Merge joreland@bk-internal.mysql.com:/home/bk/mysql-5.1-new

into  perch.ndb.mysql.com:/home/jonas/src/mysql-5.1-new

storage/ndb/include/kernel/signaldata/CmRegSignalData.hpp

@@ -30,12 +30,17 @@ class CmRegReq {
   friend class Qmgr;
   
 public:
-  STATIC_CONST( SignalLength = 3 );
+  STATIC_CONST( SignalLength = 5 + NdbNodeBitmask::Size );
 private:
   
   Uint32 blockRef;
   Uint32 nodeId;
-  Uint32 version; // See ndb_version.h
+  Uint32 version;    // See ndb_version.h
+
+  Uint32 start_type; // As specified by cmd-line or mgm, NodeState::StartType
+  Uint32 latest_gci; // 0 means no fs
+  Uint32 skip_nodes[NdbNodeBitmask::Size]; // Nodes that does not _need_ 
+                                           // to be part of restart
 };
 
 /**
@@ -59,8 +64,7 @@ private:
    * The dynamic id that the node reciving this signal has
    */
   Uint32 dynamicId;
-  
+  Uint32 allNdbNodes[NdbNodeBitmask::Size];  
-  Uint32 allNdbNodes[NdbNodeBitmask::Size];
 };
 
 /**
@@ -73,7 +77,7 @@ class CmRegRef {
   friend class Qmgr;
   
 public:
-  STATIC_CONST( SignalLength = 4 );
+  STATIC_CONST( SignalLength = 7 + NdbNodeBitmask::Size );
   
   enum ErrorCode {
     ZBUSY = 0,          /* Only the president can send this */
@@ -85,14 +89,27 @@ public:
                          * as president. */
     ZNOT_PRESIDENT = 5, /* We are not president */
     ZNOT_DEAD = 6,       /* We are not dead when we are starting  */
-    ZINCOMPATIBLE_VERSION = 7
+    ZINCOMPATIBLE_VERSION = 7,
+    ZINCOMPATIBLE_START_TYPE = 8
   };
 private:
   
   Uint32 blockRef;
   Uint32 nodeId;
   Uint32 errorCode;
+  /**
+   * Applicable if ZELECTION
+   */
   Uint32 presidentCandidate;
+  Uint32 candidate_latest_gci; // 0 means non
+
+  /**
+   * Data for sending node sending node
+   */
+  Uint32 latest_gci; 
+  Uint32 start_type; 
+  Uint32 skip_nodes[NdbNodeBitmask::Size]; // Nodes that does not _need_ 
+                                           // to be part of restart
 };
 
 class CmAdd {

storage/ndb/include/kernel/signaldata/DumpStateOrd.hpp

@@ -64,6 +64,7 @@ public:
     // 19 NDBFS Fipple with O_SYNC, O_CREATE etc.
     // 20-24 BACKUP
     NdbcntrTestStopOnError = 25,
+    NdbcntrStopNodes = 70,
     // 100-105 TUP and ACC  
     // 200-240 UTIL
     // 300-305 TRIX

storage/ndb/include/kernel/signaldata/FailRep.hpp

@@ -18,6 +18,7 @@
 #define FAIL_REP_HPP
 
 #include "SignalData.hpp"
+#include <NodeBitmask.hpp>
 
 /**
  * 
@@ -27,6 +28,7 @@ class FailRep {
    * Sender(s) & Reciver(s)
    */
   friend class Qmgr;
+  friend class Ndbcntr;
   
   /**
    * For printing
@@ -35,7 +37,8 @@ class FailRep {
 
 public:
   STATIC_CONST( SignalLength = 2 );
-
+  STATIC_CONST( ExtraLength = 1 + NdbNodeBitmask::Size );
+  
   enum FailCause {
     ZOWN_FAILURE=0,
     ZOTHER_NODE_WHEN_WE_START=1,
@@ -43,13 +46,20 @@ public:
     ZSTART_IN_REGREQ=3,
     ZHEARTBEAT_FAILURE=4,
     ZLINK_FAILURE=5,
-    ZOTHERNODE_FAILED_DURING_START=6
+    ZOTHERNODE_FAILED_DURING_START=6,
+    ZMULTI_NODE_SHUTDOWN = 7,
+    ZPARTITIONED_CLUSTER = 8
   };
-
+  
 private:
   
   Uint32 failNodeId;
   Uint32 failCause;
+  /**
+   * Used when failCause == ZPARTITIONED_CLUSTER
+   */
+  Uint32 president;
+  Uint32 partition[NdbNodeBitmask::Size];
 };
 
 

storage/ndb/include/kernel/signaldata/StopReq.hpp

@@ -32,7 +32,7 @@ class StopReq
   friend class MgmtSrvr;
 
 public:
-  STATIC_CONST( SignalLength = 9 );
+  STATIC_CONST( SignalLength = 9 + NdbNodeBitmask::Size);
   
 public:
   Uint32 senderRef;
@@ -49,29 +49,34 @@ public:
   Int32 readOperationTimeout; // Timeout before read operations are aborted
   Int32 operationTimeout;     // Timeout before all operations are aborted
 
