bug#28445 - Heartbeat does not start until first API_REGREQ is recevied

- move api failure handling into own method
- add START_ORD so that hb checking can start really early

storage/ndb/src/kernel/blocks/cmvmi/Cmvmi.cpp

@@ -421,9 +421,10 @@ void Cmvmi::execCLOSE_COMREQ(Signal* signal)
 //  Uint32 noOfNodes = closeCom->noOfNodes;
   
   jamEntry();
-  for (unsigned i = 0; i < MAX_NODES; i++){
-    if(NodeBitmask::get(closeCom->theNodes, i)){
-    
+  for (unsigned i = 0; i < MAX_NODES; i++)
+  {
+    if(NodeBitmask::get(closeCom->theNodes, i))
+    {
       jam();
 
       //-----------------------------------------------------
@@ -437,7 +438,9 @@ void Cmvmi::execCLOSE_COMREQ(Signal* signal)
       globalTransporterRegistry.do_disconnect(i);
     }
   }
-  if (failNo != 0) {
+
+  if (failNo != 0) 
+  {
     jam();
     signal->theData[0] = userRef;
     signal->theData[1] = failNo;
@@ -456,13 +459,21 @@ void Cmvmi::execOPEN_COMREQ(Signal* signal)
   jamEntry();
 
   const Uint32 len = signal->getLength();
-  if(len == 2){
-
+  if(len == 2)
+  {
 #ifdef ERROR_INSERT
     if (! ((ERROR_INSERTED(9000) || ERROR_INSERTED(9002)) 
 	   && c_error_9000_nodes_mask.get(tStartingNode)))
 #endif
     {
+      if (globalData.theStartLevel != NodeState::SL_STARTED &&
+          (getNodeInfo(tStartingNode).m_type != NodeInfo::DB &&
+           getNodeInfo(tStartingNode).m_type != NodeInfo::MGM))
+      {
+        jam();
+        goto done;
+      }
+
       globalTransporterRegistry.do_connect(tStartingNode);
       globalTransporterRegistry.setIOState(tStartingNode, HaltIO);
       
@@ -475,9 +486,11 @@ void Cmvmi::execOPEN_COMREQ(Signal* signal)
       //-----------------------------------------------------
     }
   } else {
-    for(unsigned int i = 1; i < MAX_NODES; i++ ) {
+    for(unsigned int i = 1; i < MAX_NODES; i++ ) 
+    {
       jam();
-      if (i != getOwnNodeId() && getNodeInfo(i).m_type == tData2){
+      if (i != getOwnNodeId() && getNodeInfo(i).m_type == tData2)
+      {
 	jam();
 
 #ifdef ERROR_INSERT
@@ -496,6 +509,7 @@ void Cmvmi::execOPEN_COMREQ(Signal* signal)
     }
   }
   
+done:  
   if (userRef != 0) {
     jam(); 
     signal->theData[0] = tStartingNode;
@@ -537,22 +551,8 @@ void Cmvmi::execDISCONNECT_REP(Signal *signal)
   const NodeInfo::NodeType type = getNodeInfo(hostId).getType();
   ndbrequire(type != NodeInfo::INVALID);
 
-  if(type == NodeInfo::DB || globalData.theStartLevel == NodeState::SL_STARTED){
-    jam();
-    DisconnectRep * const rep = (DisconnectRep *)&signal->theData[0];
-    rep->nodeId = hostId;
-    rep->err = errNo;
   sendSignal(QMGR_REF, GSN_DISCONNECT_REP, signal, 
              DisconnectRep::SignalLength, JBA);
-  } else if((globalData.theStartLevel == NodeState::SL_CMVMI ||
-	     globalData.theStartLevel == NodeState::SL_STARTING)
-	    && type == NodeInfo::MGM) {
-    /**
-     * Someone disconnected during cmvmi period
-     */
-    jam();
-    globalTransporterRegistry.do_connect(hostId);
-  }
   
   cancelSubscription(hostId);
 
