X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=0ee67b103990e87f6b8eafec33406a983b066a4d;hp=37a66ce30d69191ee2ce836a51beb190e18ea9fe;hb=f625f670afbe954030ff81f0f8522137d6cdd335;hpb=57118830eb55ab43b4b6fc096ad40b2bd83c6de3

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
old mode 100644
new mode 100755
index 37a66ce..0ee67b1
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,12 +11,9 @@ 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:    LU-3815
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# skip test cases failed before landing - Jinshan
-
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 31a 34 35 36"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 201 221 223a 223b 225"
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 
@@ -27,13 +24,12 @@ init_logging
 
 MULTIOP=${MULTIOP:-multiop}
 OPENFILE=${OPENFILE:-openfile}
-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
 
@@ -91,15 +87,22 @@ init_agt_vars() {
 
 	export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
 	export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
+	export HSMTOOL_UPDATE_INTERVAL=${HSMTOOL_UPDATE_INTERVAL:=""}
+	export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
+	export HSMTOOL_TESTDIR
 	export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
 	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
+
+	# Don't allow copytool error upon start/setup
+	HSMTOOL_NOERROR=false
 }
 
 # Get the backend root path for the given agent facet.
@@ -112,11 +115,28 @@ copytool_device() {
 
 # Stop copytool and unregister an existing changelog user.
 cleanup() {
+	copytool_monitor_cleanup
 	copytool_cleanup
 	changelog_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 $1}' | head -n1)
+	done
+}
+
+search_copytools() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+}
+
 search_and_kill_copytool() {
 	local agents=${1:-$(facet_active_host $SINGLEAGT)}
 
@@ -124,16 +144,73 @@ search_and_kill_copytool() {
 	do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
 }
 
+copytool_monitor_setup() {
+	local facet=${1:-$SINGLEAGT}
+	local agent=$(facet_active_host $facet)
+
+	local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
+	local test_dir=$(do_node $agent "$cmd") ||
+		error "Failed to create tempdir on $agent"
+	export HSMTOOL_MONITOR_DIR=$test_dir
+
+	# Create the fifo and a monitor (cat dies when copytool dies)
+	do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
+		error "failed to create copytool fifo on $agent"
+	cmd="cat $test_dir/fifo > $test_dir/events &"
+	cmd+=" echo \\\$! > $test_dir/monitor_pid"
+
+	if [[ $PDSH == *Rmrsh* ]]; then
+		# This is required for pdsh -Rmrsh and its handling of remote
+		# shells.
+		# Regular ssh and pdsh -Rssh work fine without this
+		# backgrounded subshell nonsense.
+		(do_node $agent "$cmd") &
+		export HSMTOOL_MONITOR_PDSH=$!
+
+		# Slightly racy, but just making a best-effort to catch obvious
+		# problems.
+		sleep 1
+		ps -p $HSMTOOL_MONITOR_PDSH >&- ||
+			error "Failed to start copytool monitor on $agent"
+	else
+		do_node $agent "$cmd"
+		if [ $? != 0 ]; then
+			error "Failed to start copytool monitor on $agent"
+		fi
+	fi
+}
+
+copytool_monitor_cleanup() {
+	local facet=${1:-$SINGLEAGT}
+	local agent=$(facet_active_host $facet)
+
+	if [ -n "$HSMTOOL_MONITOR_DIR" ]; then
+		# Should die when the copytool dies, but just in case.
+		local cmd="kill \\\$(cat $HSMTOOL_MONITOR_DIR/monitor_pid)"
+		cmd+=" 2>/dev/null || true"
+		do_node $agent "$cmd"
+		do_node $agent "rm -fr $HSMTOOL_MONITOR_DIR"
+		export HSMTOOL_MONITOR_DIR=
+	fi
+
+	# The pdsh should die on its own when the monitor dies. Just
+	# in case, though, try to clean up to avoid any cruft.
+	if [ -n "$HSMTOOL_MONITOR_PDSH" ]; then
+		kill $HSMTOOL_MONITOR_PDSH 2>/dev/null
+		export HSMTOOL_MONITOR_PDSH=
+	fi
+}
+
 copytool_setup() {
 	local facet=${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" ]] &&
 		do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
-			echo "Wakeup copytool $facet on $agent"
+			echo "Only wakeup running copytool $facet on $agent"
 			return 0
 	fi
 
