b=20918 report max recovery time estimated

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

diff --git a/lustre/tests/recovery-random-scale.sh b/lustre/tests/recovery-random-scale.sh
index dfde69a..4f8e879 100644 (file)
--- a/lustre/tests/recovery-random-scale.sh
+++ b/lustre/tests/recovery-random-scale.sh
@@ -18,9 +18,13 @@ CLEANUP=${CLEANUP:-""}
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 
-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
@@ -44,6 +48,8 @@ build_test_filter
 check_and_setup_lustre
 rm -rf $DIR/[df][0-9]*
 
+max_recov_time=$(max_recovery_time)
+
 # the test node needs to be insulated from a lustre failure as much as possible,
 # so not even loading the lustre modules is ideal.
 # -- umount lustre
@@ -128,7 +134,7 @@ summary_and_cleanup () {
     local result=PASS
     [ $rc -eq 0 ] || result=FAIL
 
-    log "Duraion:                $DURATION
+    log "Duration:                $DURATION
 Server failover period: $SERVER_FAILOVER_PERIOD seconds
 Exited after:           $ELAPSED seconds
 Number of failovers before exit:
@@ -155,13 +161,16 @@ 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) \
+                                 $(mdts_nodes) $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
 }
 
@@ -183,12 +192,8 @@ fi
 start_client_loads $NODES_TO_USE
 
 echo clients load pids:
-if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $LOAD_PID_FILE"; then
-    if [ -e $DEBUGLOG ]; then
-        exec 2<&-
-        cat $DEBUGLOG
+if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
         exit 3
-    fi
 fi
 
 START_TS=$(date +%s)
@@ -285,15 +290,22 @@ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
     # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
     if [ $sleep -lt $MINSLEEP ]; then
         reqfail=$((reqfail +1))
-        log "WARNING: failover, client reintegration and check_client_loads time
-exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
-Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
+        log "WARNING: failover, client reintegration and check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
+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.
+Estimated max recovery time : $max_recov_time
+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 " 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