@@ -587,6 +587,8 @@ void Cmvmi::execCONNECT_REP(Signal *signal){
      */
     if(type == NodeInfo::MGM){
       jam();
+      signal->theData[0] = hostId;
+      sendSignal(QMGR_REF, GSN_CONNECT_REP, signal, 1, JBA);
     } else {
       /**
        * Dont allow api nodes to connect
@@ -802,6 +804,8 @@ Cmvmi::execSTART_ORD(Signal* signal) {
         }
       }
     }
+
+    EXECUTE_DIRECT(QMGR, GSN_START_ORD, signal, 1);
     return ;
   }
   
@@ -830,9 +834,6 @@ Cmvmi::execSTART_ORD(Signal* signal) {
      * Do Restart
      */
     
-    globalScheduler.clear();
-    globalTimeQueue.clear();
-    
     // Disconnect all nodes as part of the system restart. 
     // We need to ensure that we are starting up
     // without any connected nodes.   

storage/ndb/src/kernel/blocks/qmgr/Qmgr.hpp

@@ -266,6 +266,8 @@ private:
   void execALLOC_NODEID_REF(Signal *);
   void completeAllocNodeIdReq(Signal *);
   
+  void execSTART_ORD(Signal*);
+
   // Arbitration signals
   void execARBIT_CFG(Signal* signal);
   void execARBIT_PREPREQ(Signal* signal);
@@ -281,6 +283,7 @@ private:
   void check_readnodes_reply(Signal* signal, Uint32 nodeId, Uint32 gsn);
   Uint32 check_startup(Signal* signal);
 
+  void api_failed(Signal* signal, Uint32 aFailedNode);
   void node_failed(Signal* signal, Uint16 aFailedNode);
   void checkStartInterface(Signal* signal);
   void failReport(Signal* signal,

storage/ndb/src/kernel/blocks/qmgr/QmgrInit.cpp

@@ -31,10 +31,6 @@ void Qmgr::initData()
   cnoCommitFailedNodes = 0;
   c_maxDynamicId = 0;
   c_clusterNodes.clear();
-
-  Uint32 hbDBAPI = 500;
-  setHbApiDelay(hbDBAPI);
-  c_connectedNodes.set(getOwnNodeId());
   c_stopReq.senderRef = 0;
 
   /**
@@ -43,6 +39,27 @@ void Qmgr::initData()
   ndbrequire((Uint32)NodeInfo::DB == 0);
   ndbrequire((Uint32)NodeInfo::API == 1);
   ndbrequire((Uint32)NodeInfo::MGM == 2); 
+
+  NodeRecPtr nodePtr;
+  nodePtr.i = getOwnNodeId();
+  ptrAss(nodePtr, nodeRec);
+  nodePtr.p->blockRef = reference();
+
+  c_connectedNodes.set(getOwnNodeId());
+  setNodeInfo(getOwnNodeId()).m_version = NDB_VERSION;
+
+
+  /**
+   * Timeouts
+   */
+  const ndb_mgm_configuration_iterator * p = 
+    m_ctx.m_config.getOwnConfigIterator();
+  ndbrequire(p != 0);
+  
+  Uint32 hbDBAPI = 1500;
+  ndb_mgm_get_int_parameter(p, CFG_DB_API_HEARTBEAT_INTERVAL, &hbDBAPI);
+  
+  setHbApiDelay(hbDBAPI);
 }//Qmgr::initData()
 
 void Qmgr::initRecords() 
@@ -113,6 +130,7 @@ Qmgr::Qmgr(Block_context& ctx)
   addRecSignal(GSN_DIH_RESTARTREF, &Qmgr::execDIH_RESTARTREF);
   addRecSignal(GSN_DIH_RESTARTCONF, &Qmgr::execDIH_RESTARTCONF);
   addRecSignal(GSN_NODE_VERSION_REP, &Qmgr::execNODE_VERSION_REP);
+  addRecSignal(GSN_START_ORD, &Qmgr::execSTART_ORD);
   
   initData();
 }//Qmgr::Qmgr()