X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Freplay-dual.sh;h=6197ff999781c62e46e088b0cf845b573ba686ba;hb=a42a91c783903ea15ad902032166c7e312dad7ee;hp=1f1c8ab8b71c617cf7ae892b1da5e7214fe1be02;hpb=30be03b4dd593894687773d2a460d441d85f88a2;p=fs%2Flustre-release.git

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index 1f1c8ab..6197ff9 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -1,11 +1,9 @@
 #!/bin/bash
-# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
-# vim:shiftwidth=4:softtabstop=4:tabstop=4:
 
 set -e
 
-# bug number:  LU-2012 10124
-ALWAYS_EXCEPT="14b     15c   $REPLAY_DUAL_EXCEPT"
+# bug number:  LU-2012 10124 LU-7372
+ALWAYS_EXCEPT="14b     15c   26      $REPLAY_DUAL_EXCEPT"
 
 SAVE_PWD=$PWD
 PTLDEBUG=${PTLDEBUG:--1}
@@ -22,6 +20,7 @@ init_logging
 
 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
 
+#                                   7  (min)"
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
 
 [[ $(facet_fstype $SINGLEMDS) == zfs ]] &&
@@ -254,23 +253,25 @@ test_10() {
 run_test 10 "resending a replayed unlink"
 
 test_11() {
-    replay_barrier $SINGLEMDS
-    mcreate $MOUNT1/$tfile-1
-    mcreate $MOUNT2/$tfile-2
-    mcreate $MOUNT1/$tfile-3
-    mcreate $MOUNT2/$tfile-4
-    mcreate $MOUNT1/$tfile-5
-    # drop all reint replies for a while
-    do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
-    # note that with this fail_loc set, facet_failover df will fail
-    facet_failover $SINGLEMDS
-    #sleep for while, let both clients reconnect and timeout
-    sleep $((TIMEOUT * 2))
-    do_facet $SINGLEMDS lctl set_param fail_loc=0
+	replay_barrier $SINGLEMDS
+	mcreate $DIR1/$tfile-1
+	mcreate $DIR2/$tfile-2
+	mcreate $DIR1/$tfile-3
+	mcreate $DIR2/$tfile-4
+	mcreate $DIR1/$tfile-5
+	# drop all reint replies for a while
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0119
+	# note that with this fail_loc set, facet_failover df will fail
+	facet_failover $SINGLEMDS
 
-    rm $MOUNT1/$tfile-[1-5] || return 1
+	local clients=${CLIENTS:-$HOSTNAME}
+	wait_clients_import_state "$clients" $SINGLEMDS FULL
 
-    return 0
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+	rm $DIR1/$tfile-[1-5] || return 1
+
+	return 0
 }
 run_test 11 "both clients timeout during replay"
 
@@ -437,11 +438,11 @@ test_18() { # bug 3822 - evicting client with enqueued lock
 	local DLMTRACE=$(do_facet $SINGLEMDS lctl get_param debug)
 	do_facet $SINGLEMDS lctl set_param debug=+dlmtrace
 	mkdir -p $MOUNT1/$tdir || error "mkdir $MOUNT1/$tdir failed"
-	touch $MOUNT1/$tdir/$tfile
-	#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
-	statmany -s $MOUNT1/$tdir/f 1 500 &
+	touch $MOUNT1/$tdir/${tfile}0 || error "touch file failed"
+	statmany -s $MOUNT1/$tdir/$tfile 1 500 &
 	OPENPID=$!
-	NOW=$(date +%s)
+	NOW=$SECONDS
+	#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b  # hold enqueue
 	sleep 1
 	#define OBD_FAIL_LDLM_BL_CALLBACK_NET			0x305
@@ -471,27 +472,29 @@ test_19() { # Bug 10991 - resend of open request does not fail assertion.
 run_test 19 "resend of open request"
 
