags $f "0x00000009"
+
+	copytool_cleanup
+}
+run_test 30c "Update during exec of released file must fail"
+
 restore_and_check_size() {
 	local f=$1
 	local fid=$2
@@ -1852,11 +2047,11 @@ restore_and_check_size() {
 		cpt=$((cpt + 1))
 	done
 	if [[ $cpt -lt 10 ]]; then
-		echo " restore is too long"
-	else
 		echo " "done
+	else
+		echo " restore is too long"
+		wait_request_state $fid RESTORE SUCCEED
 	fi
-	wait_request_state $fid RESTORE SUCCEED
 	return $err
 }
 
@@ -2086,7 +2281,13 @@ test_40() {
 			fid=$(copy_file /etc/hosts $f.$p.$i)
 		done
 	done
-	copytool_setup
+	# force copytool to use a local/temp archive dir to ensure best
+	# performance vs remote/NFS mounts used in auto-tests
+	if df --local $HSM_ARCHIVE >/dev/null 2>&1 ; then
+		copytool_setup
+	else
+		copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir
+	fi
 	# to be sure wait_all_done will not be mislead by previous tests
 	cdt_purge
 	wait_for_grace_delay
@@ -2382,7 +2583,7 @@ double_verify_reset_hsm_param() {
 test_100() {
 	double_verify_reset_hsm_param loop_period
 	double_verify_reset_hsm_param grace_delay
-	double_verify_reset_hsm_param request_timeout
+	double_verify_reset_hsm_param active_request_timeout
 	double_verify_reset_hsm_param max_requests
 	double_verify_reset_hsm_param default_archive_id
 }
@@ -2412,7 +2613,7 @@ test_103() {
 
 	echo "Current requests"
 	local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
 
 	[[ -z "$res" ]] || error "Some request have not been canceled"
@@ -2434,7 +2635,7 @@ test_104() {
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
 	local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep $fid | cut -f16 -d=")
 	cdt_enable
 
@@ -2455,12 +2656,12 @@ test_105() {
 		$LFS hsm_archive $DIR/$tdir/$i
 	done
 	local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_restart
 	cdt_disable
 	local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_enable
 	cdt_purge
@@ -2470,28 +2671,70 @@ test_105() {
 }
 run_test 105 "Restart of coordinator"
 
-test_106() {
-	# test needs a running copytool
-	copytool_setup
+get_agent_by_uuid_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
+		 grep $uuid"
+}
+
+check_agent_registered_by_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+	if [[ ! -z "$agent" ]]; then
+		echo "found agent $agent on $mds"
+	else
+		error "uuid $uuid not found in agent list on $mds"
+	fi
+}
 
+check_agent_unregistered_by_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+	if [[ -z "$agent" ]]; then
+		echo "uuid not found in agent list on $mds"
+	else
+		error "uuid found in agent list on $mds: $agent"
+	fi
+}
+
+check_agent_registered() {
+	local uuid=$1
+	local mdsno
+	for mdsno in $(seq 1 $MDSCOUNT); do
+		check_agent_registered_by_mdt $uuid $((mdsno - 1))
+	done
+}
+
+check_agent_unregistered() {
+	local uuid=$1
+	local mdsno
+	for mdsno in $(seq 1 $MDSCOUNT); do
+		check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
+	done
+}
+
+test_106() {
 	local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
 		get_client_uuid $MOUNT | cut -d' ' -f2)
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
+
+	copytool_setup
+	check_agent_registered $uuid
+
+	search_copytools || error "No copytool found"
+
 	copytool_cleanup
-	[[ ! -z "$agent" ]] || error "My uuid $uuid not found in agent list"
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
-	[[ -z "$agent" ]] ||
-		error "My uuid $uuid still found in agent list,"\
-		      " after copytool shutdown"
+	check_agent_unregistered $uuid
+
 	copytool_setup
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
+	check_agent_registered $uuid
+
 	copytool_cleanup
-	[[ ! -z "$agent" ]] ||
-		error "My uuid $uuid not found in agent list after"\
-		      " copytool restart"
 }
 run_test 106 "Copytool register/unregister"
 
