b=20647 Ensure all needed pids exits before failover in test set 53

[fs/lustre-release.git] / lustre / tests / replay-single.sh

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index ddba13d..1e4e91f 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -13,14 +13,15 @@ CLEANUP=${CLEANUP:-}
 . $LUSTRE/tests/test-framework.sh
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 CHECK_GRANT=${CHECK_GRANT:-"yes"}
 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
 
-remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0
+require_dsh_mds || exit 0
 
 # Skip these tests
-# bug number:  17466 18857,15962 21240
-ALWAYS_EXCEPT="61d   33a 33b     44 $REPLAY_SINGLE_EXCEPT"
+# bug number:  17466 18857
+ALWAYS_EXCEPT="61d   33a 33b     $REPLAY_SINGLE_EXCEPT"
 
 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
     CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
@@ -936,7 +937,7 @@ test_43() { # bug 2530
 }
 run_test 43 "mds osc import failure during recovery; don't LBUG"
 
-test_44a() {   # was test_44
+test_44a() { # was test_44
     local at_max_saved=0
 
     mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
@@ -950,12 +951,13 @@ test_44a() {      # was test_44
     fi
 
     for i in `seq 1 10`; do
-       echo "$i of 10 ($(date +%s))"
-       do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
-       #define OBD_FAIL_TGT_CONN_RACE     0x701
-       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
-       $LCTL --device $mdcdev recover || return 4
-       df $MOUNT
+        echo "$i of 10 ($(date +%s))"
+        do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+        #define OBD_FAIL_TGT_CONN_RACE     0x701
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
+        # lctl below may fail, it is valid case
+        $LCTL --device $mdcdev recover
+        df $MOUNT
     done
     do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
@@ -970,11 +972,12 @@ test_44b() {
 
     for i in `seq 1 10`; do
         echo "$i of 10 ($(date +%s))"
-       do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
-       #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
-       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
-       $LCTL --device $mdcdev recover || return 4
-       df $MOUNT
+        do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+        #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
+        # lctl below may fail, it is valid case
+        $LCTL --device $mdcdev recover
+        df $MOUNT
     done
     do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     return 0
@@ -1167,6 +1170,9 @@ test_53c() {
         kill -USR1 $close_pid
         cancel_lru_locks mdc    # force the close
 
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
         replay_barrier_nodf $SINGLEMDS
         fail_nodf $SINGLEMDS
         wait $open_pid || return 1
@@ -1182,6 +1188,7 @@ test_53c() {
 run_test 53c "|X| open request and close request while two MDC requests in flight"
 
 test_53d() {
+        cancel_lru_locks mdc    # cleanup locks from former test cases
         rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
 
         mkdir -p $DIR/${tdir}-1
@@ -1259,6 +1266,9 @@ test_53f() {
         kill -USR1 $close_pid
         cancel_lru_locks mdc    # force the close
 
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
         replay_barrier_nodf $SINGLEMDS
         fail_nodf $SINGLEMDS
         wait $open_pid || return 1
@@ -1291,8 +1301,11 @@ test_53g() {
         do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
         kill -USR1 $close_pid
         cancel_lru_locks mdc    # force the close
-
         do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
         replay_barrier_nodf $SINGLEMDS
         fail_nodf $SINGLEMDS
         wait $open_pid || return 1
@@ -1326,6 +1339,9 @@ test_53h() {
         cancel_lru_locks mdc    # force the close
         sleep 1
 
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
         replay_barrier_nodf $SINGLEMDS
         fail_nodf $SINGLEMDS
         wait $open_pid || return 1
@@ -1839,7 +1855,7 @@ test_70b () {
        [ "$SLOW" = "no" ] && duration=60
        local cmd="rundbench 1 -t $duration"
        local PID=""
-       do_nodes --verbose $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
+       do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
                PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
                DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
                LCTL=$LCTL $cmd" &
@@ -2017,8 +2033,8 @@ test_83b() {
 run_test 83b "fail log_add during unlink recovery"
 
 test_84a() {
-#define OBD_FAIL_MDS_OPEN_WAIT_CREATE  0x143
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000143"
+#define OBD_FAIL_MDS_OPEN_WAIT_CREATE  0x144
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
     createmany -o $DIR/$tfile- 1 &
     PID=$!
     mds_evict_client