Branch HEAD

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

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 9f04ca5..4ccfe84 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -460,6 +460,8 @@ test_20a() {        # was test_20
 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
 
 test_20b() { # bug 10480
+    # XXX increase the debug level temporary 
+    do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=0x33f0406; $LCTL set_param debug_mb=150"
     BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
 
     dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
@@ -484,6 +486,8 @@ test_20b() { # bug 10480
     log "before $BEFOREUSED, after $AFTERUSED"
     [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
         error "after $AFTERUSED > before $BEFOREUSED"
+    # XXX decrease it back
+    do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=$PTLDEBUG; $LCTL set_param debug_mb=$DEBUG_SIZE"
     return 0
 }
 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
@@ -1826,7 +1830,7 @@ test_70b () {
 
        zconf_mount_clients $clients $DIR
        
-       local duration=120
+       local duration=300
        [ "$SLOW" = "no" ] && duration=60
        local cmd="rundbench 1 -t $duration"
        local PID=""
@@ -1836,14 +1840,21 @@ test_70b () {
                LCTL=$LCTL $cmd" &
        PID=$!
        log "Started rundbench load PID=$PID ..."
-
-       sleep $((duration / 4))
-       replay_barrier $SINGLEMDS
-       sleep 3 # give clients a time to do operations
-
-       log "$TESTNAME fail mds 1"
-       fail $SINGLEMDS
-
+       ELAPSED=0
+       NUM_FAILOVERS=0
+       START_TS=$(date +%s)
+       CURRENT_TS=$START_TS
+       while [ $ELAPSED -lt $duration ]; do
+               sleep 1
+               replay_barrier $SINGLEMDS
+               sleep 1 # give clients a time to do operations
+               # Increment the number of failovers
+               NUM_FAILOVERS=$((NUM_FAILOVERS+1))
+               log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
+               facet_failover $SINGLEMDS
+               CURRENT_TS=$(date +%s)
+               ELAPSED=$((CURRENT_TS - START_TS))
+       done
        wait $PID || error "rundbench load on $CLIENTS failed!"
 }
 run_test 70b "mds recovery; $CLIENTCOUNT clients"