X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=25c9cf8a14f21bf93c2119b77382a4a5b690eda9;hb=cc6ef11d2f972ebc440013bddda87a536a09750c;hp=fd2f71121f5ee8a65492ae2314ef59a8c6c1e8aa;hpb=89a668f6c8770c5af8c99984003ffd9844dc2904;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index fd2f711..25c9cf8 100755
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,8 +11,8 @@ SRCDIR=$(dirname $0)
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test:    LU-3815
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
+# bug number for skipped test:
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
@@ -170,15 +170,18 @@ wait_copytools() {
 	local wait_timeout=200
 	local wait_start=$SECONDS
 	local wait_end=$((wait_start + wait_timeout))
+	local sleep_time=100000 # 0.1 second
 
 	while ((SECONDS < wait_end)); do
-		sleep 2
 		if ! search_copytools $hosts; then
 			echo "copytools stopped in $((SECONDS - wait_start))s"
 			return 0
 		fi
 
 		echo "copytools still running on $hosts"
+		usleep $sleep_time
+		[ $sleep_time -lt 32000000 ] && # 3.2 seconds
+			sleep_time=$(bc <<< "$sleep_time * 2")
 	done
 
 	# try to dump Copytool's stack
@@ -258,12 +261,6 @@ copytool_setup() {
 
 	local agent=$(facet_active_host $facet)
 
-	if [[ -z "$arc_id" ]] &&
-		do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
-			echo "Only wakeup running copytool $facet on $agent"
-			return 0
-	fi
-
 	if $HSM_ARCHIVE_PURGE; then
 		echo "Purging archive on $agent"
 		do_facet $facet "rm -rf $hsm_root/$HSMTMP/*"
@@ -377,6 +374,13 @@ copytool_suspend() {
 	echo "Copytool is suspended on $agents"
 }
 
+copytool_continue() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+
+	do_nodesv $agents "pkill -CONT -x $HSMTOOL_BASE" || return 0
+	echo "Copytool is continued on $agents"
+}
+
 copytool_remove_backend() {
 	local fid=$1
 	local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
@@ -681,66 +685,18 @@ check_enough_free_space() {
 	return 0
 }
 
-make_large_for_striping() {
+make_custom_file_for_progress() {
 	local file2=${1/$DIR/$DIR2}
-	local sz=$($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 ||
-		file_creation_failure dd $file2 $?
-
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress() {
-	local file2=${1/$DIR/$DIR2}
-
-	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 ||
-		file_creation_failure dd $file2 $?
-
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress_aligned() {
-	local file2=${1/$DIR/$DIR2}
-
-	cleanup_large_files
+	local fsize=${2:-"39"}
+	local blksz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
+	blksz=${3:-$blksz}
 
-	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 ||
-		file_creation_failure dd $file2 $?
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_cancel() {
-	local file2=${1/$DIR/$DIR2}
+	[[ $fsize -gt  0 ]] || error "Invalid file size"
+	[[ $blksz -gt 0 ]] || error "Invalid stripe size"
 
 	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 ||
+	check_enough_free_space $fsize $blksz || return $?
+	dd if=/dev/zero of=$file2 count=$fsize bs=$blksz conv=fsync ||
 		file_creation_failure dd $file2 $?
 	path2fid $1 || error "cannot get fid on $1"
 }
@@ -782,6 +738,15 @@ get_request_count() {
 		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
+# Ensure the number of HSM request for a given FID is correct
+# assert_request_count FID REQUEST_TYPE COUNT [ERROR_MSG]
+assert_request_count() {
+	local request_count=$(get_request_count $1 $2)
+	local default_error_msg=("expected $3 '$2' request(s) for '$1', found "
+				"'$request_count'")
+	[ $request_count -eq $3 ] || error "${4:-"${default_error_msg[@]}"}"
+}
+
 wait_all_done() {
 	local timeout=$1
 	local fid=$2
@@ -799,6 +764,11 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
+wait_for_loop_period() {
+	local val=$(get_hsm_param loop_period)
+	sleep $val
+}
+
 parse_json_event() {
 	local raw_event=$1
 