+  Uint32 nodes[NdbNodeBitmask::Size];
+
   static void setSystemStop(Uint32 & requestInfo, bool value);
   static void setPerformRestart(Uint32 & requestInfo, bool value);
   static void setNoStart(Uint32 & requestInfo, bool value);
   static void setInitialStart(Uint32 & requestInfo, bool value);
-  static void setEscalateOnNodeFail(Uint32 & requestInfo, bool value);
   /**
    * Don't perform "graceful" shutdown/restart...
    */
   static void setStopAbort(Uint32 & requestInfo, bool value);
+  static void setStopNodes(Uint32 & requestInfo, bool value);
 
   static bool getSystemStop(const Uint32 & requestInfo);
   static bool getPerformRestart(const Uint32 & requestInfo);
   static bool getNoStart(const Uint32 & requestInfo);
   static bool getInitialStart(const Uint32 & requestInfo);
-  static bool getEscalateOnNodeFail(const Uint32 & requestInfo);
   static bool getStopAbort(const Uint32 & requestInfo);
+  static bool getStopNodes(const Uint32 & requestInfo);
 };
 
 struct StopConf
 {
   STATIC_CONST( SignalLength = 2 );
   Uint32 senderData;
-  Uint32 nodeState;
+  union {
+    Uint32 nodeState;
+    Uint32 nodeId;
+  };
 };
 
 class StopRef 
@@ -87,19 +92,22 @@ class StopRef
   friend class Ndbcntr;
 
 public:
-  STATIC_CONST( SignalLength = 2 );
+  STATIC_CONST( SignalLength = 3 );
   
   enum ErrorCode {
     OK = 0,
     NodeShutdownInProgress = 1,
     SystemShutdownInProgress = 2,
     NodeShutdownWouldCauseSystemCrash = 3,
-    TransactionAbortFailed = 4
+    TransactionAbortFailed = 4,
+    UnsupportedNodeShutdown = 5,
+    MultiNodeShutdownNotMaster = 6
   };
   
 public:
   Uint32 senderData;
   Uint32 errorCode;
+  Uint32 masterNodeId;
 };
 
 inline
@@ -132,16 +140,16 @@ StopReq::getInitialStart(const Uint32 & requestInfo)
 
 inline
 bool
-StopReq::getEscalateOnNodeFail(const Uint32 & requestInfo)
+StopReq::getStopAbort(const Uint32 & requestInfo)
 {
-  return requestInfo & 16;
+  return requestInfo & 32;
 }
 
 inline
 bool
-StopReq::getStopAbort(const Uint32 & requestInfo)
+StopReq::getStopNodes(const Uint32 & requestInfo)
 {
-  return requestInfo & 32;
+  return requestInfo & 64;
 }
 
 
@@ -185,16 +193,6 @@ StopReq::setInitialStart(Uint32 & requestInfo, bool value)
     requestInfo &= ~8;
 }
 
-inline
-void
-StopReq::setEscalateOnNodeFail(Uint32 & requestInfo, bool value)
-{
-  if(value)
-    requestInfo |= 16;
-  else
-    requestInfo &= ~16;
-}
-
 inline
 void
 StopReq::setStopAbort(Uint32 & requestInfo, bool value)
@@ -205,6 +203,15 @@ StopReq::setStopAbort(Uint32 & requestInfo, bool value)
     requestInfo &= ~32;
 }
 
+inline
+void
+StopReq::setStopNodes(Uint32 & requestInfo, bool value)
+{
+  if(value)
+    requestInfo |= 64;
+  else
+    requestInfo &= ~64;
+}
 
 #endif
 

storage/ndb/include/kernel/signaldata/WaitGCP.hpp

@@ -46,7 +46,9 @@ public:
     Complete = 1,           ///< Wait for a GCP to complete
     CompleteForceStart = 2, ///< Wait for a GCP to complete start one if needed
     CompleteIfRunning = 3,  ///< Wait for ongoing GCP
-    CurrentGCI        = 8   ///< Immediately return current GCI
+    CurrentGCI        = 8,  ///< Immediately return current GCI
+    BlockStartGcp     = 9,
+    UnblockStartGcp   = 10
   };
 
   Uint32 senderRef;
@@ -70,11 +72,12 @@ class WaitGCPConf {
   //friend class Grep::PSCoord;
 
 public:
-  STATIC_CONST( SignalLength = 2 );
+  STATIC_CONST( SignalLength = 3 );
   
 public:
   Uint32 senderData;
   Uint32 gcp;
+  Uint32 blockStatus;
 };
 
 class WaitGCPRef {

storage/ndb/include/mgmapi/ndb_logevent.h

@@ -169,9 +169,13 @@ extern "C" {
     NDB_LE_BackupAborted = 57,
 
     /** NDB_MGM_EVENT_CATEGORY_INFO */
-    NDB_LE_EventBufferStatus = 58
+    NDB_LE_EventBufferStatus = 58,
 
     /* 59 used */
+
+    /** NDB_MGM_EVENT_CATEGORY_STARTUP */
+    NDB_LE_StartReport = 60
+
     /* 60 unused */
     /* 61 unused */
     /* 62 unused */
@@ -637,6 +641,13 @@ extern "C" {
         unsigned type;
         unsigned node_id;
       } SingleUser;
+      /** Log even data @ref NDB_LE_StartReport */
+      struct {
+	unsigned report_type;
+	unsigned remaining_time;
+	unsigned bitmask_size;
+	unsigned bitmask_data[1];
+      } StartReport;
 #ifndef DOXYGEN_FIX
     };
 #else

