LU-14876 out: don't connect to busy MDS-MDS export

[fs/lustre-release.git] / lustre / tests / recovery-small.sh

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 675f616..cc54479 100755 (executable)
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -1069,6 +1069,8 @@ test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
 
        check_timeout || return 1
 
+       # make sure all imports are connected and not IDLE
+       do_facet client lfs df > /dev/null
 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
        do_facet client lctl set_param fail_loc=0x505
        local before=$(date +%s)
@@ -1079,7 +1081,7 @@ test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
        # the loser might have to wait for the next ping.
        sleep $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
        do_facet client lctl set_param fail_loc=0x0
-       do_facet client df > /dev/null
+       do_facet client lfs df > /dev/null
 
        local oscs=$(lctl dl | awk '/-osc-/ {print $4}')
        check_clients_evicted $before ${oscs[@]}
@@ -1117,10 +1119,9 @@ test_26b() {      # bug 10140 - evict dead exports by pinger
        # = 9 * PING_INTERVAL + PING_INTERVAL
        # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
        # let's wait $((TIMEOUT * 3)) # bug 19887
-       local rc=0
-       wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
-               error "Client was not evicted by ost" rc=1
-       wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
+       wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) ||
+               error "Client was not evicted by ost"
+       wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) ||
                error "Client was not evicted by mds"
 }
 run_test 26b "evict dead exports"
@@ -2252,8 +2253,8 @@ test_110k() {
        [[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
                skip "Need MDS version at least 2.12.55"
 
-       stop mds2 || error "stop mds2 failed"
        umount $MOUNT
+       stop mds2 || error "stop mds2 failed"
 
 #define OBD_FAIL_FLD_QUERY_REQ 0x1103
        do_facet mds2 lctl set_param fail_loc=0x1103
@@ -2271,6 +2272,38 @@ test_110k() {
 }
 run_test 110k "FID_QUERY failed during recovery"
 
+test_110m () {
+       (( $(lustre_version_code $SINGLEMDS) >= $(version_code 2.14.52) )) ||
+               skip "Need MDS version at least 2.14.52"
+       (( $MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs"
+       local remote_dir=$DIR/$tdir/remote_dir
+       local mdccli
+       local uuid
+       local diridx
+
+       mkdir_on_mdt0 $DIR/$tdir
+
+#define OBD_FAIL_PTLRPC_RESEND_RACE 0x0525
+       do_facet mds1 $LCTL set_param fail_loc=0x80000525
+       $LFS mkdir -i 1 -c2 $remote_dir &
+       mkdir_pid=$!
+       sleep 3
+       # initiate the re-connect & re-send
+       mdccli=$(do_facet mds2 $LCTL dl |
+               awk '/MDT0000-osp-MDT0001/ {print $4;}')
+       uuid=$(do_facet mds2 $LCTL get_param -n osp.$mdccli.mdt_conn_uuid)
+       echo "conn_uuid=$uuid"
+       do_facet mds2 $LCTL set_param "osp.$mdccli.import=connection=$uuid"
+
+       wait $mkdir_pid
+       (( $? == 0 )) || error "mkdir failed"
+
+       diridx=$($LFS getstripe -m $remote_dir)
+       (( $diridx == 1 )) || error "$diridx != 1"
+       rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110m "update resent vs original RPC race"
+
 # LU-2844 mdt prepare fail should not cause umount oops
 test_111 ()
 {
@@ -2889,7 +2922,7 @@ test_138() {
        sleep 55
        stop $SINGLEMDS || error "stop MDS failed"
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0
-       start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
+       start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
                error "start MDS failed"
        zconf_mount_clients $CLIENTS $MOUNT
 }
@@ -2928,7 +2961,7 @@ test_140a() {
        mount_mds_client
 
        local cnt
-       cnt=$(do_facet mds1 $LCTL get_param "mdt.*.exports.*.export" |
+       cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
                grep export_flags.*no_recovery | wc -l)
        echo "$cnt clients with recovery disabled"
        umount_mds_client
@@ -2939,7 +2972,7 @@ test_140a() {
        do_facet mds1 $LCTL set_param mdt.*.local_recovery=1
        mount_mds_client
 
-       cnt=$(do_facet mds1 $LCTL get_param "mdt.*.exports.*.export" |
+       cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
                grep export_flags.*no_recovery | wc -l)
        echo "$cnt clients with recovery disabled"
        umount_mds_client
@@ -2966,11 +2999,12 @@ test_140b() {
        replay_barrier mds1
        umount_mds_client
        fail mds1
+       # Lustre: tfs-MDT0000: Recovery over after 0:03, of 2 clients 2 rec...
        local recovery=$(do_facet mds1 dmesg |
-                        awk -F: '/Recovery over after/ { print $4 }' |
-                        cut -d, -f1 | tail -1)
-       (( $recovery < $TIMEOUT*2 )) ||
-               error "recovery took too long $recovery > $((TIMEOUT * 2))"
+                        awk '/Recovery over after/ { print $6 }' | tail -1 |
+                        awk -F: '{ print $1 * 60 + $2 }')
+       (( recovery < TIMEOUT * 2 + 5 )) ||
+               error "recovery took too long $recovery > $((TIMEOUT * 2 + 5))"
 }
 run_test 140b "local mount is excluded from recovery"
 
@@ -3152,6 +3186,35 @@ test_148() {
 }
 run_test 148 "data corruption through resend"
 
+test_149() {
+       [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+
+       test_mkdir -i 0 -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
+
+       # make an orphan striped dir
+       $MULTIOP $DIR/$tdir D_c &
+       local PID=$!
+       sleep 0.3
+       rmdir $DIR/$tdir || error "can't rmdir"
+
+       # stop a slave MDT where one ons stripe is located
+       stop mds2 -f
+
+       # stopping should not cause orphan as another MDT can
+       # be stopped yet
+       stop mds1 -f
+
+       start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
+       start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS || error "mds1 start fail"
+
+       kill -USR1 $PID
+       wait $PID
+
+       clients_up
+       return 0
+}
+run_test 149 "skip orphan removal at umount"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status