X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-dual.sh;h=d2285009e593c03dec2aa90f3d4872e543e268e2;hp=c4f12d12c6c36fc4f59e20050e34d0d570130ede;hb=aa651cabdd8aa4d5539c66e3abbc5f5b6dc6a614;hpb=ae5d700fdbff4a47740d50a4c486b764823b97bb

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index c4f12d1..d228500 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -2,27 +2,42 @@
 
 set -e
 
-# bug number:  13129 13129 6088 10124 
-ALWAYS_EXCEPT="2     3     8    15c   $REPLAY_DUAL_EXCEPT"
+# bug number:  13129 13129 10124 
+ALWAYS_EXCEPT="2     3     15c   $REPLAY_DUAL_EXCEPT"
 
 SAVE_PWD=$PWD
 PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 SETUP=${SETUP:-""}
 CLEANUP=${CLEANUP:-""}
 MOUNT_2=${MOUNT_2:-"yes"}
 . $LUSTRE/tests/test-framework.sh
 
+if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
+    CONFIG_EXCEPTIONS="17"
+    echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
+    echo "Except the tests: $CONFIG_EXCEPTIONS"
+    ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
+fi
+
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 
-#
+remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
+
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
 
 build_test_filter
 
-cleanup_and_setup_lustre
+check_and_setup_lustre
+MOUNTED=$(mounted_lustre_filesystems)
+if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
+    zconf_mount $HOSTNAME $MOUNT2
+    MOUNTED2=yes
+fi
+
+assert_DIR
 rm -rf $DIR/[df][0-9]*
 
 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
@@ -92,10 +107,8 @@ run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
 test_5() {
     # multiclient version of replay_single.sh/test_8
     mcreate $MOUNT1/a
-    multiop $MOUNT2/a o_tSc &
+    multiop_bg_pause $MOUNT2/a o_tSc || return  1
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $MOUNT1/a
     replay_barrier $SINGLEMDS
     kill -USR1 $pid
@@ -110,12 +123,10 @@ run_test 5 "open, unlink |X| close"
 
 test_6() {
     mcreate $MOUNT1/a
-    multiop $MOUNT2/a o_c &
+    multiop_bg_pause $MOUNT2/a o_c || return 1
     pid1=$!
-    multiop $MOUNT1/a o_c &
+    multiop_bg_pause $MOUNT1/a o_c || return 1
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $MOUNT1/a
     replay_barrier $SINGLEMDS
     kill -USR1 $pid1
@@ -145,9 +156,9 @@ test_9() {
     mcreate $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
     fail $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     rm $MOUNT1/$tfile-[1,2] || return 1
 
@@ -161,9 +172,9 @@ test_10() {
     munlink $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
     fail $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     checkstat $MOUNT1/$tfile-1 && return 1
     checkstat $MOUNT1/$tfile-2 || return 2
@@ -181,12 +192,12 @@ test_11() {
     mcreate $MOUNT2/$tfile-4
     mcreate $MOUNT1/$tfile-5
     # drop all reint replies for a while
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
+    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 sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     rm $MOUNT1/$tfile-[1-5] || return 1
 
@@ -197,14 +208,13 @@ run_test 11 "both clients timeout during replay"
 test_12() {
     replay_barrier $SINGLEMDS
 
-    multiop $DIR/$tfile mo_c &
+    multiop_bg_pause $DIR/$tfile mo_c || return 1
     MULTIPID=$!
-    sleep 5
 
 #define OBD_FAIL_LDLM_ENQUEUE            0x302
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
     facet_failover $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
     df $MOUNT || return 1
 
     ls $DIR/$tfile
@@ -218,9 +228,8 @@ test_12() {
 run_test 12 "open resend timeout"
 
 test_13() {
-    multiop $DIR/$tfile mo_c &
+    multiop_bg_pause $DIR/$tfile mo_c || return 1
     MULTIPID=$!
-    sleep 5
 
     replay_barrier $SINGLEMDS
 
@@ -228,9 +237,9 @@ test_13() {
     wait $MULTIPID || return 4
 
     # drop close 
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
     facet_failover $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
     df $MOUNT || return 1
 
     ls $DIR/$tfile
@@ -261,7 +270,7 @@ test_14() {
 }
 run_test 14 "timeouts waiting for lost client during replay"
 
-test_15() {
+test_15a() {	# was test_15
     replay_barrier $SINGLEMDS
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
@@ -276,72 +285,7 @@ test_15() {
     zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
     return 0
 }
-run_test 15 "timeout waiting for lost client during replay, 1 client completes"
-
-test_15a() {
-    local ost_last_id=""
-    local osc_last_id=""
-    
-    replay_barrier $SINGLEMDS
-    echo "data" > "$MOUNT2/${tfile}-m2"
-
-    umount $MOUNT2
-    facet_failover $SINGLEMDS
-    df $MOUNT || return 1
-    
-    ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
-    mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
-    
-    echo "Ids after MDS<->OST synchonizing"
-    echo "--------------------------------"
-    echo "MDS last_id:"
-    echo $mds_last_id
-    echo "OST last_id:"
-    echo $ost_last_id
-
-    local i=0
-    echo $ost_last_id | while read id; do
-	ost_ids[$i]=$id
-	((i++))
-    done
-    
-    i=0
-    echo $mds_last_id | while read id; do
-	mds_ids[$i]=$id
-	((i++))
-    done
-    
-    local arr_len=${#mds_ids[*]}
-    for ((i=0;i<$arr_len;i++)); do
-	    mds_id=${mds_ids[i]}
-	    ost_id=${ost_ids[i]}
-	    
-	    test $mds_id -ge $ost_id || {
-		echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
-		return 2
-	    }
-    done
-
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    return 0
-}
-#CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
-
-test_15b() {
-    replay_barrier $SINGLEMDS
-    echo "data" > "$MOUNT2/${tfile}-m2"
-    umount $MOUNT2
-
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
-    facet_failover $SINGLEMDS
-
-    df $MOUNT || return 1
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
-    
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    return 0
-}
-#CROW run_test 15b "multiple delayed OST clear orphans"
+run_test 15a "timeout waiting for lost client during replay, 1 client completes"
 
