ndb: new arbitrator behaviour for >=3-way: < 1/2 nodes can survive

ndb/include/kernel/signaldata/ArbitSignalData.hpp

@@ -94,13 +94,14 @@ public:
 
     // arbitration result
     LoseNodes = 41,             // lose on ndb node count
-    WinGroups = 42,             // we win, no need for arbitration
-    LoseGroups = 43,            // we lose, missing node group
-    Partitioning = 44,          // possible network partitioning
-    WinChoose = 45,             // positive reply
-    LoseChoose = 46,            // negative reply
-    LoseNorun = 47,             // arbitrator required but not running
-    LoseNocfg = 48,             // arbitrator required but none configured
+    WinNodes = 42,              // win on ndb node count
+    WinGroups = 43,             // we win, no need for arbitration
+    LoseGroups = 44,            // we lose, missing node group
+    Partitioning = 45,          // possible network partitioning
+    WinChoose = 46,             // positive reply
+    LoseChoose = 47,            // negative reply
+    LoseNorun = 48,             // arbitrator required but not running
+    LoseNocfg = 49,             // arbitrator required but none configured
 
     // general error codes
     ErrTicket = 91,             // invalid arbitrator-ticket

ndb/src/common/debugger/EventLogger.cpp

@@ -421,6 +421,11 @@ EventLogger::getText(char * m_text, size_t m_text_len,
 	  "%sArbitration check lost - less than 1/2 nodes left",
 	  theNodeId);
 	break;
+      case ArbitCode::WinNodes:
+	BaseString::snprintf(m_text, m_text_len,
+	  "%sArbitration check won - all node groups and more than 1/2 nodes left",
+	  theNodeId);
+	break;
       case ArbitCode::WinGroups:
 	BaseString::snprintf(m_text, m_text_len,
 	  "%sArbitration check won - node group majority",

ndb/src/kernel/blocks/qmgr/QmgrMain.cpp

@@ -2946,6 +2946,12 @@ void Qmgr::sendPrepFailReq(Signal* signal, Uint16 aNode)
  * the "handle" routines.
  */
 
+/**
+ * Should < 1/2 nodes die unconditionally.  Affects only >= 3-way
+ * replication.
+ */
+static const bool g_ndb_arbit_one_half_rule = false;
+
 /**
  * Config signals are logically part of CM_INIT.
  */
@@ -3157,7 +3163,8 @@ Qmgr::handleArbitCheck(Signal* signal)
   ndbrequire(cpresident == getOwnNodeId());
   NodeBitmask ndbMask;
   computeArbitNdbMask(ndbMask);
-  if (2 * ndbMask.count() < cnoOfNodes) {
+  if (g_ndb_arbit_one_half_rule &&
+      2 * ndbMask.count() < cnoOfNodes) {
     jam();
     arbitRec.code = ArbitCode::LoseNodes;
   } else {
@@ -3181,6 +3188,11 @@ Qmgr::handleArbitCheck(Signal* signal)
     case CheckNodeGroups::Partitioning:
       jam();
       arbitRec.code = ArbitCode::Partitioning;
+      if (g_ndb_arbit_one_half_rule &&
+          2 * ndbMask.count() > cnoOfNodes) {
+        jam();
+        arbitRec.code = ArbitCode::WinNodes;
+      }
       break;
     default:
       ndbrequire(false);
@@ -3190,8 +3202,12 @@ Qmgr::handleArbitCheck(Signal* signal)
   switch (arbitRec.code) {
   case ArbitCode::LoseNodes:
     jam();
+  case ArbitCode::LoseGroups:
+    jam();
     goto crashme;
-  case ArbitCode::WinGroups:  
+  case ArbitCode::WinNodes:
+    jam();
+  case ArbitCode::WinGroups:
     jam();
     if (arbitRec.state == ARBIT_RUN) {
       jam();
@@ -3200,9 +3216,6 @@ Qmgr::handleArbitCheck(Signal* signal)
     arbitRec.state = ARBIT_INIT;
     arbitRec.newstate = true;
     break;
-  case ArbitCode::LoseGroups:
-    jam();
-    goto crashme;
   case ArbitCode::Partitioning:
     if (arbitRec.state == ARBIT_RUN) {
       jam();
@@ -3762,8 +3775,7 @@ Qmgr::execARBIT_CHOOSEREF(Signal* signal)
 }
 
 /**
- * Handle CRASH state.  We must crash immediately.  But it
- * would be nice to wait until event reports have been sent.
+ * Handle CRASH state.  We must crash immediately.
  * XXX tell other nodes in our party to crash too.
  */
 void
@@ -3773,12 +3785,11 @@ Qmgr::stateArbitCrash(Signal* signal)
   if (arbitRec.newstate) {
     jam();
     CRASH_INSERTION((Uint32)910 + arbitRec.state);
-
     arbitRec.setTimestamp();
     arbitRec.code = 0;
     arbitRec.newstate = false;
   }
-#if 0
+#ifdef ndb_arbit_crash_wait_for_event_report_to_get_out
   if (! (arbitRec.getTimediff() > getArbitTimeout()))
     return;
 #endif