 test_20() { #16389
-    BEFORE=`date +%s`
-    replay_barrier $SINGLEMDS
-    touch $MOUNT1/a
-    touch $MOUNT2/b
-    umount $MOUNT2
-    fail $SINGLEMDS
-    rm $MOUNT1/a
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    TIER1=$((`date +%s` - BEFORE))
-    BEFORE=`date +%s`
-    replay_barrier $SINGLEMDS
-    touch $MOUNT1/a
-    touch $MOUNT2/b
-    umount $MOUNT2
-    fail $SINGLEMDS
-    rm $MOUNT1/a
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    TIER2=$((`date +%s` - BEFORE))
-    [ $TIER2 -ge $((TIER1 * 2)) ] && \
-        error "recovery time is growing $TIER2 > $TIER1"
-    return 0
+	local before=$SECONDS
+	replay_barrier $SINGLEMDS
+	touch $DIR1/$tfile.a
+	touch $DIR2/$tfile.b
+	umount $DIR2
+	fail $SINGLEMDS
+	rm $DIR1/$tfile.a
+	zconf_mount $HOSTNAME $DIR2 || error "mount $DIR2 fail"
+	local tier1=$((SECONDS - before))
+
+	before=$SECONDS
+	replay_barrier $SINGLEMDS
+	touch $DIR1/$tfile.a
+	touch $DIR2/$tfile.b
+	umount $DIR2
+	fail $SINGLEMDS
+	rm $DIR1/$tfile.a
+	zconf_mount $HOSTNAME $DIR2 || error "mount $DIR2 fail"
+	local tier2=$((SECONDS - before))
+
+	# timeout is more than 1.5x original timeout
+	((tier2 < tier1 * 6 / 4)) ||
+		error "recovery time $tier2 >= 1.5x original time $tier1"
 }
 run_test 20 "recovery time is not increasing"
 
@@ -523,75 +526,76 @@ test_21a() {
 run_test 21a "commit on sharing"
 
 test_21b_sub () {
-    local mds=$1 
-    do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
+	local mds=$1
+	do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
 
-    do_facet $mds sync
-    do_node $CLIENT1 touch  $MOUNT1/$tfile-1
-    do_node $CLIENT2 mv  $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
-    do_node $CLIENT1 mv  $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+	do_facet $mds sync
+	do_node $CLIENT1 touch $MOUNT1/$tfile-1
+	do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
+	do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
 
-    replay_barrier_nosync $mds
-    shutdown_client $CLIENT2 $MOUNT1
+	replay_barrier_nosync $mds
+	shutdown_client $CLIENT2 $MOUNT1
 
-    facet_failover $mds
+	facet_failover $mds
 
-    # were renames replayed?
-    local rc=0
-    echo UNLINK $MOUNT1/$tfile-3 
-    do_node $CLIENT1 unlink  $MOUNT1/$tfile-3 || \
-        { echo "unlink $tfile-3 fail!" && rc=1; }
+	# were renames replayed?
+	local rc=0
+	echo UNLINK $MOUNT1/$tfile-3
+	do_node $CLIENT1 unlink  $MOUNT1/$tfile-3 ||
+		{ echo "unlink $tfile-3 fail!" && rc=1; }
 
-    boot_node $CLIENT2
-    zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail" 
+	boot_node $CLIENT2
+	zconf_mount_clients $CLIENT2 $MOUNT1 ||
+		error "mount $CLIENT2 $MOUNT1 fail"
 
-    return $rc
+	return $rc
 }
 
 test_21b() {
-    [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
-    [ $CLIENTCOUNT -lt 2 ] && \
-        { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
-
-    if [ "$FAILURE_MODE" = "HARD" ] &&  mixed_mdt_devs; then
-        skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
-        return 0
-    fi
+	[ -z "$CLIENTS" ] && skip "Need two or more clients" && return
+	[ $CLIENTCOUNT -lt 2 ] &&
+		{ skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
 