@@ -887,9 +857,6 @@ echo "Set HSM on and start"
 cdt_set_mount_state enabled
 cdt_check_state enabled
 
-echo "Start copytool"
-copytool_setup
-
 echo "Set sanity-hsm HSM policy"
 cdt_set_sanity_policy
 
@@ -936,6 +903,8 @@ test_1a() {
 	local f=$DIR/$tdir/$tfile
 	local fid=$(make_small $f)
 
+	copytool_setup
+
 	$LFS hsm_archive $f || error "could not archive file"
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -945,9 +914,43 @@ test_1a() {
 	check_hsm_flags $f "0x0000000d"
 
 	$MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
+
+	copytool_cleanup
 }
 run_test 1a "mmap & cat a HSM released file"
 
+test_1b() {
+	mkdir -p $DIR/$tdir
+	$LFS setstripe -E 1M -E 64M -c 2 -E -1 -c 4 $DIR/$tdir ||
+		error "failed to set default stripe"
+	local f=$DIR/$tdir/$tfile
+	rm -f $f
+
+	dd if=/dev/random of=$f bs=1M count=1 conv=sync ||
+		error "failed to create file"
+	local fid=$(path2fid $f)
+
+	copytool_setup
+
+	echo "archive $f"
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	echo "release $f"
+	$LFS hsm_release $f || error "could not release file"
+	echo "verify released state: "
+	check_hsm_flags $f "0x0000000d" && echo "pass"
+
+	echo "restore $f"
+	$LFS hsm_restore $f || error "could not restore file"
+	wait_request_state $fid RESTORE SUCCEED
+	echo "verify restored state: "
+	check_hsm_flags $f "0x00000009" && echo "pass"
+
+	copytool_cleanup
+}
+run_test 1b "Archive, Release & Restore composite file"
+
 test_2() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
@@ -1305,7 +1308,7 @@ test_12c() {
 	local f=$DIR/$tdir/$tfile
 	$LFS setstripe -c 2 $f
 	local fid
-	fid=$(make_large_for_striping $f)
+	fid=$(make_custom_file_for_progress $f 5)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local FILE_CRC=$(md5sum $f)
@@ -1773,7 +1776,7 @@ test_16() {
 	# Add 1 to account for rounding errors between start and end (LU-8155)
 	local duration=$((end - start + 1))
 
-	[[ $duration -ge $goal ]] ||
+	[[ $duration -ge $((goal - 1)) ]] ||
 		error "Transfer is too fast $duration < $goal"
 
 	copytool_cleanup
@@ -2173,6 +2176,7 @@ run_test 24c "check that user,group,other request masks work"
 cleanup_test_24d() {
 	trap 0
 	mount -o remount,rw $MOUNT2
+	zconf_umount $(facet_host $SINGLEAGT) "$MOUNT3"
 }
 
 test_24d() {
@@ -2181,16 +2185,24 @@ test_24d() {
 	local fid1
 	local fid2
 
-	copytool_setup
-
 	mkdir -p $DIR/$tdir
 	rm -f $file1
 	fid1=$(make_small $file1)
 
+	echo $fid1
+	$LFS getstripe $file1
+
 	trap cleanup_test_24d EXIT
+	zconf_mount $(facet_host $SINGLEAGT) "$MOUNT3" ||
+		error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
+	copytool_setup $SINGLEAGT "$MOUNT3" ||
+		error "unable to setup a copytool for the test"
 
 	mount -o remount,ro $MOUNT2
 
+	do_nodes $(comma_list $(nodes_list)) $LCTL clear
+	start_full_debug_logging
+
 	fid2=$(path2fid $file2)
 	[ "$fid1" == "$fid2" ] ||
 		error "FID mismatch '$fid1' != '$fid2'"
@@ -2199,15 +2211,17 @@ test_24d() {
 		error "archive should fail on read-only mount"
 	check_hsm_flags $file1 "0x00000000"
 
-	$LFS hsm_archive $file1
+	$LFS hsm_archive $file1 || error "Fail to archive $file1"
 	wait_request_state $fid1 ARCHIVE SUCCEED
 