storage/ndb/src/common/debugger/EventLogger.cpp

@@ -743,6 +743,90 @@ void getTextSingleUser(QQQQ) {
   }
 }
 
+void getTextStartReport(QQQQ) {
+  Uint32 time = theData[2];
+  Uint32 sz = theData[3];
+  char mask1[100];
+  char mask2[100];
+  char mask3[100];
+  char mask4[100];
+  BitmaskImpl::getText(sz, theData + 4 + (0 * sz), mask1);
+  BitmaskImpl::getText(sz, theData + 4 + (1 * sz), mask2);
+  BitmaskImpl::getText(sz, theData + 4 + (2 * sz), mask3);
+  BitmaskImpl::getText(sz, theData + 4 + (3 * sz), mask4);
+  switch(theData[1]){
+  case 1: // Wait initial
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Initial start, waiting for %s to connect, "
+       " nodes [ all: %s connected: %s no-wait: %s ]",
+       mask4, mask1, mask2, mask3);
+    break;
+  case 2: // Wait partial
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Waiting until nodes: %s connects, "
+       "nodes [ all: %s connected: %s no-wait: %s ]",
+       mask4, mask1, mask2, mask3);
+    break;
+  case 3: // Wait partial timeout
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Waiting %u sec for nodes %s to connect, "
+       "nodes [ all: %s connected: %s no-wait: %s ]",
+       
+       time, mask4, mask1, mask2, mask3);
+    break;
+  case 4: // Wait partioned
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Waiting for non partitioned start, "
+       "nodes [ all: %s connected: %s missing: %s no-wait: %s ]",
+       
+       mask1, mask2, mask4, mask3);
+    break;
+  case 5:
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Waiting %u sec for non partitioned start, "
+       "nodes [ all: %s connected: %s missing: %s no-wait: %s ]",
+       
+       time, mask1, mask2, mask4, mask3);
+    break;
+  case 0x8000: // Do initial
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Initial start with nodes %s [ missing: %s no-wait: %s ]",
+       mask2, mask4, mask3);
+    break;
+  case 0x8001: // Do start
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Start with all nodes %s",
+       mask2);
+    break;
+  case 0x8002: // Do partial
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Start with nodes %s [ missing: %s no-wait: %s ]",
+       mask2, mask4, mask3);
+    break;
+  case 0x8003: // Do partioned
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Start potentially partitioned with nodes %s "
+       " [ missing: %s no-wait: %s ]",
+       mask2, mask4, mask3);
+    break;
+  default:
+    BaseString::snprintf
+      (m_text, m_text_len,
+       "Unknown startreport: 0x%x [ %s %s %s %s ]", 
+       theData[1],
+       mask1, mask2, mask3, mask4);
+  }
+}
+
 #if 0
 BaseString::snprintf(m_text, 
 		     m_text_len, 
@@ -791,6 +875,7 @@ const EventLoggerBase::EventRepLogLevelMatrix EventLoggerBase::matrix[] = {
   ROW(StartREDOLog,            LogLevel::llStartUp,    10, Logger::LL_INFO ),
   ROW(StartLog,                LogLevel::llStartUp,    10, Logger::LL_INFO ),
   ROW(UNDORecordsExecuted,     LogLevel::llStartUp,    15, Logger::LL_INFO ),
+  ROW(StartReport,             LogLevel::llStartUp,     4, Logger::LL_INFO ),
   
   // NODERESTART
   ROW(NR_CopyDict,             LogLevel::llNodeRestart, 8, Logger::LL_INFO ),

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

@@ -134,6 +134,9 @@ Cmvmi::~Cmvmi()
   m_shared_page_pool.clear();
 }
 
+#ifdef ERROR_INSERT
+NodeBitmask c_error_9000_nodes_mask;
+#endif
 
 void Cmvmi::execNDB_TAMPER(Signal* signal) 
 {
@@ -441,21 +444,33 @@ void Cmvmi::execOPEN_COMREQ(Signal* signal)
 
   const Uint32 len = signal->getLength();
   if(len == 2){
-    globalTransporterRegistry.do_connect(tStartingNode);
-    globalTransporterRegistry.setIOState(tStartingNode, HaltIO);
 
-    //-----------------------------------------------------
+#ifdef ERROR_INSERT
-    // Report that the connection to the node is opened
+    if (! (ERROR_INSERTED(9000) && c_error_9000_nodes_mask.get(tStartingNode)))
-    //-----------------------------------------------------
+#endif
-    signal->theData[0] = NDB_LE_CommunicationOpened;
+    {
-    signal->theData[1] = tStartingNode;
+      globalTransporterRegistry.do_connect(tStartingNode);
-    sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 2, JBB);
+      globalTransporterRegistry.setIOState(tStartingNode, HaltIO);
-    //-----------------------------------------------------
+      
+      //-----------------------------------------------------
+      // Report that the connection to the node is opened
+      //-----------------------------------------------------
+      signal->theData[0] = NDB_LE_CommunicationOpened;
+      signal->theData[1] = tStartingNode;
+      sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 2, JBB);
+      //-----------------------------------------------------
+    }
   } else {
     for(unsigned int i = 1; i < MAX_NODES; i++ ) {
       jam();
       if (i != getOwnNodeId() && getNodeInfo(i).m_type == tData2){
 	jam();
+
+#ifdef ERROR_INSERT
+	if (ERROR_INSERTED(9000) && c_error_9000_nodes_mask.get(i))
+	  continue;
+#endif
+	
 	globalTransporterRegistry.do_connect(i);
 	globalTransporterRegistry.setIOState(i, HaltIO);
 	
@@ -1064,7 +1079,8 @@ Cmvmi::execDUMP_STATE_ORD(Signal* signal)
   }
 
   DumpStateOrd * const & dumpState = (DumpStateOrd *)&signal->theData[0];
-  if (dumpState->args[0] == DumpStateOrd::CmvmiDumpConnections){
+  Uint32 arg = dumpState->args[0];
+  if (arg == DumpStateOrd::CmvmiDumpConnections){
     for(unsigned int i = 1; i < MAX_NODES; i++ ){
       const char* nodeTypeStr = "";
       switch(getNodeInfo(i).m_type){
@@ -1094,7 +1110,7 @@ Cmvmi::execDUMP_STATE_ORD(Signal* signal)
     }
   }
   
-  if (dumpState->args[0] == DumpStateOrd::CmvmiDumpLongSignalMemory){
+  if (arg == DumpStateOrd::CmvmiDumpLongSignalMemory){
     infoEvent("Cmvmi: g_sectionSegmentPool size: %d free: %d",
 	      g_sectionSegmentPool.getSize(),
 	      g_sectionSegmentPool.getNoOfFree());
@@ -1131,7 +1147,7 @@ Cmvmi::execDUMP_STATE_ORD(Signal* signal)
     return;
   }
   
-  if (dumpState->args[0] == DumpStateOrd::CmvmiSetRestartOnErrorInsert)
+  if (arg == DumpStateOrd::CmvmiSetRestartOnErrorInsert)
   {
     if(signal->getLength() == 1)
     {
@@ -1151,7 +1167,7 @@ Cmvmi::execDUMP_STATE_ORD(Signal* signal)
     }
   }
 