+	if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
+		skip "Several MDTs on one MDS with FAILURE_MODE=$FAILURE_MODE"
+		return 0
+	fi
 
-    zconf_umount_clients $CLIENTS $MOUNT2
-    zconf_mount_clients $CLIENTS $MOUNT1
+	zconf_umount_clients $CLIENTS $MOUNT2
+	zconf_mount_clients $CLIENTS $MOUNT1
 
-    local param_file=$TMP/$tfile-params
+	local param_file=$TMP/$tfile-params
 
-    local num=$(get_mds_dir $MOUNT1)
+	local mdtidx=$($LFS getstripe -M $MOUNT1)
+	local facet=mds$((mdtidx + 1))
 
-	save_lustre_params mds$num "mdt.*.commit_on_sharing" > $param_file
+	save_lustre_params $facet "mdt.*.commit_on_sharing" > $param_file
 
-    # COS enabled
-    local COS=1
-    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+	# COS enabled
+	local COS=1
+	do_facet $facet lctl set_param mdt.*.commit_on_sharing=$COS
 
-    test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
+	test_21b_sub $facet || error "Not all renames are replayed. COS=$COS"
 
-    # COS disabled (should fail)
-    COS=0
-    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+	# there is still a window when transactions may be written to disk
+	# before the mds device is set R/O. To avoid such a rare test failure,
+	# the check is repeated several times.
+	COS=0
+	local n_attempts=1
+	while true; do
+		# COS disabled (should fail)
+		do_facet $facet lctl set_param mdt.*.commit_on_sharing=$COS
 
-    # there is still a window when transactions may be written to disk before
-    # the mds device is set R/O. To avoid such a rare test failure, the check
-    # is repeated several times.
-    local n_attempts=1
-    while true; do
-    test_21b_sub mds$num || break;
-    let n_attempts=n_attempts+1
-    [ $n_attempts -gt 3 ] &&
-        error "The test cannot check whether COS works or not: all renames are replied w/o COS"
-    done
-    zconf_mount_clients $CLIENTS $MOUNT2
-    restore_lustre_params < $param_file
-    rm -f $param_file
-    return 0
+		test_21b_sub $facet || break
+		n_attempts=$((n_attempts + 1))
+		[ $n_attempts -gt 3 ] &&
+			error "can't check if COS works: rename replied w/o COS"
+	done
+	zconf_mount_clients $CLIENTS $MOUNT2
+	restore_lustre_params < $param_file
+	rm -f $param_file
+	return 0
 }
 run_test 21b "commit on sharing, two clients"
 
@@ -865,7 +869,7 @@ run_test 23d "c1 rmdir d1, M0 drop update reply and fail M0/M1, c2 mkdir d1"
 
 test_24 () {
 	[[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.5.2) ]] ||
-                { skip "Need MDS version newer than 2.5.2"; return 0; }
+		{ skip "Need MDS version newer than 2.5.2"; return 0; }
 
 	touch $MOUNT/$tfile
 	stat $MOUNT/$tfile >&/dev/null
@@ -898,9 +902,15 @@ test_25() {
 	drop_ldlm_cancel "multiop $DIR2/$tfile Ow512" &
 	sleep 1
 
-#define OBD_FAIL_OST_LDLM_REPLY_NET      0x213
 	# failover, replay and resend replayed waiting locks
-	do_facet ost1 lctl set_param fail_loc=0x80000213
+	if [ $(lustre_version_code ost1) -ge $(version_code 2.6.90) ]; then
+		#define OBD_FAIL_LDLM_SRV_CP_AST      0x325
+		do_facet ost1 lctl set_param fail_loc=0x80000325
+	else
+		#define OBD_FAIL_OST_LDLM_REPLY_NET	0x213
+		do_facet ost1 lctl set_param fail_loc=0x80000213
+	fi
+
 	fail ost1
 