@@ -148,6 +225,10 @@ copytool_setup() {
 	# independent of hardware
 	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
 	[[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
+	[[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
+		cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
+	[[ -z "$HSMTOOL_EVENT_FIFO" ]] ||
+		cmd+=" --event-fifo $HSMTOOL_EVENT_FIFO"
 	cmd+=" --bandwidth 1 $lustre_mntpnt"
 
 	# Redirect the standard output and error to a log file which
@@ -156,18 +237,78 @@ copytool_setup() {
 	[[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
 	local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
 
-	do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1" ||
-		error "start copytool $facet on $agent failed"
+	do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1"
+	if [[ $? !=  0 ]]; then
+		[[ $HSMTOOL_NOERROR == true ]] ||
+			error "start copytool $facet on $agent failed"
+		echo "start copytool $facet on $agent failed"
+	fi
+
 	trap cleanup EXIT
 }
 
+get_copytool_event_log() {
+	local facet=${1:-$SINGLEAGT}
+	local agent=$(facet_active_host $facet)
+
+	[ -z "$HSMTOOL_MONITOR_DIR" ] &&
+		error "Can't get event log: No monitor directory!"
+
+	do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
+		error "Could not collect event log from $agent"
+}
+
 copytool_cleanup() {
 	trap - EXIT
-	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	local facet=$SINGLEAGT
+	local agents=${1:-$(facet_active_host $facet)}
+	local mdtno
+	local idx
+	local oldstate
+	local mdt_hsmctrl
+	local hsm_root=$(copytool_device $facet)
+	local end_wait=$(( SECONDS + TIMEOUT ))
 
 	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
-	sleep 1
-	echo "Copytool is stopped on $agents"
+
+	while (( SECONDS < end_wait )); do
+		sleep 2
+		do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+		if [ $? -ne 0 ]; then
+			echo "Copytool is stopped on $agents"
+			break
+		fi
+		echo "Copytool still running on $agents"
+	done
+	if do_nodesv $agents "pgrep -x $HSMTOOL_BASE"; then
+		error "Copytool failed to stop in ${TIMEOUT}s ..."
+	else
+		echo "Copytool has stopped in " \
+		     "$((TIMEOUT - (end_wait - SECONDS)))s."
+	fi
+
+	# 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
+	if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then
+		do_facet $facet "rm -rf $hsm_root/*"
+	fi
 }
 
 copytool_suspend() {
@@ -179,7 +320,7 @@ 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 $be
 }
@@ -195,7 +336,7 @@ make_archive() {
 	local file=$HSM_ARCHIVE/$1
 	do_facet $SINGLEAGT mkdir -p $(dirname $file)
 	do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
-		error "cannot create $file"
+		file_creation_failure dd $file $?
 }
 
 copy2archive() {
@@ -204,20 +345,66 @@ 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 +425,57 @@ 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() {
@@ -331,17 +511,19 @@ needclients() {
 
 path2fid() {
 	$LFS path2fid $1 | tr -d '[]'
+	return ${PIPESTATUS[0]}
 }
 
 get_hsm_flags() {
 	local f=$1
 	local u=$2
+	local st
 
 	if [[ $u == "user" ]]; then
-		local st=$($RUNAS $LFS hsm_state $f)
+		st=$($RUNAS $LFS hsm_state $f)
 	else
-		local st=$($LFS hsm_state $f)
 		u=root
+		st=$($LFS hsm_state $f)
 	fi
 