@@ -2767,7 +3010,7 @@ test_220() {
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 	changelog_cleanup
 
 	local target=0x0
@@ -2794,7 +3037,7 @@ test_221() {
 	wait_request_state $fid ARCHIVE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x7d
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
@@ -2819,7 +3062,7 @@ test_222a() {
 	$LFS hsm_restore $f
 	wait_request_state $fid RESTORE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x80
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
@@ -2845,7 +3088,7 @@ test_222b() {
 
 	wait_request_state $fid RESTORE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x80
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
@@ -2874,7 +3117,7 @@ test_223a() {
 	wait_request_state $fid RESTORE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0xfd
 	[[ $flags == $target ]] ||
@@ -2903,7 +3146,7 @@ test_223b() {
 	wait_request_state $fid RESTORE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0xfd
 	[[ $flags == $target ]] ||
@@ -2929,7 +3172,7 @@ test_224() {
 	$LFS hsm_remove $f
 	wait_request_state $fid REMOVE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
 
 	local target=0x200
 	[[ $flags == $target ]] ||
@@ -2965,9 +3208,9 @@ test_225() {
 	wait_request_state $fid REMOVE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	flags=$(changelog_get_flags $MDT0 RENME $fid2)
-	local flags=$($LFS changelog $MDT0 | grep HSM | grep $fid | tail -1 |
-		awk '{print $5}')
+	flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
+	local flags=$($LFS changelog ${MDT[0]} | grep HSM | grep $fid |
+		tail -n 1 | awk '{print $5}')
 
 	local target=0x27d
 	[[ $flags == $target ]] ||
@@ -2999,7 +3242,7 @@ test_226() {
 
 	rm $f1 || error "rm $f1 failed"
 
-	local flags=$(changelog_get_flags $MDT0 UNLNK $fid1)
+	local flags=$(changelog_get_flags ${MDT[0]} UNLNK $fid1)
 
 	local target=0x3
 	[[ $flags == $target ]] ||
@@ -3007,7 +3250,7 @@ test_226() {
 
 	mv $f3 $f2 || error "mv $f3 $f2 failed"
 
-	flags=$(changelog_get_flags $MDT0 RENME $fid2)
+	flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
 
 	target=0x3
 	[[ $flags == $target ]] ||
@@ -3027,7 +3270,7 @@ check_flags_changes() {
 	local target=0x280
 	$LFS hsm_set --$hsm_flag $f ||
 		error "Cannot set $hsm_flag on $f"
-	local flags=($(changelog_get_flags $MDT0 HSM $fid))
+	local flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
 	local seen=${#flags[*]}
 	cnt=$((fst + cnt))
 	[[ $seen == $cnt ]] ||
@@ -3038,7 +3281,7 @@ check_flags_changes() {
 
 	$LFS hsm_clear --$hsm_flag $f ||
 		error "Cannot clear $hsm_flag on $f"
-	flags=($(changelog_get_flags $MDT0 HSM $fid))
+	flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
 	seen=${#flags[*]}
 	cnt=$(($cnt + 1))
 	[[ $cnt == $seen ]] ||
@@ -3078,10 +3321,9 @@ test_228() {
 	# test needs a running copytool
 	copytool_setup
 
-	dd if=/dev/urandom of=$DIR/$tfile bs=1M count=1 conv=sync ||
-		error "creating $DIR/$tfile"
+	local fid=$(make_small_sync $DIR/$tfile)
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
-	wait_request_state $(path2fid $DIR/$tfile) ARCHIVE SUCCEED
+	wait_request_state $fid ARCHIVE SUCCEED
 
 	$LFS hsm_release $DIR/$tfile
 	check_hsm_flags $DIR/$tfile "0x0000000d"
@@ -3098,13 +3340,15 @@ test_228() {
 	$LFS hsm_release $DIR/$tfile
 	check_hsm_flags $DIR/$tfile "0x0000000d"
 
-	mkdir $DIR/$tdir
+	mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
 
 	tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
 		error "tar failed"
 	cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
 		error "comparing untarred $DIR/$tfile"
 
+	rm -f $DIR/$tfile $DIR/$tfile.2 ||
+		error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
 	copytool_cleanup
 }
 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
@@ -3133,12 +3377,12 @@ test_250() {
 	while [[ $cnt != 0 || $wt != 0 ]]; do
 		sleep 1
 		cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep STARTED | grep -v CANCEL | wc -l")
 		[[ $cnt -le $maxrequest ]] ||
 			error "$cnt > $maxrequest too many started requests"
 		wt=$(do_facet $SINGLEMDS "$LCTL get_param\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 		echo "max=$maxrequest started=$cnt waiting=$wt"
 	done
@@ -3157,8 +3401,8 @@ test_251() {
 
 	cdt_disable
 	# to have a short test
-	local old_to=$(get_hsm_param request_timeout)
-	set_hsm_param request_timeout 4
+	local old_to=$(get_hsm_param active_request_timeout)
+	set_hsm_param active_request_timeout 4
 	# to be sure the cdt will wake up frequently so
 	# it will be able to cancel the "old" request
 	local old_loop=$(get_hsm_param loop_period)
@@ -3170,7 +3414,7 @@ test_251() {
 	sleep 5
 	wait_request_state $fid ARCHIVE CANCELED
 
