X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-random-scale.sh;h=384c82b4836c5f4250935e4b5303c41e63acd938;hb=1e4cc6b12d7aa5266597540590a2a182edb41289;hp=061a2b9b83bd2e27019d4dfbfaf66dc31559c344;hpb=0f391e372b759b0f7dac6c924da1b8944532cc62;p=fs%2Flustre-release.git

diff --git a/lustre/tests/recovery-random-scale.sh b/lustre/tests/recovery-random-scale.sh
index 061a2b9..384c82b 100644
--- a/lustre/tests/recovery-random-scale.sh
+++ b/lustre/tests/recovery-random-scale.sh
@@ -19,8 +19,11 @@ init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-random-scale}
+TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
 DEBUGLOG=$TESTSUITELOG.debug
+
+cleanup_logs
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
@@ -112,7 +115,7 @@ summary_and_cleanup () {
     # actually failed though.  the first node in the END_RUN_NODE is
     # the one we are really interested in.
         if [ -n "$END_RUN_NODE" ]; then
-            var=${END_RUN_NODE}_load
+            var=$(client_var_name $END_RUN_NODE)_load
             echo "Client load failed on node $END_RUN_NODE" 
             echo
             echo "client $END_RUN_NODE load stdout and debug files :
@@ -155,13 +158,16 @@ Status: $result: rc=$rc"
 
     if [ $rc -ne 0 ]; then
         print_logs $NODES_TO_USE
+        # we are interested in only on failed clients and servers
+        local failedclients=$(cat $END_RUN_FILE | grep -v $0)
+        # FIXME: need ostfailover-s nodes also for FLAVOR=OST
+        local product=$(gather_logs $(comma_list $(osts_nodes) \
+                                 $mds_HOST $mdsfailover_HOST $failedclients))
+        echo logs files $product
     fi
 
-    if [ $rc -eq 0 ]; then
-        zconf_mount $(hostname) $MOUNT
-    else
-        error "exited with rc=$rc"
-    fi
+    [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
+
     exit $rc
 }
 
@@ -294,6 +300,10 @@ Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
     log " Number of failovers:
 $(numfailovers)                and counting..."
 
+    if [ $((ELAPSED + sleep)) -gt $DURATION ]; then
+         break
+    fi
+
     if [ $sleep -gt 0 ]; then 
         echo "sleeping $sleep seconds ... "
         sleep $sleep