-  if (dumpState->args[0] == DumpStateOrd::CmvmiTestLongSigWithDelay) {
+  if (arg == DumpStateOrd::CmvmiTestLongSigWithDelay) {
     unsigned i;
     Uint32 loopCount = dumpState->args[1];
     const unsigned len0 = 11;
@@ -1179,6 +1195,30 @@ Cmvmi::execDUMP_STATE_ORD(Signal* signal)
     sendSignal(reference(), GSN_TESTSIG, signal, 8, JBB, ptr, 2);
   }
 
+#ifdef ERROR_INSERT
+  if (arg == 9000)
+  {
+    SET_ERROR_INSERT_VALUE(9000);
+    for (Uint32 i = 1; i<signal->getLength(); i++)
+      c_error_9000_nodes_mask.set(signal->theData[i]);
+  }
+  
+  if (arg == 9001)
+  {
+    CLEAR_ERROR_INSERT_VALUE;
+    for (Uint32 i = 0; i<MAX_NODES; i++)
+    {
+      if (c_error_9000_nodes_mask.get(i))
+      {
+	signal->theData[0] = 0;
+	signal->theData[1] = i;
+	EXECUTE_DIRECT(CMVMI, GSN_OPEN_COMREQ, signal, 2);
+      }
+    }
+    c_error_9000_nodes_mask.clear();
+  }
+#endif
+
 #ifdef VM_TRACE
 #if 0
   {

storage/ndb/src/kernel/blocks/dbdih/DbdihInit.cpp

@@ -58,6 +58,7 @@ void Dbdih::initData()
   cwaitLcpSr       = false;
   c_blockCommit    = false;
   c_blockCommitNo  = 1;
+  cntrlblockref    = RNIL;
 }//Dbdih::initData()
 
 void Dbdih::initRecords() 

storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp

@@ -11995,7 +11995,7 @@ void Dbdih::makeNodeGroups(Uint32 nodeArray[])
   Uint32 tmngNode;
   Uint32 tmngNodeGroup;
   Uint32 tmngLimit;
-  Uint32 i;
+  Uint32 i, j;
 
   /**-----------------------------------------------------------------------
    * ASSIGN ALL ACTIVE NODES INTO NODE GROUPS. HOT SPARE NODES ARE ASSIGNED 
@@ -12041,6 +12041,38 @@ void Dbdih::makeNodeGroups(Uint32 nodeArray[])
       Sysfile::setNodeGroup(mngNodeptr.i, SYSFILE->nodeGroups, mngNodeptr.p->nodeGroup);
     }//if
   }//for
+
+  for (i = 0; i<cnoOfNodeGroups; i++)
+  {
+    jam();
+    bool alive = false;
+    NodeGroupRecordPtr NGPtr;
+    NGPtr.i = i;
+    ptrCheckGuard(NGPtr, MAX_NDB_NODES, nodeGroupRecord);
+    for (j = 0; j<NGPtr.p->nodeCount; j++)
+    {
+      jam();
+      mngNodeptr.i = NGPtr.p->nodesInGroup[j];
+      ptrCheckGuard(mngNodeptr, MAX_NDB_NODES, nodeRecord);
+      if (checkNodeAlive(NGPtr.p->nodesInGroup[j]))
+      {
+	alive = true;
+	break;
+      }
+    }
+
+    if (!alive)
+    {
+      char buf[255];
+      BaseString::snprintf
+	(buf, sizeof(buf), 
+	 "Illegal initial start, no alive node in nodegroup %u", i);
+      progError(__LINE__, 
+		NDBD_EXIT_SR_RESTARTCONFLICT,
+		buf);
+      
+    }
+  }
 }//Dbdih::makeNodeGroups()
 
 /**
@@ -12851,7 +12883,6 @@ void Dbdih::sendStartFragreq(Signal* signal,
 void Dbdih::setInitialActiveStatus()
 {
   NodeRecordPtr siaNodeptr;
-  Uint32 tsiaNodeActiveStatus;
   Uint32 tsiaNoActiveNodes;
 
   tsiaNoActiveNodes = csystemnodes - cnoHotSpare;
@@ -12859,39 +12890,34 @@ void Dbdih::setInitialActiveStatus()
     SYSFILE->nodeStatus[i] = 0;
   for (siaNodeptr.i = 1; siaNodeptr.i < MAX_NDB_NODES; siaNodeptr.i++) {
     ptrAss(siaNodeptr, nodeRecord);
-    if (siaNodeptr.p->nodeStatus == NodeRecord::ALIVE) {
+    switch(siaNodeptr.p->nodeStatus){
+    case NodeRecord::ALIVE:
+    case NodeRecord::DEAD:
       if (tsiaNoActiveNodes == 0) {
         jam();
         siaNodeptr.p->activeStatus = Sysfile::NS_HotSpare;
       } else {
         jam();
         tsiaNoActiveNodes = tsiaNoActiveNodes - 1;
-        siaNodeptr.p->activeStatus = Sysfile::NS_Active;
+        if (siaNodeptr.p->nodeStatus == NodeRecord::ALIVE)
-      }//if
+	{
-    } else {
+	  jam();
-      jam();
+	  siaNodeptr.p->activeStatus = Sysfile::NS_Active;
-      siaNodeptr.p->activeStatus = Sysfile::NS_NotDefined;
+	} 
-    }//if
+	else
-    switch (siaNodeptr.p->activeStatus) {
+	{
-    case Sysfile::NS_Active:
+	  siaNodeptr.p->activeStatus = Sysfile::NS_NotActive_NotTakenOver;
-      jam();
+	}
-      tsiaNodeActiveStatus = Sysfile::NS_Active;
+      }
-      break;
-    case Sysfile::NS_HotSpare:
-      jam();
-      tsiaNodeActiveStatus = Sysfile::NS_HotSpare;
-      break;
-    case Sysfile::NS_NotDefined:
-      jam();
-      tsiaNodeActiveStatus = Sysfile::NS_NotDefined;
       break;
     default:
-      ndbrequire(false);
+      jam();
-      return;
+      siaNodeptr.p->activeStatus = Sysfile::NS_NotDefined;
       break;
-    }//switch
+    }//if
-    Sysfile::setNodeStatus(siaNodeptr.i, SYSFILE->nodeStatus,
+    Sysfile::setNodeStatus(siaNodeptr.i, 
-                           tsiaNodeActiveStatus);
+			   SYSFILE->nodeStatus,
+                           siaNodeptr.p->activeStatus);
   }//for
 }//Dbdih::setInitialActiveStatus()
 
