b=23515 recovery_*_scale tests need more than 2 clients

[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 a332acd..0aeb321 100644 (file)
--- a/lustre/tests/recovery-double-scale.sh
+++ b/lustre/tests/recovery-double-scale.sh
@@ -17,17 +17,26 @@ CLEANUP=${CLEANUP:-""}
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-double-scale}
+init_logging
+TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
 DEBUGLOG=$TESTSUITELOG.debug
+
+cleanup_logs
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
 echo "--- env ---" >&2
 set -x
 
-[ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
+[ "$SHARED_DIRECTORY" ] || \
+    { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
+
+[ -n "$CLIENTS" ] || \
+    { FAIL_ON_ERROR=true skip_env "$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}
@@ -37,6 +46,9 @@ 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
@@ -67,17 +79,23 @@ 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 $TMP/client-load.pid && \
+                             { kill -s TERM \$(cat $TMP/client-load.pid) || true; }"
                       shutdown_client $c
                       boot_node $c
                       echo "Reintegrating $c"
-                      zconf_mount $c $MOUNT || return $?
+                      # one client fails; need dk logs from this client only 
+                      zconf_mount $c $MOUNT || NODES="$c $(mdts_nodes) $(osts_nodes)" error_exit "zconf_mount failed"
                  done
-                 start_client_loads $item || return $?
+                 start_client_loads $item
                  ;;
-       * )      error "reboot_recover_node: nodetype=$nodetype. Must be one of 'MDS', 'OST', or 'clients'."
+                # script failure:
+                # don't use error (), the logs from all nodes not needed
+       * )      echo "reboot_recover_node: nodetype=$nodetype. Must be one of 'MDS', 'OST', or 'clients'."
                 exit 1;;
     esac
 }
@@ -92,7 +110,9 @@ get_item_type () {
        OST )    list=$OSTS;;
        clients) list=$NODES_TO_USE
                 ;;
-       * )      error "Invalid type=$type. Must be one of 'MDS', 'OST', or 'clients'."
+                # script failure:
+                # don't use error (), the logs from all nodes not needed
+       * )      echo "Invalid type=$type. Must be one of 'MDS', 'OST', or 'clients'."
                 exit 1;;
     esac
 
@@ -148,7 +168,7 @@ failover_pair() {
 
     log "Done checking client loads. Failing type1=$type1 item1=$item1 ... "
 
-    reboot_recover_node $item1 $type1 || return $?
+    reboot_recover_node $item1 $type1
 
     # Hendrix test17 description: 
     # Introduce a failure, wait at
@@ -163,7 +183,7 @@ failover_pair() {
     # do not need a sleep between failures for "double failures"
 
     log "                            Failing type2=$type2 item2=$item2 ... "    
-    reboot_recover_node $item2 $type2 || return $?
+    reboot_recover_node $item2 $type2
 
     # Client loads are allowed to die while in recovery, so we just
     # restart them.
@@ -189,7 +209,7 @@ 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 debug output :"
@@ -219,6 +239,16 @@ 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))
+        echo logs files $product
+    fi
+
     [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
     exit $rc
 }
@@ -245,12 +275,8 @@ FAILOVER_PERIOD=${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
+if ! do_nodesv $NODES_TO_USE "cat $TMP/client-load.pid"; then
         exit 3
-    fi
 fi
 
 # FIXME: Do we want to have an initial sleep period where the clients