b=1600: my version of Andreas' patch to use write_record to ensure that the

        MDS server data is written at creation time
b=1597: cope with entering recovery/replay without a preceding timeout

We pass this initial single-node/single-mount replay test, yay.

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 3203044..6fe23bd 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -41,7 +41,7 @@ replay_barrier() {
 }
 
 fail() {
-    stop mds
+    stop mds -f --failover
     start mds
     df $MOUNTPT
 }
@@ -67,15 +67,56 @@ gen_config() {
     do_lmc --add mtpt --node client_facet --path $MOUNTPT --mds mds1 --ost ost1
 }
 
+error() {
+    echo '**** FAIL:' $@
+    exit 1
+}
+
+EQUALS="======================================================================"
+
+run_test() {
+    testnum=$1
+    message=$2
+    
+    # Pretty tests run faster.
+    echo '=====' $testnum: $message
+    local suffixlen=$((65 - `echo -n $2 | wc -c | awk '{print $1}'`))
+    printf '%.*s\n' $suffixlen $EQUALS
+
+    test_${testnum} || error "test_$testnum failed with $?"
+}
+
 gen_config
-start mds
-start ost
-start client
+start mds --reformat $MDSLCONFARGS
+start ost --reformat $OSTLCONFARGS
+start client --gdb $CLIENTLCONFARGS
 
-touch $MOUNTPT/lustre-works
-replay_barrier mds
-touch $MOUNTPT/lustre-does-not-work
+test_1() {
+    replay_barrier mds
+    mcreate $MOUNTPT/f1
+    fail
+    rm $MOUNTPT/f1
+}
+run_test 1 "simple create"
+
+test_2() {
+    replay_barrier mds
+    mkdir $MOUNTPT/d2
+    mcreate $MOUNTPT/d2/f2
+    fail
+    rm -fr $MOUNTPT/d2
+}
+run_test 2 "mkdir + contained create"
+
+test_3() {
+    mkdir $MOUNTPT/d3
+    replay_barrier mds
+    mcreate $MOUNTPT/d3/f3
+    fail
+    rm -fr $MOUNTPT/d3
+}
+run_test 3 "mkdir |X| contained create"
 
-stop client
+stop client $CLIENTLCONFARGS
 stop ost
 stop mds