@@ -14613,11 +14639,36 @@ void Dbdih::execWAIT_GCP_REQ(Signal* signal)
     jam();
     conf->senderData = senderData;
     conf->gcp = cnewgcp;
+    conf->blockStatus = cgcpOrderBlocked;
     sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, 
 	       WaitGCPConf::SignalLength, JBB);
     return;
   }//if
 
+  if (requestType == WaitGCPReq::BlockStartGcp)
+  {
+    jam();
+    conf->senderData = senderData;
+    conf->gcp = cnewgcp;
+    conf->blockStatus = cgcpOrderBlocked;
+    sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, 
+	       WaitGCPConf::SignalLength, JBB);
+    cgcpOrderBlocked = 1;
+    return;
+  }
+
+  if (requestType == WaitGCPReq::UnblockStartGcp)
+  {
+    jam();
+    conf->senderData = senderData;
+    conf->gcp = cnewgcp;
+    conf->blockStatus = cgcpOrderBlocked;
+    sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, 
+	       WaitGCPConf::SignalLength, JBB);
+    cgcpOrderBlocked = 0;
+    return;
+  }
+  
   if(isMaster()) {
     /**
      * Master
@@ -14629,6 +14680,7 @@ void Dbdih::execWAIT_GCP_REQ(Signal* signal)
       jam();
       conf->senderData = senderData;
       conf->gcp = coldgcp;
+      conf->blockStatus = cgcpOrderBlocked;
       sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, 
 		 WaitGCPConf::SignalLength, JBB);
       return;
@@ -14715,6 +14767,7 @@ void Dbdih::execWAIT_GCP_CONF(Signal* signal)
 
   conf->senderData = ptr.p->clientData;
   conf->gcp = gcp;
+  conf->blockStatus = cgcpOrderBlocked;
   sendSignal(ptr.p->clientRef, GSN_WAIT_GCP_CONF, signal,
 	     WaitGCPConf::SignalLength, JBB);
   
@@ -14782,6 +14835,7 @@ void Dbdih::emptyWaitGCPMasterQueue(Signal* signal)
 
     c_waitGCPMasterList.next(ptr);    
     conf->senderData = clientData;
+    conf->blockStatus = cgcpOrderBlocked;
     sendSignal(clientRef, GSN_WAIT_GCP_CONF, signal,
 	       WaitGCPConf::SignalLength, JBB);
     

storage/ndb/src/kernel/blocks/ndbcntr/Ndbcntr.hpp

@@ -204,6 +204,7 @@ private:
   void execWAIT_GCP_CONF(Signal* signal);
 
   void execSTOP_REQ(Signal* signal);
+  void execSTOP_CONF(Signal* signal);
   void execRESUME_REQ(Signal* signal);
 
   void execCHANGE_NODE_STATE_CONF(Signal* signal);
@@ -339,6 +340,16 @@ public:
     void progError(int line, int cause, const char * extra) { 
       cntr.progError(line, cause, extra); 
     }
+
+    enum StopNodesStep {
+      SR_BLOCK_GCP_START_GCP = 0,
+      SR_WAIT_COMPLETE_GCP = 1,
+      SR_UNBLOCK_GCP_START_GCP = 2,
+      SR_QMGR_STOP_REQ = 3,
+      SR_WAIT_NODE_FAILURES = 4,
+      SR_CLUSTER_SHUTDOWN = 12
+    } m_state;
+    SignalCounter m_stop_req_counter;
   };
 private:
   StopRecord c_stopRec;

storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrInit.cpp

@@ -88,6 +88,7 @@ Ndbcntr::Ndbcntr(Block_context& ctx):
   addRecSignal(GSN_STOP_ME_CONF, &Ndbcntr::execSTOP_ME_CONF);
 
   addRecSignal(GSN_STOP_REQ, &Ndbcntr::execSTOP_REQ);
+  addRecSignal(GSN_STOP_CONF, &Ndbcntr::execSTOP_CONF);
   addRecSignal(GSN_RESUME_REQ, &Ndbcntr::execRESUME_REQ);
 
   addRecSignal(GSN_WAIT_GCP_REF, &Ndbcntr::execWAIT_GCP_REF);

storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp

@@ -42,6 +42,8 @@
 #include <signaldata/FsRemoveReq.hpp>
 #include <signaldata/ReadConfig.hpp>
 
+#include <signaldata/FailRep.hpp>
+
 #include <AttributeHeader.hpp>
 #include <Configuration.hpp>
 #include <DebuggerNames.hpp>
@@ -853,17 +855,9 @@ Ndbcntr::trySystemRestart(Signal* signal){
     return false;
   }
   
-  if(!allNodes && c_start.m_startPartialTimeout > now){
-    jam();
-    return false;
-  }
-
   NodeState::StartType srType = NodeState::ST_SYSTEM_RESTART;
-  if(c_start.m_waiting.equal(c_start.m_withoutLog)){
+  if(c_start.m_waiting.equal(c_start.m_withoutLog))
-    if(!allNodes){
+  {
-      jam();
-      return false;
-    }
     jam();
     srType = NodeState::ST_INITIAL_START;
     c_start.m_starting = c_start.m_withoutLog; // Used for starting...
@@ -893,10 +887,6 @@ Ndbcntr::trySystemRestart(Signal* signal){
 	ndbrequire(false); // All nodes -> partitioning, which is not allowed
       }
       
-      if(c_start.m_startPartitionedTimeout > now){
-	jam();
-	return false;
-      }
       break;
     }    
     