+	stop_full_debug_logging
+
 	$LFS hsm_release $file1
 	$LFS hsm_restore $file2
 	wait_request_state $fid1 RESTORE SUCCEED
 
 	$LFS hsm_release $file1 || error "cannot release '$file1'"
-	dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
+	dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
 
 	$LFS hsm_release $file2 &&
 		error "release should fail on read-only mount"
@@ -2320,7 +2334,7 @@ test_26() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2335,6 +2349,222 @@ test_26() {
 }
 run_test 26 "Remove the archive of a valid file"
 
+cleanup_test_26a() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
+	copytool_cleanup
+}
+
+test_26a() {
+	local raolu=$(get_hsm_param remove_archive_on_last_unlink)
+	[[ $raolu -eq 0 ]] || error "RAoLU policy should be off"
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local f2=$DIR/$tdir/${tfile}_2
+	local fid2=$(copy_file /etc/passwd $f2)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
+	wait_request_state $fid2 ARCHIVE SUCCEED
+
+	local f3=$DIR/$tdir/${tfile}_3
+	local fid3=$(copy_file /etc/passwd $f3)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
+	wait_request_state $fid3 ARCHIVE SUCCEED
+
+	trap cleanup_test_26a EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
+
+	rm -f $f
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	ln "$f3" "$f3"_bis || error "Unable to create hard-link"
+	rm -f $f3
+
+	rm -f $f2
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid2 REMOVE SUCCEED
+
+	assert_request_count $fid REMOVE 0 \
+		"Unexpected archived data remove request for $f"
+	assert_request_count $fid3 REMOVE 0 \
+		"Unexpected archived data remove request for $f3"
+
+	cleanup_test_26a
+}
+run_test 26a "Remove Archive On Last Unlink (RAoLU) policy"
+
+cleanup_test_26b() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	copytool_cleanup
+}
+
+test_26b() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	trap cleanup_test_26b EXIT
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	cdt_shutdown
+	cdt_check_state stopped
+
+	rm -f $f
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid REMOVE WAITING
+
+	cdt_enable
+	# copytool must re-register
+	kill_copytools
+	wait_copytools || error "copytool failed to stop"
+	HSM_ARCHIVE_PURGE=false copytool_setup
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	cleanup_test_26b
+}
+run_test 26b "RAoLU policy when CDT off"
+
+cleanup_test_26c() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
+	copytool_cleanup
+}
+
+test_26c() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local f2=$DIR/$tdir/${tfile}_2
+	local fid2=$(copy_file /etc/passwd $f2)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
+	wait_request_state $fid2 ARCHIVE SUCCEED
+
+	trap cleanup_test_26c EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	multiop_bg_pause $f O_c || error "open $f failed"
+	local pid=$!
+
+	rm -f $f
+	rm -f $f2
+
+	wait_request_state $fid2 REMOVE SUCCEED
+	assert_request_count $fid REMOVE 0 \
+		"Unexpected archived data remove request for $f"
+
+	kill -USR1 $pid || error "multiop early exit"
+	# should reach autotest timeout if multiop fails to trap
+	# signal, close file, and exit ...
+	wait $pid || error
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	cleanup_test_26c
+}
+run_test 26c "RAoLU effective when file closed"
+
+cleanup_test_26d() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
+	copytool_cleanup
+}
+
+test_26d() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/motd $f 1)
+
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	trap cleanup_test_26d EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	multiop_bg_pause $f O_c || error "multiop failed"
+	local MULTIPID=$!
+
+	rm -f $f
+
+	mds_evict_client
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	client_up || client_up || true
+
+	kill -USR1 $MULTIPID
+	wait $MULTIPID || error "multiop close failed"
+
+	cleanup_test_26d
+}
+run_test 26d "RAoLU when Client eviction"
+
 test_27a() {
 	# test needs a running copytool
 	copytool_setup
@@ -2360,7 +2590,7 @@ test_27b() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2382,7 +2612,7 @@ test_28() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2471,6 +2701,78 @@ test_29c() {
 }
 run_test 29c "Archive/delete/remove by FID, using a file list."
 
