X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=fd2f71121f5ee8a65492ae2314ef59a8c6c1e8aa;hp=bdabe108ba514eb4e3ce70f6207f61d90505d379;hb=89a668f6c8770c5af8c99984003ffd9844dc2904;hpb=74d92933108dc64b110a843352cf3336dca249d0

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index bdabe10..fd2f711 100755
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -103,7 +103,16 @@ init_agt_vars() {
 	export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
 	export HSMTOOL_TESTDIR
 	export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
+	# $hsm_root/$HSMTMP Makes $hsm_root dir path less generic to ensure
+	# rm -rf $hsm_root/* is safe even if $hsm_root becomes unset to avoid
+	# deleting everything in filesystem, independent of any copytool.
+	export HSMTMP=${HSMTMP:-"shsm"}
+
 	HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
+
+	[ -z "${HSM_ARCHIVE// /}" ] && error "HSM_ARCHIVE is empty!"
+	HSM_ARCHIVE=$HSM_ARCHIVE/$HSMTMP
+
 	HSM_ARCHIVE_NUMBER=2
 
 	# The test only support up to 10 MDTs
@@ -244,6 +253,9 @@ copytool_setup() {
 	local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
 	local arc_id=$3
 	local hsm_root=${4:-$(copytool_device $facet)}
+
+	[ -z "${hsm_root// /}" ] && error "copytool_setup: hsm_root empty!"
+
 	local agent=$(facet_active_host $facet)
 
 	if [[ -z "$arc_id" ]] &&
@@ -254,14 +266,16 @@ copytool_setup() {
 
 	if $HSM_ARCHIVE_PURGE; then
 		echo "Purging archive on $agent"
-		do_facet $facet "rm -rf $hsm_root/*"
+		do_facet $facet "rm -rf $hsm_root/$HSMTMP/*"
 	fi
 
 	echo "Starting copytool $facet on $agent"
-	do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
+	do_facet $facet "mkdir -p $hsm_root/$HSMTMP/" ||
+			error "mkdir '$hsm_root/$HSMTMP' failed"
 	# bandwidth is limited to 1MB/s so the copy time is known and
 	# independent of hardware
-	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
+	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon"
+	cmd+=" --hsm-root $hsm_root/$HSMTMP"
 	[[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
 	[[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
 		cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
@@ -301,6 +315,9 @@ copytool_cleanup() {
 	local agt_facet=$SINGLEAGT
 	local agt_hosts=${1:-$(facet_active_host $agt_facet)}
 	local hsm_root=$(copytool_device $agt_facet)
+
+	[ -z "${hsm_root// /}" ] && error "copytool_cleanup: hsm_root empty!"
+
 	local i
 	local facet
 	local param
@@ -349,7 +366,7 @@ copytool_cleanup() {
 	done
 
 	if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
-		do_facet $agt_facet "rm -rf $hsm_root/*"
+		do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
 	fi
 }
 
@@ -794,12 +811,72 @@ parse_json_event() {
 	echo $raw_event | python -c "$json_parser"
 }
 
-# populate MDT device array
-get_mdt_devices
+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 $((mdsno - 1))
+	done
+}
+
+get_agent_uuid() {
+	local agent=${1:-$(facet_active_host $SINGLEAGT)}
+
+	# Lustre mount-point is mandatory and last parameter on
+	# copytool cmd-line.
+	local mntpnt=$(do_rpc_nodes $agent ps -C $HSMTOOL_BASE -o args= |
+		       awk '{print $NF}')
+	[ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
+				  "parameter"
+	do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
+}
 
 # initiate variables
 init_agt_vars
 
+# populate MDT device array
+get_mdt_devices
+
 # cleanup from previous bad setup
 kill_copytools
 
@@ -982,13 +1059,19 @@ test_8() {
 run_test 8 "Test default archive number"
 
 test_9() {
-	mkdir -p $DIR/$tdir
-	local f=$DIR/$tdir/$tfile
-	local fid=$(copy_file /etc/passwd $f)
 	# we do not use the default one to be sure
 	local new_an=$((HSM_ARCHIVE_NUMBER + 1))
 	copytool_cleanup
 	copytool_setup $SINGLEAGT $MOUNT $new_an
+
+	# give time for CT to register with MDTs
+	sleep $(($MDSCOUNT*2))
+	local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
+	check_agent_registered $uuid
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
 	$LFS hsm_archive --archive $new_an $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -1687,7 +1770,8 @@ test_16() {
 	$LFS hsm_archive $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	local end=$(date +%s)
-	local duration=$((end - start))
+	# Add 1 to account for rounding errors between start and end (LU-8155)
+	local duration=$((end - start + 1))
 
 	[[ $duration -ge $goal ]] ||
 		error "Transfer is too fast $duration < $goal"
@@ -1921,8 +2005,7 @@ test_24a() {
 	[ $ctime0 -eq $ctime1 ] ||
 		error "release changed ctime from $ctime0 to $ctime1"
 
-	# Restore should not change atime or mtime and should not
-	# decrease ctime.
+	# Restore should not change any timestamps.
 	$LFS hsm_restore $file
 	wait_request_state $fid RESTORE SUCCEED
 
@@ -2049,7 +2132,6 @@ test_24c() {
 	chown $RUNAS_ID:nobody $file ||
 		error "cannot chown '$file' to '$RUNAS_ID:nobody'"
 
-	set_hsm_param user_request_mask ""
 	$RUNAS $LFS hsm_$action $file &&
 		error "$action by user should fail"
 