 	[[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
@@ -352,7 +534,8 @@ get_hsm_flags() {
 
 get_hsm_archive_id() {
 	local f=$1
-	local st=$($LFS hsm_state $f)
+	local st
+	st=$($LFS hsm_state $f)
 	[[ $? == 0 ]] || error "$LFS hsm_state $f failed"
 
 	local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
@@ -376,6 +559,15 @@ check_hsm_flags_user() {
 	[[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
 }
 
+file_creation_failure() {
+	local cmd=$1
+	local f=$2
+	local err=$3
+
+	df $MOUNT $MOUNT2 >&2
+	error "cannot create $f with $cmd, status=$err"
+}
+
 copy_file() {
 	local f=
 
@@ -389,30 +581,52 @@ copy_file() {
 		f=${f/$DIR/$DIR2}
 	fi
 	rm -f $f
-	cp $1 $f || error "cannot copy $1 to $f"
+	cp $1 $f || file_creation_failure cp $f $?
+
 	path2fid $f || error "cannot get fid on $f"
 }
 
 make_small() {
         local file2=${1/$DIR/$DIR2}
         dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
-		error "cannot create $file2"
+		file_creation_failure dd $file2 $?
+
         path2fid $1 || error "cannot get fid on $1"
 }
 
+make_small_sync() {
+	dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
+		file_creation_failure dd $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 {} \;
 }
 
+check_enough_free_space() {
+	local nb=$1
+	local unit=$2
+	local need=$((nb * unit /1024))
+	local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}')
+	(( $need >= $free )) && return 1
+	return 0
+}
+
 make_large_for_striping() {
 	local file2=${1/$DIR/$DIR2}
-	local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -1)
+	local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
 
 	cleanup_large_files
 
+	check_enough_free_space 5 $sz
+	[ $? != 0 ] && return $?
+
 	dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
-		error "cannot create $file2"
+		file_creation_failure dd $file2 $?
+
 	path2fid $1 || error "cannot get fid on $1"
 }
 
@@ -421,12 +635,16 @@ make_large_for_progress() {
 
 	cleanup_large_files
 
+	check_enough_free_space 39 1000000
+	[ $? != 0 ] && return $?
+
 	# big file is large enough, so copy time is > 30s
 	# so copytool make 1 progress
 	# size is not a multiple of 1M to avoid stripe
 	# aligment
 	dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
-		error "cannot create $file2"
+		file_creation_failure dd $file2 $?
+
 	path2fid $1 || error "cannot get fid on $1"
 }
 
@@ -435,12 +653,15 @@ make_large_for_progress_aligned() {
 
 	cleanup_large_files
 
+	check_enough_free_space 33 1048576
+	[ $? != 0 ] && return $?
+
 	# big file is large enough, so copy time is > 30s
 	# so copytool make 1 progress
 	# size is a multiple of 1M to have stripe
 	# aligment
 	dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
-		error "cannot create $file2"
+		file_creation_failure dd $file2 $?
 	path2fid $1 || error "cannot get fid on $1"
 }
 
@@ -449,9 +670,12 @@ make_large_for_cancel() {
 
 	cleanup_large_files
 
+	check_enough_free_space 103 1048576
+	[ $? != 0 ] && return $?
+
 	# Copy timeout is 100s. 105MB => 105s
 	dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
-		error "cannot create $file2"
+		file_creation_failure dd $file2 $?
 	path2fid $1 || error "cannot get fid on $1"
 }
 
@@ -465,19 +689,22 @@ 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 200 ||
+		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 +712,16 @@ 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 fid=$2
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
+	[[ -n $fid ]] && cmd+=" | grep '$fid'"
 	cmd+=" | egrep 'WAITING|STARTED'"
 
 	wait_result $SINGLEMDS "$cmd" "" $timeout ||
@@ -504,8 +733,20 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-	awk '{gsub(/_UUID/,""); print $1}' | head -1)
+parse_json_event() {
+	local raw_event=$1
+
+	# python2.6 in EL6 includes an internal json module
+	local json_parser='import json; import fileinput;'
+	json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
+	json_parser+='json.loads([line for line in '
+	json_parser+='fileinput.input()][0]).items()])'
+
+	echo $raw_event | python -c "$json_parser"
+}
+
+# populate MDT device array
+get_mdt_devices
 