+test_29d() {
+	# test needs more than one CT
+	needclients 3 || return 0
+
+	local n
+	local file
+	local fid
+
+	copytool_cleanup $(comma_list $(agts_nodes))
+
+	# start all of the copytools
+	for n in $(seq $AGTCOUNT); do
+		copytool_setup agt$n $MOUNT2 $n
+	done
+
+	trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
+	# archive files
+	mkdir -p $DIR/$tdir
+	file=$DIR/$tdir/$tfile
+	fid=$(make_small $file)
+
+	$LFS hsm_archive $file
+	wait_request_state $fid ARCHIVE SUCCEED
+	check_hsm_flags $file "0x00000009"
+
+	rm -f $file
+
+	$LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
+		error "cannot hsm_remove '$fid'"
+
+	# give time for CDT to handle remove request and create broadcasted
+	sleep 2
+
+	# remove request has been broadcasted ?
+	local cnt=$(get_request_count $fid REMOVE)
+	# broadcasted requests + original
+	[[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
+		error "remove not broadcasted to all CTs"
+
+	# give time for CDT and CTs to handle broadcasted
+	wait_for_loop_period
+
+	# each agent serves one different archive_id, so broadcasted
+	# hsm_remove request should only succeed once and fail at all others
+	local res
+	local scnt=0
+	local fcnt=0
+	for n in $(seq $AGTCOUNT); do
+		res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+			       $HSM_PARAM.actions | awk \
+			       '/'$fid'.*action=REMOVE archive#='$n'/ \
+			       {print \\\$13}' | cut -f2 -d=")
+		if [[ "$res" == "SUCCEED" ]]; then
+			scnt=$((scnt + 1))
+		elif [[ "$res" == "FAILED" ]]; then
+			fcnt=$((fcnt + 1))
+		fi
+	done
+
+	[[ $scnt -ne 1 ]] &&
+		error "one and only CT should have removed successfully"
+
+	[[ $AGTCOUNT -ne $((scnt + fcnt)) ]] &&
+		error "all but one CT should have failed to remove"
+
+	trap - EXIT
+	copytool_cleanup $(comma_list $(agts_nodes))
+
+}
+run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
+	     "request to be sent once for each registered archive_id"
+
 test_30a() {
 	# restore at exec cannot work on agent node (because of Linux kernel
 	# protection of executables)
@@ -2588,19 +2890,18 @@ restore_and_check_size() {
 	while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
 	do
 		n=$(stat -c "%s" $f)
-		# we echo in both cases to show stat is not
-		# hang
+		# we echo in both cases to show stat is not hang
 		if [[ $n != $s ]]; then
 			echo "size seen is $n != $s"
 			err=1
 		else
 			echo "size seen is right: $n == $s"
 		fi
-		st=$(get_hsm_flags $f)
 		sleep 10
 		cpt=$((cpt + 1))
+		st=$(get_hsm_flags $f)
 	done
-	if [[ $cpt -lt 10 ]]; then
+	if [[ "$st" = "0x00000009" ]]; then
 		echo " "done
 	else
 		echo " restore is too long"
@@ -2639,7 +2940,7 @@ test_31b() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2663,7 +2964,7 @@ test_31c() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress_aligned $f)
+	fid=$(make_custom_file_for_progress $f 33 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2687,7 +2988,7 @@ test_33() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2754,7 +3055,7 @@ test_34() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2790,7 +3091,7 @@ test_35() {
 	local f=$DIR/$tdir/$tfile
 	local f1=$DIR/$tdir/$tfile-1
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local fid1=$(copy_file /etc/passwd $f1)
@@ -2829,7 +3130,7 @@ test_36() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2873,6 +3174,9 @@ test_37() {
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f || error "cannot release $f"
 
+	# Allow previous archive request to expire from the actions log.
+	wait_for_grace_delay
+
 	# Dirty file.
 	dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
 
@@ -2894,6 +3198,12 @@ multi_archive() {
 	echo "$count archive requests submitted"
 }
 