@@ -1512,13 +1502,74 @@ void Ndbcntr::execNODE_FAILREP(Signal* signal)
   sendSignal(QMGR_REF, GSN_NODE_FAILREP, signal,
 	     NodeFailRep::SignalLength, JBB);
 
+  if (c_stopRec.stopReq.senderRef)
+  {
+    jam();
+    switch(c_stopRec.m_state){
+    case StopRecord::SR_WAIT_NODE_FAILURES:
+    {
+      jam();
+      NdbNodeBitmask tmp;
+      tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+      tmp.bitANDC(allFailed);      
+      tmp.copyto(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+      
+      if (tmp.isclear())
+      {
+	jam();
+	if (c_stopRec.stopReq.senderRef != RNIL)
+	{
+	  jam();
+	  StopConf * const stopConf = (StopConf *)&signal->theData[0];
+	  stopConf->senderData = c_stopRec.stopReq.senderData;
+	  stopConf->nodeState  = (Uint32) NodeState::SL_SINGLEUSER;
+	  sendSignal(c_stopRec.stopReq.senderRef, GSN_STOP_CONF, signal, 
+		     StopConf::SignalLength, JBB);
+	}
+
+	c_stopRec.stopReq.senderRef = 0;
+	WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
+	req->senderRef = reference();
+	req->senderData = StopRecord::SR_UNBLOCK_GCP_START_GCP;
+	req->requestType = WaitGCPReq::UnblockStartGcp;
+	sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
+		   WaitGCPReq::SignalLength, JBA);
+      }
+      break;
+    }
+    case StopRecord::SR_QMGR_STOP_REQ:
+    {
+      NdbNodeBitmask tmp;
+      tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+      tmp.bitANDC(allFailed);      
+
+      if (tmp.isclear())
+      {
+	Uint32 nodeId = allFailed.find(0);
+	tmp.set(nodeId);
+
+	StopConf* conf = (StopConf*)signal->getDataPtrSend();
+	conf->senderData = c_stopRec.stopReq.senderData;
+	conf->nodeId = nodeId;
+	sendSignal(reference(), 
+		   GSN_STOP_CONF, signal, StopConf::SignalLength, JBB);
+      }
+
+      tmp.copyto(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+      
+      break;
+    }
+    }
+  }
+  
+  signal->theData[0] = NDB_LE_NODE_FAILREP;
+  signal->theData[2] = 0;
+  
   Uint32 nodeId = 0;
   while(!allFailed.isclear()){
     nodeId = allFailed.find(nodeId + 1);
     allFailed.clear(nodeId);
-    signal->theData[0] = NDB_LE_NODE_FAILREP;
     signal->theData[1] = nodeId;
-    signal->theData[2] = 0;
     sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 3, JBB);
   }//for
 
@@ -1964,13 +2015,15 @@ void
 Ndbcntr::execDUMP_STATE_ORD(Signal* signal)
 {
   DumpStateOrd * const & dumpState = (DumpStateOrd *)&signal->theData[0];
-  if(signal->theData[0] == 13){
+  Uint32 arg = dumpState->args[0];
+
+  if(arg == 13){
     infoEvent("Cntr: cstartPhase = %d, cinternalStartphase = %d, block = %d", 
 	      cstartPhase, cinternalStartphase, cndbBlocksCount);
     infoEvent("Cntr: cmasterNodeId = %d", cmasterNodeId);
   }
 
-  if (dumpState->args[0] == DumpStateOrd::NdbcntrTestStopOnError){
+  if (arg == DumpStateOrd::NdbcntrTestStopOnError){
     if (m_ctx.m_config.stopOnError() == true)
       ((Configuration&)m_ctx.m_config).stopOnError(false);
     
@@ -1983,6 +2036,28 @@ Ndbcntr::execDUMP_STATE_ORD(Signal* signal)
 	       SystemError::SignalLength, JBA);
   }
 
+  if (arg == DumpStateOrd::NdbcntrStopNodes)
+  {
+    NdbNodeBitmask mask;
+    for(Uint32 i = 1; i<signal->getLength(); i++)
+      mask.set(signal->theData[i]);
+
+    StopReq* req = (StopReq*)signal->getDataPtrSend();
+    req->senderRef = RNIL;
+    req->senderData = 123;
+    req->requestInfo = 0;
+    req->singleuser = 0;
+    req->singleUserApi = 0;
+    mask.copyto(NdbNodeBitmask::Size, req->nodes);
+    StopReq::setPerformRestart(req->requestInfo, 1);
+    StopReq::setNoStart(req->requestInfo, 1);
+    StopReq::setStopNodes(req->requestInfo, 1);
+    StopReq::setStopAbort(req->requestInfo, 1);
+    
+    sendSignal(reference(), GSN_STOP_REQ, signal,
+	       StopReq::SignalLength, JBB);
+    return;
+  }
 
 }//Ndbcntr::execDUMP_STATE_ORD()
 
@@ -2043,9 +2118,12 @@ Ndbcntr::execSTOP_REQ(Signal* signal){
   Uint32 senderData = req->senderData;
   BlockReference senderRef = req->senderRef;
   bool abort = StopReq::getStopAbort(req->requestInfo);
+  bool stopnodes = StopReq::getStopNodes(req->requestInfo);
 
-  if(getNodeState().startLevel < NodeState::SL_STARTED || 
+  if(!singleuser && 
-     abort && !singleuser){
+     (getNodeState().startLevel < NodeState::SL_STARTED || 
+      (abort && !stopnodes)))
+  {
     /**
      * Node is not started yet
      *
@@ -2087,21 +2165,74 @@ Ndbcntr::execSTOP_REQ(Signal* signal){
     else
       ref->errorCode = StopRef::NodeShutdownInProgress;
     ref->senderData = senderData;
-    sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
+    ref->masterNodeId = cmasterNodeId;
+    
+    if (senderRef != RNIL)
+      sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
+    return;
+  }
+
+  if (stopnodes && !abort)
+  {
+    jam();
+    ref->errorCode = StopRef::UnsupportedNodeShutdown;
+    ref->senderData = senderData;
+    ref->masterNodeId = cmasterNodeId;
+    if (senderRef != RNIL)
+      sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
+    return;
+  }
+
+  if (stopnodes && cmasterNodeId != getOwnNodeId())
+  {
+    jam();
+    ref->errorCode = StopRef::MultiNodeShutdownNotMaster;
+    ref->senderData = senderData;
+    ref->masterNodeId = cmasterNodeId;
+    if (senderRef != RNIL)
+      sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
     return;
   }
   
   c_stopRec.stopReq = * req;
   c_stopRec.stopInitiatedTime = NdbTick_CurrentMillisecond();
   