 # initiate variables
 init_agt_vars
@@ -619,7 +860,7 @@ test_3() {
 		error "user could not change hsm flags"
 	dd if=/etc/passwd of=$f.append bs=1 count=3\
 	   conv=notrunc oflag=append status=noxfer ||
-		error "could not append to test file"
+		file_creation_failure dd $f.append $?
 	check_hsm_flags $f.append "0x00000003"
 
 	# Modify a file sets it dirty
@@ -628,7 +869,7 @@ test_3() {
 		error "user could not change hsm flags"
 	dd if=/dev/zero of=$f.modify bs=1 count=3\
 	   conv=notrunc status=noxfer ||
-		error "could not modify test file"
+		file_creation_failure dd $f.modify $?
 	check_hsm_flags $f.modify "0x00000003"
 
 	# Open O_TRUNC sets dirty
@@ -689,7 +930,7 @@ test_9() {
 
 	copytool_cleanup
 }
-run_test 9 "Use of explict archive number, with dedicated copytool"
+run_test 9 "Use of explicit archive number, with dedicated copytool"
 
 test_9a() {
 	needclients 3 || return 0
@@ -802,7 +1043,7 @@ test_10d() {
 }
 run_test 10d "Archive a file on the default archive id"
 
-test_11() {
+test_11a() {
 	mkdir -p $DIR/$tdir
 	copy2archive /etc/hosts $tdir/$tfile
 	local f=$DIR/$tdir/$tfile
@@ -826,7 +1067,31 @@ test_11() {
 	local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
 		error "fid $fid not in archive $HSM_ARCHIVE"
 }
-run_test 11 "Import a file"
+run_test 11a "Import a file"
+
+test_11b() {
+	# 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 -a $HSM_ARCHIVE_NUMBER $f ||
+		error "hsm_archive failed"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local FILE_HASH=$(md5sum $f)
+	rm -f $f
+
+	import_file $fid $f
+
+	echo "$FILE_HASH" | md5sum -c
+
+	[[ $? -eq 0 ]] || error "Restored file differs"
+
+	copytool_cleanup
+}
+run_test 11b "Import a deleted file using its FID"
 
 test_12a() {
 	# test needs a running copytool
@@ -837,16 +1102,16 @@ test_12a() {
 
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
-	local f=$DIR2/$tdir/$tfile
+	local f2=$DIR2/$tdir/$tfile
 	echo "Verifying released state: "
-	check_hsm_flags $f "0x0000000d"
+	check_hsm_flags $f2 "0x0000000d"
 
-	local fid=$(path2fid $f)
-	$LFS hsm_restore $f
+	local fid=$(path2fid $f2)
+	$LFS hsm_restore $f2
 	wait_request_state $fid RESTORE SUCCEED
 
 	echo "Verifying file state: "
-	check_hsm_flags $f "0x00000009"
+	check_hsm_flags $f2 "0x00000009"
 
 	do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
 
@@ -890,7 +1155,10 @@ test_12c() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	$LFS setstripe -c 2 $f
-	local fid=$(make_large_for_striping $f)
+	local fid
+	fid=$(make_large_for_striping $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	local FILE_CRC=$(md5sum $f)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -1066,6 +1334,77 @@ 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
+
+	# to be sure previous RESTORE result is gone
+	cdt_purge
+	wait_for_grace_delay
+
+	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_12p() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir $DIR/$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
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+	$LFS hsm_release $f || error "cannot release $f"
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+	$LFS hsm_release $f || error "cannot release $f"
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+
+	copytool_cleanup
+}
+run_test 12p "implicit restore of a file on copytool mount point"
+
 test_13() {
 	# test needs a running copytool
 	copytool_setup
@@ -1082,8 +1421,8 @@ test_13() {
 			CURR_FILE="$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 > $CURR_FILE"
 		done
 	done
 	# import to Lustre
@@ -1260,19 +1599,41 @@ test_21() {
 	local fid=$(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 +1651,8 @@ 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"
@@ -1471,7 +1834,7 @@ test_24b() {
 	copytool_setup
 	mkdir -p $DIR/$tdir
 
