b=20918 improve log warning

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

diff --git a/lustre/tests/recovery-mds-scale.sh b/lustre/tests/recovery-mds-scale.sh
index 598620b..0ab9999 100644 (file)
--- a/lustre/tests/recovery-mds-scale.sh
+++ b/lustre/tests/recovery-mds-scale.sh
@@ -14,9 +14,13 @@ CLEANUP=${CLEANUP:-""}
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
+TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
 DEBUGLOG=$TESTSUITELOG.debug
+
+cleanup_logs
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
@@ -28,9 +32,9 @@ set -x
 
 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
 [ $CLIENTCOUNT -ge 3 ] || \
-    { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
+    { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
 
-END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY}/end_run_file}
+END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
 
 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
@@ -106,7 +110,7 @@ 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 :
@@ -145,6 +149,15 @@ 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))
+        echo logs files $product
+    fi
+
     [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
 
     exit $rc
@@ -207,6 +220,19 @@ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
         exit 4
     fi
 
+    log "Wait $SERVERFACET recovery complete before doing next failover ...."
+    if [[ $NUM_FAILOVERS != 0 ]]; then
+        if ! wait_recovery_complete $SERVERFACET ; then
+            echo "$SERVERFACET recovery is not completed!"
+            exit 7
+        fi
+    fi
+
+    log "Checking clients are in FULL state before doing next failover"
+    if ! wait_clients_import_state $NODES_TO_USE $SERVERFACET FULL; then
+        echo "Clients import not FULL, please consider to increase SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD !"
+        
+    fi
     log "Starting failover on $SERVERFACET"
 
     facet_failover "$SERVERFACET" || exit 1
@@ -236,11 +262,19 @@ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
     if [ $sleep -lt $MINSLEEP ]; then
         reqfail=$((reqfail +1))
         log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
-Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
+Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
+This iteration, the load was only applied for sleep=$sleep seconds.
+Probably the hardware is taking excessively long to boot.
+Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
         [ $reqfail -gt $REQFAIL ] && exit 6 
     fi  
 
     log "$SERVERFACET has failed over ${!var} times, and counting..."
+
+    if [ $((ELAPSED + sleep)) -gt $DURATION ]; then
+         break
+    fi
+
     if [ $sleep -gt 0 ]; then 
         echo "sleeping $sleep seconds ... "
         sleep $sleep