@@ -2063,7 +2145,6 @@ test_24c() {
 	chown nobody:$RUNAS_GID $file ||
 		error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
 
-	set_hsm_param group_request_mask ""
 	$RUNAS $LFS hsm_$action $file &&
 		error "$action by group should fail"
 
@@ -2077,7 +2158,6 @@ test_24c() {
 	chown nobody:nobody $file ||
 		error "cannot chown '$file' to 'nobody:nobody'"
 
-	set_hsm_param other_request_mask ""
 	$RUNAS $LFS hsm_$action $file &&
 		error "$action by other should fail"
 
@@ -2255,270 +2335,6 @@ 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
-
-	rm -f $f
-
-	local f2=$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
-
-	cat $f2 > /dev/null
-
-	local f3=$DIR/$tdir/${tfile}_3
-	local fid3=$(copy_file /etc/passwd $f3)
-
-	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
-	wait_request_state $fid3 ARCHIVE SUCCEED
-
-	local f4=$DIR/$tdir/${tfile}_4
-	local fid4=$(copy_file /etc/passwd $f4)
-
-	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f4
-	wait_request_state $fid4 ARCHIVE SUCCEED
-
-	trap cleanup_test_26a EXIT
-
-	# set a long grace_delay vs short loop_period
-	local orig_loop_period=$(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
-
-	set_hsm_param remove_archive_on_last_unlink 1
-
-	rm -f $f3
-	cat $f4 > /dev/null
-	local f4bis=$DIR/$tdir/${tfile}_4bis
-	ln $f4 $f4bis
-	[[ $? -eq 0 ]] || error "Unable to create hard-link"
-	rm -f $f4
-
-	# Since CDT is not signaled for RAoLU requests to be sure it
-	# will wake-up to send remove request and copytool will process
-	# it, wait for loop_period + some extra-time.
-	local loop_period=$(get_hsm_param loop_period)
-	sleep $((loop_period + 5))
-
-	set_hsm_param remove_archive_on_last_unlink 0
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid
-	[[ $? -eq 0 ]] || error "File being removed on archive"
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid2
-	[[ $? -eq 0 ]] || error "File being removed on archive"
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid3
-	[[ $? -eq 0 ]] && error "File not being removed on archive"
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid4
-	[[ $? -eq 0 ]] || error "File being removed on archive"
-
-	# previous actions elapsed time should be < grace_delay
-	wait_request_state $fid3 REMOVE SUCCEED
-
-	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
-	search_and_kill_copytool
-	sleep 5
-	search_copytools && error "Copytool should have stopped"
-	HSM_ARCHIVE_PURGE=false copytool_setup
-
-	wait_request_state $fid REMOVE SUCCEED
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid
-	[[ $? -eq 0 ]] && error "File not being removed on archive"
-
-	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 $fid ARCHIVE SUCCEED
-
-	trap cleanup_test_26c EXIT
-
-	# set a long grace_delay vs short loop_period
-	local orig_loop_period=$(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 $f failed"
-	local pid=$!
-	# give multiop a chance to open
-	sleep 2
-
-	rm -f $f
-
-	# Since CDT is not signaled for RAoLU requests to be sure it
-	# will wake-up to send remove request and copytool will process
-	# it, wait for loop_period + some extra-time.
-	local loop_period=$(get_hsm_param loop_period)
-	sleep $((loop_period + 5))
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid
-	[[ $? -eq 0 ]] || error "File being removed on archive"
-
-	kill -USR1 $pid || error "multiop early exit"
-	# should reach autotest timeout if multiop fails to trap
-	# signal, close file, and exit ...
-	wait $pid || error
-
-	# again, wait for loop_period + some extra-time, to allow
-	# CDT enough time to handle remove request.
-	sleep $((loop_period + 5))
-
-	set_hsm_param remove_archive_on_last_unlink 0
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid
-	[[ $? -eq 0 ]] && error "File not being removed on archive"
-
-	# previous actions elapsed time should be < grace_delay
-	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
-	check_hsm_flags $f "0x00000009"
-
-	trap cleanup_test_26d EXIT
-
-	# set a long grace_delay vs short loop_period
-	local orig_loop_period=$(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 $f O_c || error "multiop failed"
-	local MULTIPID=$!
-	# give multiop a chance to open
-	sleep 2
-
-	rm -f $f
-
-	mds_evict_client
-
-	# Since CDT is not signaled for RAoLU requests to be sure it
-	# will wake-up to send remove request and copytool will process
-	# it, wait for loop_period + some extra-time.
-	local loop_period=$(get_hsm_param loop_period)
-	sleep $((loop_period + 5))
-
-	set_hsm_param remove_archive_on_last_unlink 0
-
-	do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid
-	[[ $? -eq 0 ]] && error "File not being removed on archive"
-
-	# previous actions elapsed time should be < grace_delay
-	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
@@ -3815,66 +3631,6 @@ test_105() {
 }
 run_test 105 "Restart of coordinator"
 
-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 $((mdsno - 1))
-	done
-}
-
-get_agent_uuid() {
-	local agent=${1:-$(facet_active_host $SINGLEAGT)}
-
-	# Lustre mount-point is mandatory and last parameter on
-	# copytool cmd-line.
-	local mntpnt=$(do_rpc_nodes $agent pgrep -fl $HSMTOOL_BASE |
-		       grep -v pgrep | awk '{print $NF}')
-	[ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
-				  "parameter"
-	do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
-}
-
 test_106() {
 	# test needs a running copytool
 	copytool_setup