-	# Check that root can do HSM actions on a ordinary user's file.
+	# Check that root can do HSM actions on a regular user's file.
 	rm -f $file
 	fid=$(make_small $file)
 	sum0=$(md5sum $file)
@@ -1690,7 +2053,10 @@ test_26() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -1727,7 +2093,10 @@ test_27b() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1746,7 +2115,10 @@ test_28() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -1827,6 +2199,47 @@ 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 slp_sum1=$(md5sum /bin/sleep)
+	local fid=$(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
+	if [[ $? != 0 ]]; then
+		local slp_sum2=$(md5sum /bin/sleep)
+		# in case sleep file is modified during the test
+		[[ $slp_sum1 == $slp_sum2 ]] &&
+			error "Binary overwritten during exec"
+	fi
+
+	# 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=$2
@@ -1852,11 +2265,11 @@ restore_and_check_size() {
 		cpt=$((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
 }
 
@@ -1889,7 +2302,10 @@ test_31b() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1910,7 +2326,10 @@ test_31c() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress_aligned $f)
+	local fid
+	fid=$(make_large_for_progress_aligned $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1931,11 +2350,23 @@ test_33() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
 
+	# to be sure wait_all_done will not be mislead by previous tests
+	# and ops.
+	cdt_purge
+	wait_for_grace_delay
+	# Also raise grace_delay significantly so the Canceled
+	# Restore action will stay enough long avail.
+	local old_grace=$(get_hsm_param grace_delay)
+	set_hsm_param grace_delay 100
+
 	md5sum $f >/dev/null &
 	local pid=$!
 	wait_request_state $fid RESTORE STARTED
@@ -1948,8 +2379,29 @@ test_33() {
 
 	$LFS hsm_cancel $f
 
-	wait_request_state $fid RESTORE CANCELED
-	wait_request_state $fid CANCEL SUCCEED
+	# instead of waiting+checking both Restore and Cancel ops
+	# sequentially, wait for both to be finished and then check
+	# each results.
+	wait_all_done 100 $fid
+	local rstate=$(get_request_state $fid RESTORE)
+	local cstate=$(get_request_state $fid CANCEL)
+
+	# restore orig grace_delay.
+	set_hsm_param grace_delay $old_grace
+
+	if [[ "$rstate" == "CANCELED" ]] ; then
+		[[ "$cstate" == "SUCCEED" ]] ||
+			error "Restore state is CANCELED and Cancel state " \
+			       "is not SUCCEED but $cstate"
+		echo "Restore state is CANCELED, Cancel state is SUCCEED"
+	elif [[ "$rstate" == "SUCCEED" ]] ; then
+		[[ "$cstate" == "FAILED" ]] ||
+			error "Restore state is SUCCEED and Cancel state " \
+				"is not FAILED but $cstate"
+		echo "Restore state is SUCCEED, Cancel state is FAILED"
+	else
+		error "Restore state is $rstate and Cancel state is $cstate"
+	fi
 
 	[ -z $killed ] ||
 		error "Cannot kill process waiting for restore ($killed)"
@@ -1965,7 +2417,10 @@ test_34() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1998,7 +2453,10 @@ test_35() {
 
 	local f=$DIR/$tdir/$tfile
 	local f1=$DIR/$tdir/$tfile-1
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	local fid1=$(copy_file /etc/passwd $f1)
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
@@ -2034,7 +2492,10 @@ test_36() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -2086,7 +2547,13 @@ test_40() {
 			fid=$(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 do_facet $SINGLEAGT "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
@@ -2221,7 +2688,9 @@ test_56() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2330,10 +2799,314 @@ test_58() {
 }
 run_test 58 "Truncate a released file will trigger restore"
 
