X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=a69302a8d41779dd3e4aa1b354b2c59ac78f58be;hb=cff9f1e7c6a41bfa05d1455b8964860803d12612;hp=dd700c3ee8d0c400704740fc32cfe46c8309292f;hpb=2dbb4d1ae3bcda5c733c512df141837289e03b7a;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index dd700c3..a69302a 100755
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -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/*"
@@ -741,6 +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
@@ -851,9 +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
 
@@ -900,6 +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
 
@@ -909,9 +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
@@ -2137,6 +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() {
@@ -2145,16 +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'"
@@ -2163,9 +2211,11 @@ 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
@@ -2299,6 +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
@@ -2462,7 +2728,8 @@ test_29d() {
 
 	rm -f $file
 
-	$LFS hsm_remove -a 0 $fid
+	$LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
+		error "cannot hsm_remove '$fid'"
 
 	# give time for CDT to handle remove request and create broadcasted
 	sleep 2
@@ -3292,6 +3559,23 @@ test_60() {
 }
 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
@@ -4533,6 +4817,43 @@ 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"
@@ -4681,7 +5002,7 @@ mdc_change_state() # facet, MDT_pattern, activate|deactivate
 	done
 }
 
-test_402() {
+test_402a() {
 	# make sure there is no running copytool
 	copytool_cleanup
 
@@ -4698,7 +5019,34 @@ 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