-  if(!singleuser) {
+  if (stopnodes)
-    if(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)) {
+  {
+    jam();
+
+    if(!c_stopRec.checkNodeFail(signal))
+    {
       jam();
-      if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){
+      return;
+    }
+
+    char buf[100];
+    NdbNodeBitmask mask;
+    mask.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+    infoEvent("Initiating shutdown abort of %s", mask.getText(buf));
+    ndbout_c("Initiating shutdown abort of %s", mask.getText(buf));    
+
+    WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
+    req->senderRef = reference();
+    req->senderData = StopRecord::SR_BLOCK_GCP_START_GCP;
+    req->requestType = WaitGCPReq::BlockStartGcp;
+    sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
+	       WaitGCPReq::SignalLength, JBB);
+    return;
+  }
+  else if(!singleuser) 
+  {
+    if(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)) 
+    {
+      jam();
+      if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo))
+      {
 	((Configuration&)m_ctx.m_config).stopOnError(false);
       }
     }
-    if(!c_stopRec.checkNodeFail(signal)){
+    if(!c_stopRec.checkNodeFail(signal))
+    {
       jam();
       return;
     }
@@ -2171,7 +2302,17 @@ Ndbcntr::StopRecord::checkNodeFail(Signal* signal){
    */
   NodeBitmask ndbMask; 
   ndbMask.assign(cntr.c_startedNodes);
-  ndbMask.clear(cntr.getOwnNodeId());
+
+  if (StopReq::getStopNodes(stopReq.requestInfo))
+  {
+    NdbNodeBitmask tmp;
+    tmp.assign(NdbNodeBitmask::Size, stopReq.nodes);
+    ndbMask.bitANDC(tmp);
+  }
+  else
+  {
+    ndbMask.clear(cntr.getOwnNodeId());
+  }
   
   CheckNodeGroups* sd = (CheckNodeGroups*)&signal->theData[0];
   sd->blockRef = cntr.reference();
@@ -2191,9 +2332,11 @@ Ndbcntr::StopRecord::checkNodeFail(Signal* signal){
   
   ref->senderData = stopReq.senderData;
   ref->errorCode = StopRef::NodeShutdownWouldCauseSystemCrash;
+  ref->masterNodeId = cntr.cmasterNodeId;
   
   const BlockReference bref = stopReq.senderRef;
-  cntr.sendSignal(bref, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
+  if (bref != RNIL)
+    cntr.sendSignal(bref, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
   
   stopReq.senderRef = 0;
 
@@ -2243,23 +2386,23 @@ Ndbcntr::StopRecord::checkTcTimeout(Signal* signal){
     if(stopReq.getSystemStop(stopReq.requestInfo)  || stopReq.singleuser){
       jam();
       if(stopReq.singleuser) 
-	{
+      {
-	  jam();
+	jam();
-	   AbortAllReq * req = (AbortAllReq*)&signal->theData[0];
+	AbortAllReq * req = (AbortAllReq*)&signal->theData[0];
-	   req->senderRef = cntr.reference();
+	req->senderRef = cntr.reference();
-	   req->senderData = 12;
+	req->senderData = 12;
-	   cntr.sendSignal(DBTC_REF, GSN_ABORT_ALL_REQ, signal, 
+	cntr.sendSignal(DBTC_REF, GSN_ABORT_ALL_REQ, signal, 
-		      AbortAllReq::SignalLength, JBB);
+			AbortAllReq::SignalLength, JBB);
-	} 
+      } 
       else
-	{
+      {
-	  WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
+	WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
-	  req->senderRef = cntr.reference();
+	req->senderRef = cntr.reference();
-	  req->senderData = 12;
+	req->senderData = StopRecord::SR_CLUSTER_SHUTDOWN;
-	  req->requestType = WaitGCPReq::CompleteForceStart;
+	req->requestType = WaitGCPReq::CompleteForceStart;
-	  cntr.sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
+	cntr.sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
-			  WaitGCPReq::SignalLength, JBB);
+			WaitGCPReq::SignalLength, JBB);
-	}
+      }
     } else {
       jam();
       StopPermReq * req = (StopPermReq*)&signal->theData[0];
@@ -2338,6 +2481,7 @@ void Ndbcntr::execABORT_ALL_REF(Signal* signal){
   StopRef * const stopRef = (StopRef *)&signal->theData[0];
   stopRef->senderData = c_stopRec.stopReq.senderData;
   stopRef->errorCode = StopRef::TransactionAbortFailed;
+  stopRef->masterNodeId = cmasterNodeId;
   sendSignal(c_stopRec.stopReq.senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB);
 }
 
@@ -2421,7 +2565,7 @@ void Ndbcntr::execWAIT_GCP_REF(Signal* signal){
 
   WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
   req->senderRef = reference();
-  req->senderData = 12;
+  req->senderData = StopRecord::SR_CLUSTER_SHUTDOWN;
   req->requestType = WaitGCPReq::CompleteForceStart;
   sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
 	     WaitGCPReq::SignalLength, JBB);
@@ -2430,29 +2574,129 @@ void Ndbcntr::execWAIT_GCP_REF(Signal* signal){
 void Ndbcntr::execWAIT_GCP_CONF(Signal* signal){
   jamEntry();
 
-  ndbrequire(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo));
+  WaitGCPConf* conf = (WaitGCPConf*)signal->getDataPtr();
-  NodeState newState(NodeState::SL_STOPPING_3, true); 
 