-test_90() {
-	file_count=57
+test_60() {
+	# This test validates the fix for LU-4512. Ensure that the -u
+	# option changes the progress reporting interval from the
+	# default (30 seconds) to the user-specified interval.
+	local interval=5
+	local progress_timeout=$((interval * 4))
+
+	# test needs a new running copytool
+	copytool_cleanup
+	HSMTOOL_UPDATE_INTERVAL=$interval copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	local mdtidx=0
+	local mdt=${MDT_PREFIX}${mdtidx}
+	local mds=mds$((mdtidx + 1))
+
+	# Wait for copytool to register
+	wait_update_facet $mds \
+		"$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
+		uuid 100 || error "coyptool failed to register with $mdt"
+
+	local start_at=$(date +%s)
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
+		error "could not archive file"
+
+	local agent=$(facet_active_host $SINGLEAGT)
+	local prefix=$TESTLOG_PREFIX
+	[[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
+	local copytool_log=$prefix.copytool_log.$agent.log
+
+
+	wait_update $agent \
+	    "grep -o start.copy $copytool_log" "start copy" 100 ||
+		error "copytool failed to start"
+
+	local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
+	cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
+
+	local RESULT
+	local WAIT=0
+	local sleep=1
+
+	echo -n "Expecting a progress update within $progress_timeout seconds... "
+	while [ true ]; do
+		RESULT=$(do_node $(facet_active_host $mds) "$cmd")
+		if [ $RESULT -gt 0 ]; then
+			echo "$RESULT bytes copied in $WAIT seconds."
+			break
+		elif [ $WAIT -ge $progress_timeout ]; then
+			error "Timed out waiting for progress update!"
+			break
+		fi
+		WAIT=$((WAIT + sleep))
+		sleep $sleep
+	done
+
+	local finish_at=$(date +%s)
+	local elapsed=$((finish_at - start_at))
+
+	# Ensure that the progress update occurred within the expected window.
+	if [ $elapsed -lt $interval ]; then
+		error "Expected progress update after at least $interval seconds"
+	fi
+
+	cdt_clear_no_retry
+	copytool_cleanup
+}
+run_test 60 "Changing progress update interval from default"
+
+test_70() {
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+
+	# Just start and stop the copytool to generate events.
+	cdt_clear_no_retry
+
+	# Wait for the copytool to register.
+	wait_update --verbose $(facet_active_host mds1) \
+		"$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
+		uuid 100 ||
+		error "copytool failed to register with MDT0000"
+
+	copytool_cleanup
+
+	local REGISTER_EVENT
+	local UNREGISTER_EVENT
+	while read event; do
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "REGISTER" ]; then
+			REGISTER_EVENT=$event
+		elif [ $event_type == "UNREGISTER" ]; then
+			UNREGISTER_EVENT=$event
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$REGISTER_EVENT" ]; then
+		error "Copytool failed to send register event to FIFO"
+	fi
+
+	if [ -z "$UNREGISTER_EVENT" ]; then
+		error "Copytool failed to send unregister event to FIFO"
+	fi
+
+	copytool_monitor_cleanup
+	echo "Register/Unregister events look OK."
+}
+run_test 70 "Copytool logs JSON register/unregister events to FIFO"
+
+test_71() {
+	# Bump progress interval for livelier events.
+	local interval=5
+
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_UPDATE_INTERVAL=$interval \
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
+		error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local expected_fields="event_time data_fid source_fid"
+	expected_fields+=" total_bytes current_bytes"
+
+	local START_EVENT
+	local FINISH_EVENT
+	while read event; do
+		# Make sure we're not getting anything from previous events.
+		for field in $expected_fields; do
+			unset $field
+		done
+
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "ARCHIVE_START" ]; then
+			START_EVENT=$event
+			continue
+		elif [ $event_type == "ARCHIVE_FINISH" ]; then
+			FINISH_EVENT=$event
+			continue