 test_15c() {
     replay_barrier $SINGLEMDS
@@ -379,6 +323,8 @@ test_16() {
 run_test 16 "fail MDS during recovery (3571)"
 
 test_17() {
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
 
@@ -410,18 +356,18 @@ test_18() { # bug 3822 - evicting client with enqueued lock
     statmany -s $MOUNT1/$tdir/f 1 500 &
     OPENPID=$!
     NOW=`date +%s`
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b  # hold enqueue
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b  # hold enqueue
     sleep 1
 #define OBD_FAIL_LDLM_BL_CALLBACK        0x305
-    do_facet client sysctl -w lustre.fail_loc=0x80000305  # drop cb, evict
+    do_facet client lctl set_param fail_loc=0x80000305  # drop cb, evict
     cancel_lru_locks mdc
     usleep 500 # wait to ensure first client is one that will be evicted
     openfile -f O_RDONLY $MOUNT2/$tdir/f0
     wait $OPENPID
     dmesg | grep "entering recovery in server" && \
         error "client not evicted" || true
-    do_facet client "sysctl -w lustre.fail_loc=0"
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet client "lctl set_param fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 }
 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
 
@@ -436,9 +382,170 @@ 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
+    facet_failover $SINGLEMDS
+    df $MOUNT1 || return 1
+    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
+    facet_failover $SINGLEMDS
+    df $MOUNT1 || return 1
+    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 > $TIER1"
+    return 0
+}
+run_test 20 "recovery time is not increasing"
+
+# commit on sharing tests
+test_21a() {
+    local param_file=$TMP/$tfile-params
+
+    save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
+    do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
+    touch  $MOUNT1/$tfile-1
+    mv  $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
+    mv  $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+    replay_barrier_nosync $SINGLEMDS
+    umount $MOUNT2
+
+    facet_failover $SINGLEMDS
+
+    # all renames are replayed
+    unlink  $MOUNT1/$tfile-3 || return 2
+
+    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+
+    do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
+    rm -rf $MOUNT1/$tfile-*
+    restore_lustre_params < $param_file
+    rm -f $param_file
+    return 0
+}
+run_test 21a "commit on sharing"
+
+shutdown_client() {
+    local client=$1
+    local mnt=$2
+
+    if [ "$FAILURE_MODE" = HARD ]; then
+       $POWER_DOWN $client
+       while ping -w 3 -c 1 $client > /dev/null 2>&1; do
+           echo "waiting for node $client to fail"
+           sleep 1
+       done
+    else
+       zconf_umount_clients $client $mnt -f
+    fi
+}
+
+# CMD: determine mds index where directory inode presents 
+get_mds_dir () {
+    local dir=$1
+    local file=$dir/$tfile
+
+    rm -f $file
+    local iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
+    local oldused=($iused)
+
+    touch $file
+    sleep 1
+    iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
+    local newused=($iused)
+
+    local num=0
+    for ((i=0; i<${#newused[@]}; i++)); do
+         if [ ${oldused[$i]} -lt ${newused[$i]} ];  then
+             echo $(( i + 1 ))
+             rm -f $dir/$tfile
+             return 0
+         fi 
+    done
+    error "mdt-s : inodes count OLD ${oldused[@]} NEW ${newused[@]}"
+}
+
+test_21b_sub () {
+    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
+
+    replay_barrier_nosync $mds
+    shutdown_client $CLIENT2 $MOUNT1
+
+    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; }
+
+    boot_node $CLIENT2
+    zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail" 
+
+    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
+
+
+    zconf_umount_clients $CLIENTS $MOUNT2
+    zconf_mount_clients $CLIENTS $MOUNT1
+
+    local param_file=$TMP/$tfile-params
+
+    local num=$(get_mds_dir $MOUNT1)
+
+    save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
+
+    # COS enabled
+    local COS=1
+    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+
+    test_21b_sub mds$num || 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=$COS
+
+    test_21b_sub mds$num && error "Not all renames are replayed. COS=$COS" 
+
+    restore_lustre_params < $param_file
+    rm -f $param_file
+    return 0
+}
+run_test 21b "commit on sharing, two clients"
+
+# end commit on sharing tests 
+
 equals_msg `basename $0`: test complete, cleaning up
 SLEEP=$((`date +%s` - $NOW))
 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
+[ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
 check_and_cleanup_lustre
 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true