+cleanup_test_40() {
+	trap 0
+	set_hsm_param max_requests $max_requests
+	copytool_cleanup
+}
+
 test_40() {
 	local stream_count=4
 	local file_count=100
@@ -2902,6 +3212,17 @@ test_40() {
 	local i=""
 	local p=""
 	local fid=""
+	local max_requests=$(get_hsm_param max_requests)
+
+	# Increase the number of HSM request that can be performed in
+	# parallel. With the coordinator running once per second, this
+	# also limits the number of requests per seconds that can be
+	# performed, so we pick a decent number. But we also need to keep
+	# that number low because the copytool has no rate limit and will
+	# fail some requests if if gets too many at once.
+	set_hsm_param max_requests 300
+
+	trap cleanup_test_40 EXIT
 
 	for i in $(seq 1 $file_count); do
 		for p in $(seq 1 $stream_count); do
@@ -2928,7 +3249,8 @@ test_40() {
 	wait ${pids[*]}
 	echo OK
 	wait_all_done 100
-	copytool_cleanup
+
+	cleanup_test_40
 }
 run_test 40 "Parallel archive requests"
 
@@ -2993,7 +3315,7 @@ test_54() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -3021,7 +3343,7 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -3050,7 +3372,7 @@ test_56() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
@@ -3191,7 +3513,7 @@ test_60() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 10)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local mdtidx=0
@@ -3246,11 +3568,32 @@ test_60() {
 		error "Expected progress update after at least $interval seconds"
 	fi
 
+	echo "Wait for on going archive hsm action to complete"
+	wait_update $agent "grep -o copied $copytool_log" "copied" 10 ||
+		echo "File archiving not completed even after 10 secs"
+
 	cdt_clear_no_retry
 	copytool_cleanup
 }
 run_test 60 "Changing progress update interval from default"
 
+test_61() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+	cdt_disable
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	rm -f $f
+	cdt_enable
+	wait_request_state $fid ARCHIVE FAILED
+
+	copytool_cleanup
+}
+run_test 61 "Waiting archive of a removed file should fail"
+
 test_70() {
 	# test needs a new running copytool
 	copytool_cleanup
@@ -3310,7 +3653,7 @@ test_71() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
@@ -3582,34 +3925,45 @@ run_test 103 "Purge all requests"
 DATA=CEA
 DATAHEX='[434541]'
 test_104() {
-	# test needs a running copytool
-	copytool_setup
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 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.actions |\
 			grep $fid | cut -f16 -d=")
-	cdt_enable
 
 	[[ "$data1" == "$DATAHEX" ]] ||
 		error "Data field in records is ($data1) and not ($DATAHEX)"
 
+	# archive the file
+	copytool_setup
+
+	wait_request_state $fid ARCHIVE SUCCEED
+
 	copytool_cleanup
 }
 run_test 104 "Copy tool data field"
 
+cleanup_test_105() {
+	trap 0
+	set_hsm_param max_requests $max_requests
+	copytool_cleanup
+}
+
 test_105() {
+	local max_requests=$(get_hsm_param max_requests)
 	mkdir -p $DIR/$tdir
 	local i=""
 
+	set_hsm_param max_requests 300
+
+	trap cleanup_test_105 EXIT
+
 	cdt_disable
 	for i in $(seq -w 1 10); do
 		cp /etc/passwd $DIR/$tdir/$i
@@ -3619,6 +3973,7 @@ test_105() {
 			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_restart
+
 	cdt_disable
 	local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
 			$HSM_PARAM.actions |\
@@ -3628,6 +3983,8 @@ test_105() {
 	[[ "$reqcnt1" == "$reqcnt2" ]] ||
 		error "Requests count after shutdown $reqcnt2 != "\
 		      "before shutdown $reqcnt1"
+
+	cleanup_test_105
 }
 run_test 105 "Restart of coordinator"
 
@@ -3856,7 +4213,7 @@ test_200() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(make_custom_file_for_progress $f 103 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	# test with cdt on is made in test_221
@@ -3904,7 +4261,7 @@ test_202() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -3946,6 +4303,43 @@ test_220() {
 }
 run_test 220 "Changelog for archive"
 