+		elif [ $event_type != "ARCHIVE_RUNNING" ]; then
+			continue
+		fi
+
+		# Do some simple checking of the progress update events.
+		for expected_field in $expected_fields; do
+			if [ -z ${!expected_field+x} ]; then
+				error "Missing $expected_field field in event"
+			fi
+		done
+
+		if [ $total_bytes -eq 0 ]; then
+			error "Expected total_bytes to be > 0"
+		fi
+
+		# These should be identical throughout an archive
+		# operation.
+		if [ $source_fid != $data_fid ]; then
+			error "Expected source_fid to equal data_fid"
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$START_EVENT" ]; then
+		error "Copytool failed to send archive start event to FIFO"
+	fi
+
+	if [ -z "$FINISH_EVENT" ]; then
+		error "Copytool failed to send archive finish event to FIFO"
+	fi
+
+	echo "Archive events look OK."
+
+	cdt_clear_no_retry
+	copytool_cleanup
+	copytool_monitor_cleanup
+}
+run_test 71 "Copytool logs JSON archive events to FIFO"
+
+test_72() {
+	# Bump progress interval for livelier events.
+	local interval=5
+
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_UPDATE_INTERVAL=$interval \
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+	local test_file=$HSMTOOL_MONITOR_DIR/file
+
+	local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
+	cmd+="conv=fsync"
+	do_facet $SINGLEAGT "$cmd" ||
+		error "cannot create $test_file on $SINGLEAGT"
+	copy2archive $test_file $tdir/$tfile
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	import_file $tdir/$tfile $f
+	f=$DIR2/$tdir/$tfile
+	echo "Verifying released state: "
+	check_hsm_flags $f "0x0000000d"
+
+	local fid=$(path2fid $f)
+	$LFS hsm_restore $f
+	wait_request_state $fid RESTORE SUCCEED
+
+	local expected_fields="event_time data_fid source_fid"
+	expected_fields+=" total_bytes current_bytes"
+
+	local START_EVENT
+	local FINISH_EVENT
+	while read event; do
+		# Make sure we're not getting anything from previous events.
+		for field in $expected_fields; do
+			unset $field
+		done
+
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "RESTORE_START" ]; then
+			START_EVENT=$event
+			if [ $source_fid != $data_fid ]; then
+				error "source_fid should == data_fid at start"
+			fi
+			continue
+		elif [ $event_type == "RESTORE_FINISH" ]; then
+			FINISH_EVENT=$event
+			if [ $source_fid != $data_fid ]; then
+				error "source_fid should == data_fid at finish"
+			fi
+			continue
+		elif [ $event_type != "RESTORE_RUNNING" ]; then
+			continue
+		fi
+
+		# Do some simple checking of the progress update events.
+		for expected_field in $expected_fields; do
+			if [ -z ${!expected_field+x} ]; then
+				error "Missing $expected_field field in event"
+			fi
+		done
+
+		if [ $total_bytes -eq 0 ]; then
+			error "Expected total_bytes to be > 0"
+		fi
+
+		# When a restore starts out, the data fid is the same as the
+		# source fid. After the restore has gotten going, we learn
+		# the new data fid. Once the restore has finished, the source
+		# fid is set to the new data fid.
+		#
+		# We test this because some monitoring software may depend on
+		# this behavior. If it changes, then the consumers of these
+		# events may need to be modified.
+		if [ $source_fid == $data_fid ]; then
+			error "source_fid should != data_fid during restore"
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$START_EVENT" ]; then
+		error "Copytool failed to send restore start event to FIFO"
+	fi
+
+	if [ -z "$FINISH_EVENT" ]; then
+		error "Copytool failed to send restore finish event to FIFO"
+	fi
+
+	echo "Restore events look OK."
+
+	cdt_clear_no_retry
+	copytool_cleanup
+	copytool_monitor_cleanup
+
+	rm -rf $test_dir
+}
+run_test 72 "Copytool logs JSON restore events to FIFO"
+
+test_90() {
+	file_count=51 # Max number of files constrained by LNET message size
+	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	local f=$DIR/$tdir/$tfile
 	local FILELIST=/tmp/filelist.txt
 	local i=""
 