 	# multiop does not finish because CP AST is skipped;
@@ -911,8 +921,111 @@ test_25() {
 }
 run_test 25 "replay|resend"
 
+cleanup_26() {
+	trap 0
+	kill -9 $tar_26_pid
+	kill -9 $dbench_26_pid
+	killall -9 dbench
+}
+
+test_26() {
+	local clients=${CLIENTS:-$HOSTNAME}
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=600
+	[ "$SLOW" = "no" ] && duration=200
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	local start_ts=$SECONDS
+	local rc=0
+
+	trap cleanup_26	EXIT
+	(
+		local tar_dir=$DIR/$tdir/run_tar
+		while true; do
+			test_mkdir -p -c$MDSCOUNT $tar_dir || break
+			if [ $MDSCOUNT -ge 2 ]; then
+				$LFS setdirstripe -D -c$MDSCOUNT $tar_dir ||
+					error "set default dirstripe failed"
+			fi
+			cd $tar_dir || break
+			tar cf - /etc | tar xf - || error "tar failed"
+			cd $DIR/$tdir || break
+			rm -rf $tar_dir || break
+		done
+	)&
+	tar_26_pid=$!
+	echo "Started tar $tar_26_pid"
+
+	(
+		local dbench_dir=$DIR2/$tdir/run_dbench
+		while true; do
+			test_mkdir -p -c$MDSCOUNT $dbench_dir || break
+			if [ $MDSCOUNT -ge 2 ]; then
+				$LFS setdirstripe -D -c$MDSCOUNT $dbench_dir ||
+					error "set default dirstripe failed"
+			fi
+			cd $dbench_dir || break
+			rundbench 1 -D $dbench_dir -t 100 > /dev/null 2&>1 ||
+									break
+			cd $DIR/$tdir || break
+			rm -rf $dbench_dir || break
+		done
+	)&
+	dbench_26_pid=$!
+	echo "Started dbench $dbench_26_pid"
+
+	local num_failovers=0
+	local fail_index=1
+	while [ $((SECONDS - start_ts)) -lt $duration ]; do
+		kill -0 $tar_26_pid || error "tar $tar_26_pid missing"
+		kill -0 $dbench_26_pid || error "dbench $dbench_26_pid missing"
+		sleep 2
+		replay_barrier mds$fail_index
+		sleep 2 # give clients a time to do operations
+		# Increment the number of failovers
+		num_failovers=$((num_failovers + 1))
+		log "$TESTNAME fail mds$fail_index $num_failovers times"
+		fail mds$fail_index
+		if [ $fail_index -ge $MDSCOUNT ]; then
+			fail_index=1
+		else
+			fail_index=$((fail_index + 1))
+		fi
+	done
+	# stop the client loads
+	kill -0 $tar_26_pid || error "tar $tar_26_pid stopped"
+	kill -0 $dbench_26_pid || error "dbench $dbench_26_pid stopped"
+	cleanup_26 || true
+}
+run_test 26 "dbench and tar with mds failover"
+
+test_28() {
+	$SETSTRIPE -i 0 -c 1 $DIR2/$tfile
+	dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1
+
+	#define OBD_FAIL_LDLM_SRV_BL_AST	 0x324
+	do_facet ost1 $LCTL set_param fail_loc=0x80000324
+
+	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 &
+	local pid=$!
+	sleep 2
+
+	#define OBD_FAIL_LDLM_GRANT_CHECK        0x32a
+	do_facet ost1 $LCTL set_param fail_loc=0x32a
+
+	fail ost1
+
+	sleep 2
+	cancel_lru_locks OST0000-osc
+	wait $pid || error "dd failed"
+}
+run_test 28 "lock replay should be ordered: waiting after granted"
+
 complete $SECONDS
-SLEEP=$((`date +%s` - $NOW))
+SLEEP=$((SECONDS - $NOW))
 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
 check_and_cleanup_lustre