-	set_hsm_param request_timeout $old_to
+	set_hsm_param active_request_timeout $old_to
 	set_hsm_param loop_period $old_loop
 
 	copytool_cleanup
@@ -3227,7 +3471,11 @@ test_302() {
 	cdt_shutdown
 
 	set_hsm_param default_archive_id $new -P
-	fail $SINGLEMDS
+
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		fail mds${mdtno}
+	done
 
 	# check cdt is on
 	cdt_check_state enabled
@@ -3241,6 +3489,168 @@ test_302() {
 }
 run_test 302 "HSM tunnable are persistent when CDT is off"
 
+test_400() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	local dir_mdt1=$DIR/$tdir/mdt1
+
+	# create 1 dir per MDT
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+	$LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
+
+	# create 1 file in each MDT
+	local fid1=$(make_small $dir_mdt0/$tfile)
+	local fid2=$(make_small $dir_mdt1/$tfile)
+
+	# check that hsm request on mdt0 is sent to the right MDS
+	$LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+
+	# check that hsm request on mdt1 is sent to the right MDS
+	$LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
+	wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
+		echo "archive successful on mdt1"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0 $dir_mdt1
+}
+run_test 400 "Single request is sent to the right MDT"
+
+test_401() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	local dir_mdt1=$DIR/$tdir/mdt1
+
+	# create 1 dir per MDT
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+	$LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
+
+	# create 1 file in each MDT
+	local fid1=$(make_small $dir_mdt0/$tfile)
+	local fid2=$(make_small $dir_mdt1/$tfile)
+
+	# check that compound requests are shunt to the rights MDTs
+	$LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
+		error "lfs hsm_archive"
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+	wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
+		echo "archive successful on mdt1"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0 $dir_mdt1
+}
+run_test 401 "Compound requests split and sent to their respective MDTs"
+
+mdc_change_state() # facet, MDT_pattern, activate|deactivate
+{
+	local facet=$1
+	local pattern="$2"
+	local state=$3
+	local node=$(facet_active_host $facet)
+	local mdc
+	for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
+			awk '{print $4}'); do
+		echo "$3 $mdc on $node"
+		do_facet $facet "$LCTL --device $mdc $state" || return 1
+	done
+}
+
+test_402() {
+	# make sure there is no running copytool
+	copytool_cleanup
+
+	# deactivate all mdc on agent1
+	mdc_change_state $SINGLEAGT "MDT000." "deactivate"
+
+	copytool_setup $SINGLEAGT
+
+	check_agent_unregistered "uuid" # match any agent
+
+	# no expected running copytool
+	search_copytools $agent && error "Copytool start should have failed"
+
+	# reactivate MDCs
+	mdc_change_state $SINGLEAGT "MDT000." "activate"
+}
+run_test 402 "Copytool start fails if all MDTs are inactive"
+
+test_403() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	# make sure there is no running copytool
+	copytool_cleanup
+
+        local agent=$(facet_active_host $SINGLEAGT)
+	local uuid=$(do_rpc_nodes $agent get_client_uuid | cut -d' ' -f2)
+
+	# deactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "deactivate"
+
+	copytool_setup
+	# check the agent is registered on MDT0000, and not on MDT0001
+	check_agent_registered_by_mdt $uuid 0
+	check_agent_unregistered_by_mdt $uuid 1
+
+	# check running copytool process
+	search_copytools $agent || error "No running copytools on $agent"
+
+	# reactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "activate"
+
+	# make sure the copytool is now registered to all MDTs
+	check_agent_registered $uuid
+
+	copytool_cleanup
+}
+run_test 403 "Copytool starts with inactive MDT and register on reconnect"
+
+test_404() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	# create files on both MDT0000 and MDT0001
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+
+	# create 1 file on mdt0
+	local fid1=$(make_small $dir_mdt0/$tfile)
+
+	# deactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "deactivate"
+
+	# send an HSM request for files in MDT0000
+	$LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
+
+	# check for completion of files in MDT0000
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+
+	# reactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "activate"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0
+}
+run_test 404 "Inactive MDT does not block requests for active MDTs"
+
 copytool_cleanup
 
 complete $SECONDS