@@ -2342,7 +3115,14 @@ test_90() {
 		fid=$(copy_file /etc/hosts $f.$i)
 		echo $f.$i >> $FILELIST
 	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 do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
+		copytool_setup
+	else
+		local dai=$(get_hsm_param default_archive_id)
+		copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir
+	fi
 	# to be sure wait_all_done will not be mislead by previous tests
 	cdt_purge
 	wait_for_grace_delay
@@ -2382,7 +3162,7 @@ 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 +3192,7 @@ 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"
@@ -2429,12 +3209,15 @@ test_104() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	# if cdt is on, it can serve too quickly the request
 	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 +3238,12 @@ 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 +3253,70 @@ 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"
 
@@ -2698,7 +3523,10 @@ test_200() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_large_for_cancel $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	# test with cdt on is made in test_221
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2739,7 +3567,10 @@ test_202() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -2767,7 +3598,7 @@ 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
@@ -2784,7 +3615,9 @@ test_221() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_large_for_cancel $f)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 
@@ -2794,7 +3627,7 @@ 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 +3652,7 @@ 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 +3678,7 @@ 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 +3707,7 @@ 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 ]] ||
@@ -2891,7 +3724,9 @@ test_223b() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2903,7 +3738,7 @@ 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 +3764,7 @@ 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 ]] ||
@@ -2951,7 +3786,9 @@ test_225() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_large_for_progress $f)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2965,9 +3802,9 @@ 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 +3836,7 @@ 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 +3844,7 @@ 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 +3864,7 @@ 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 +3875,7 @@ 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 +3915,9 @@ 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 +3934,15 @@ 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 +3971,12 @@ 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
@@ -3153,24 +3991,30 @@ test_251() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_large_for_cancel $f)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	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)
 	set_hsm_param loop_period 2
 	cdt_enable
 
+	# clear locks to avoid extra delay caused by flush/cancel
+	# and thus prevent early copytool death to timeout.
+	cancel_lru_locks osc
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE STARTED
 	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 +4071,11 @@ 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 +4089,231 @@ 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 $SINGLEAGT "$FSNAME-MDT000." "deactivate"
+
+	HSMTOOL_NOERROR=true 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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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"
+
+test_405() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local striped_dir=$DIR/$tdir/striped_dir
+
+	# create striped dir on all of MDTs
+	$LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
+
+	local fid1=$(make_small_sync $striped_dir/${tfile}_0)
+	local fid2=$(make_small_sync $striped_dir/${tfile}_1)
+	local fid3=$(make_small_sync $striped_dir/${tfile}_2)
+	local fid4=$(make_small_sync $striped_dir/${tfile}_3)
+
+	local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0)
+	local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1)
+	local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2)
+	local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3)
+
+	# check that compound requests are shunt to the rights MDTs
+	$LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1  \
+			 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
+		error "lfs hsm_archive"
+
+	wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
+		echo "archive successful on $fid1"
+	wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
+		echo "archive successful on $fid2"
+	wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
+		echo "archive successful on $fid3"
+	wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
+		echo "archive successful on $fid4"
+
+	$LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
+	$LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
+	$LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
+	$LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
+
+	cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
+	cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
+	cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
+	cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
+
+	copytool_cleanup
+}
+run_test 405 "archive and release under striped directory"
+
+test_500()
+{
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&
+		skip "HSM migrate is not supported" && return
+
+	# Stop the existing copytool
+	copytool_cleanup
+
+	test_mkdir -p $DIR/$tdir
+	llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed"
+}
+run_test 500 "various LLAPI HSM tests"
+
 copytool_cleanup
 
 complete $SECONDS