+test_220a() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+
+	# wait request to reach CT
+	wait_request_state $fid ARCHIVE STARTED
+
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid ARCHIVE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+	changelog_cleanup
+
+	# HE_ARCHIVE|ENOENT
+	local target=0x2
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	copytool_cleanup
+}
+run_test 220a "Changelog for failed archive"
+
 test_221() {
 	# test needs a running copytool
 	copytool_setup
@@ -3954,7 +4348,7 @@ test_221() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(make_custom_file_for_progress $f 103 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -4025,6 +4419,72 @@ test_222b() {
 }
 run_test 222b "Changelog for implicit restore"
 
+test_222c() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/passwd $tdir/$tfile
+
+	local f=$DIR/$tdir/$tfile
+	import_file $tdir/$tfile $f
+	local fid=$(path2fid $f)
+
+	changelog_setup
+
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
+	$LFS hsm_restore $f
+
+	# wait request to reach CT
+	wait_request_state $fid RESTORE STARTED
+
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid RESTORE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+
+	# HE_RESTORE|ENOENT
+	local target=0x82
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 222c "Changelog for failed explicit restore"
+
+test_222d() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+
+	copytool_remove_backend $fid
+	md5sum $f
+
+	wait_request_state $fid RESTORE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+
+	# HE_RESTORE|ENOENT
+	local target=0x82
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 222d "Changelog for failed implicit restore"
+
 test_223a() {
 	# test needs a running copytool
 	copytool_setup
@@ -4063,7 +4523,7 @@ test_223b() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -4112,6 +4572,47 @@ test_224() {
 }
 run_test 224 "Changelog for remove"
 
+test_224a() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	copytool_remove_backend $fid
+
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
+	$LFS hsm_remove $f
+
+	# wait for request to reach CT
+	wait_request_state $fid REMOVE STARTED
+
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid REMOVE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
+
+	# HE_REMOVE|ENOENT
+	local target=0x202
+	[[ $flags == $target ]] ||
+		error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 224a "Changelog for failed remove"
+
 test_225() {
 	# test needs a running copytool
 	copytool_setup
@@ -4125,7 +4626,7 @@ test_225() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -4330,7 +4831,7 @@ test_251() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(make_custom_file_for_progress $f 103 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	cdt_disable
@@ -4359,6 +4860,43 @@ test_251() {
 }
 run_test 251 "Coordinator request timeout"
 
+test_252() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(make_custom_file_for_progress $f 103 1048576)
+
+	cdt_disable
+	# to have a short test
+	local old_to=$(get_hsm_param active_request_timeout)
+	set_hsm_param active_request_timeout 20
+	# 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
+	rm -f $f
+
+	# wait but less than active_request_timeout+grace_delay
+	sleep 25
+	wait_request_state $fid ARCHIVE CANCELED
+
+	set_hsm_param active_request_timeout $old_to
+	set_hsm_param loop_period $old_loop
+
+	copytool_cleanup
+}
+run_test 252 "Timeout'ed running archive of a removed file should be canceled"
+
 test_300() {
 	# the only way to test ondisk conf is to restart MDS ...
 	echo "Stop coordinator and remove coordinator state at mount"
@@ -4507,7 +5045,7 @@ mdc_change_state() # facet, MDT_pattern, activate|deactivate
 	done
 }
 
-test_402() {
+test_402a() {
 	# make sure there is no running copytool
 	copytool_cleanup
 
@@ -4524,7 +5062,34 @@ test_402() {
 	# reactivate MDCs
 	mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
 }
-run_test 402 "Copytool start fails if all MDTs are inactive"
+run_test 402a "Copytool start fails if all MDTs are inactive"
+
+test_402b() {
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	touch $f || error "touch $f failed"
+	local fid=$(path2fid $f)
+
+#define OBD_FAIL_MDS_HSM_CT_REGISTER_NET	0x14d
+	do_facet $SINGLEAGT lctl set_param fail_loc=0x14d
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+
+	# give time for CDT to send request and to keep it for retry
+	wait_for_loop_period
+
+	wait_request_state $fid ARCHIVE WAITING
+
+	do_facet $SINGLEAGT lctl set_param fail_loc=0
+
+	# request should succeed now
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 402b "CDT must retry request upon slow start of CT"
 
 test_403() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return