LU-734 tests: save recovery-*-scale debug logs into $LOGDIR

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

diff --git a/lustre/tests/recovery-double-scale.sh b/lustre/tests/recovery-double-scale.sh
index 81757e2..805fddc 100644 (file)
--- a/lustre/tests/recovery-double-scale.sh
+++ b/lustre/tests/recovery-double-scale.sh
@@ -17,8 +17,10 @@ CLEANUP=${CLEANUP:-""}
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-double-scale}
-DEBUGLOG=$TESTSUITELOG.debug
+init_logging
+
+DEBUGLOG=$TESTLOG_PREFIX.suite_debug_log.$(hostname -s).log
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
@@ -26,11 +28,16 @@ echo "--- env ---" >&2
 set -x
 
 [ "$SHARED_DIRECTORY" ] || \
-    { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
+    { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
+
+check_shared_dir $SHARED_DIRECTORY ||
+    error "$SHARED_DIRECTORY isn't a shared directory"
+
+[ -n "$CLIENTS" ] || \
+    { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients" && exit 0; }
 
-[ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
 [ $CLIENTCOUNT -ge 3 ] || \
-    { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
+    { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients, have $((CLIENTCOUNT - 1))" && exit 0; }
 
 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
@@ -40,6 +47,9 @@ remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
 
 check_timeout || exit 1
 
+[[ $FAILURE_MODE = SOFT ]] && \
+    log "WARNING: $0 is not functional with FAILURE_MODE = SOFT, bz22797"
+
 build_test_filter
 
 check_and_setup_lustre
@@ -70,9 +80,13 @@ reboot_recover_node () {
     # MDS, OST item contains the facet
     case $nodetype in
        MDS|OST )    facet_failover $item
-                [ "$SERIAL" ] && wait_recovery_complete $item $((timeout * 4)) || true
+                [ "$SERIAL" ] && wait_recovery_complete $item || true
                 ;;
        clients) for c in ${item//,/ }; do
+                      # make sure the client loads die
+                      do_nodes $c "set -x; test -f $LOAD_PID_FILE &&
+                          { kill -s TERM \\\$(cat $LOAD_PID_FILE);
+                          rm -f $LOAD_PID_FILE || true; }"
                       shutdown_client $c
                       boot_node $c
                       echo "Reintegrating $c"
@@ -183,6 +197,7 @@ failover_pair() {
 
 summary_and_cleanup () {
     local rc=$?
+    local var
     trap 0
 
     # Having not empty END_RUN_FILE means the failed loads only
@@ -192,17 +207,16 @@ summary_and_cleanup () {
         local END_RUN_NODE=
         read END_RUN_NODE < $END_RUN_FILE
 
-        # a client load will end (i.e. fail) if it finds
-        # the end run file.  that does not mean that that client load
-        # actually failed though.  the first node in the END_RUN_NODE is
-        # the one we are really interested in.
+        # A client load will stop if it found the END_RUN_FILE file.
+        # That does not mean the client load actually failed though.
+        # The first node in END_RUN_FILE is the one we are interested in.
         if [ -n "$END_RUN_NODE" ]; then
-            var=$(client_var_name $END_RUN_NODE)_load
+            var=$(node_var_name $END_RUN_NODE)_load
             echo "Client load failed on node $END_RUN_NODE"
             echo
-            echo "client $END_RUN_NODE load debug output :"
-            local logfile=${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug 
-            do_node ${END_RUN_NODE} "set -x; [ -e $logfile ] && cat $logfile " || true
+            echo "Client $END_RUN_NODE load stdout and debug files:
+                $TESTLOG_PREFIX.run_${!var}_stdout.$END_RUN_NODE.log
+                $TESTLOG_PREFIX.run_${!var}_debug.$END_RUN_NODE.log"
         fi
         rc=1
     fi
@@ -218,8 +232,9 @@ Exited after:           $ELAPSED seconds
 Status: $result: rc=$rc"
 
     # make sure the client loads die
-    do_nodes $NODES_TO_USE "set -x; test -f $TMP/client-load.pid && \
-        { kill -s TERM \$(cat $TMP/client-load.pid) || true; }"
+    do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE &&
+        { kill -s TERM \\\$(cat $LOAD_PID_FILE);
+        rm -f $LOAD_PID_FILE || true; }"
 
     # and free up the pdshes that started them, if any are still around
     if [ -n "$CLIENT_LOAD_PIDS" ]; then
@@ -227,6 +242,16 @@ Status: $result: rc=$rc"
         sleep 5
         kill -9 $CLIENT_LOAD_PIDS || true
     fi
+
+    if [ $rc -ne 0 ]; then
+        # we are interested in only on failed clients and servers
+        local failedclients=$(cat $END_RUN_FILE | grep -v $0)
+        # FIXME: need ostfailover-s nodes also for FLAVOR=OST
+        local product=$(gather_logs $(comma_list $(osts_nodes) \
+                        $(mdts_nodes) $mdsfailover_HOST $failedclients) 1)
+        echo $product
+    fi
+
     [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
     exit $rc
 }
@@ -252,13 +277,10 @@ FAILOVER_PERIOD=${FAILOVER_PERIOD:-$((60*5))} # 5 minutes
 
 # Start client loads.
 start_client_loads $NODES_TO_USE
+
 echo clients load pids:
-if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $TMP/client-load.pid"; then
-    if [ -e $DEBUGLOG ]; then
-        exec 2<&-
-        cat $DEBUGLOG
-        exit 3
-    fi
+if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
+    exit 3
 fi
 
 # FIXME: Do we want to have an initial sleep period where the clients