-  /**
+  switch(conf->senderData){
-   * Inform QMGR so that arbitrator won't kill us
+  case StopRecord::SR_BLOCK_GCP_START_GCP:
-   */
+  {
-  NodeStateRep * rep = (NodeStateRep *)&signal->theData[0];
+    jam();
-  rep->nodeState = newState;
+    /**
-  rep->nodeState.masterNodeId = cmasterNodeId;
+     * 
-  rep->nodeState.setNodeGroup(c_nodeGroup);
+     */
-  EXECUTE_DIRECT(QMGR, GSN_NODE_STATE_REP, signal, NodeStateRep::SignalLength);
+    if(!c_stopRec.checkNodeFail(signal))
+    {
+      jam();
+      goto unblock;
+    }
+    
+    WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
+    req->senderRef = reference();
+    req->senderData = StopRecord::SR_WAIT_COMPLETE_GCP;
+    req->requestType = WaitGCPReq::CompleteIfRunning;
 
-  if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){
+    sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
-    jam();
+	       WaitGCPReq::SignalLength, JBB);
-    StartOrd * startOrd = (StartOrd *)&signal->theData[0];
+    return;
-    startOrd->restartInfo = c_stopRec.stopReq.requestInfo;
+  }
-    sendSignalWithDelay(CMVMI_REF, GSN_START_ORD, signal, 500, 
+  case StopRecord::SR_UNBLOCK_GCP_START_GCP:
-			StartOrd::SignalLength);
+  {
-  } else {
+    jam();
-    jam();
+    return;
-    sendSignalWithDelay(CMVMI_REF, GSN_STOP_ORD, signal, 500, 1);
+  }
+  case StopRecord::SR_WAIT_COMPLETE_GCP:
+  {
+    jam();
+    if(!c_stopRec.checkNodeFail(signal))
+    {
+      jam();
+      goto unblock;
+    }
+
+    NdbNodeBitmask tmp;
+    tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+    c_stopRec.m_stop_req_counter = tmp;
+    NodeReceiverGroup rg(QMGR, tmp);
+    StopReq * stopReq = (StopReq *)&signal->theData[0];
+    * stopReq = c_stopRec.stopReq;
+    stopReq->senderRef = reference();
+    sendSignal(rg, GSN_STOP_REQ, signal, StopReq::SignalLength, JBA);
+    c_stopRec.m_state = StopRecord::SR_QMGR_STOP_REQ; 
+    return;
+  }
+  case StopRecord::SR_CLUSTER_SHUTDOWN:
+  {
+    jam();
+    break;
+  }
+  }
+  
+  {  
+    ndbrequire(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo));
+    NodeState newState(NodeState::SL_STOPPING_3, true); 
+    
+    /**
+     * Inform QMGR so that arbitrator won't kill us
+     */
+    NodeStateRep * rep = (NodeStateRep *)&signal->theData[0];
+    rep->nodeState = newState;
+    rep->nodeState.masterNodeId = cmasterNodeId;
+    rep->nodeState.setNodeGroup(c_nodeGroup);
+    EXECUTE_DIRECT(QMGR, GSN_NODE_STATE_REP, signal, 
+		   NodeStateRep::SignalLength);
+    
+    if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){
+      jam();
+      StartOrd * startOrd = (StartOrd *)&signal->theData[0];
+      startOrd->restartInfo = c_stopRec.stopReq.requestInfo;
+      sendSignalWithDelay(CMVMI_REF, GSN_START_ORD, signal, 500, 
+			  StartOrd::SignalLength);
+    } else {
+      jam();
+      sendSignalWithDelay(CMVMI_REF, GSN_STOP_ORD, signal, 500, 1);
+    }
+    return;
+  }
+  
+unblock:
+  WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0];
+  req->senderRef = reference();
+  req->senderData = StopRecord::SR_UNBLOCK_GCP_START_GCP;
+  req->requestType = WaitGCPReq::UnblockStartGcp;
+  sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, 
+	     WaitGCPReq::SignalLength, JBB);
+}
+
+void
+Ndbcntr::execSTOP_CONF(Signal* signal)
+{
+  jamEntry();
+  StopConf *conf = (StopConf*)signal->getDataPtr();
+  ndbrequire(c_stopRec.m_state == StopRecord::SR_QMGR_STOP_REQ);
+  c_stopRec.m_stop_req_counter.clearWaitingFor(conf->nodeId);
+  if (c_stopRec.m_stop_req_counter.done())
+  {
+    char buf[100];
+    NdbNodeBitmask mask;
+    mask.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes);
+    infoEvent("Stopping of %s", mask.getText(buf));
+    ndbout_c("Stopping of %s", mask.getText(buf));    
+
+    /**
+     * Kill any node...
+     */
+    FailRep * const failRep = (FailRep *)&signal->theData[0];
+    failRep->failCause = FailRep::ZMULTI_NODE_SHUTDOWN;
+    NodeReceiverGroup rg(QMGR, c_clusterNodes);
+    Uint32 nodeId = 0;
+    while ((nodeId = NdbNodeBitmask::find(c_stopRec.stopReq.nodes, nodeId+1))
+	   != NdbNodeBitmask::NotFound)
+    {
+      failRep->failNodeId = nodeId;
+      sendSignal(rg, GSN_FAIL_REP, signal, FailRep::SignalLength, JBA);
+    }
+    c_stopRec.m_state = StopRecord::SR_WAIT_NODE_FAILURES;
+    return;
   }
-  return;
 }
 
 void Ndbcntr::execSTTORRY(Signal* signal){

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

@@ -33,6 +33,7 @@
 
 #include <SafeCounter.hpp>
 #include <RequestTracker.hpp>
+#include <signaldata/StopReq.hpp>
 
 #include "timer.hpp"
 
@@ -53,6 +54,7 @@
 #define ZAPI_HB_HANDLING 3
 #define ZTIMER_HANDLING 4
 #define ZARBIT_HANDLING 5
+#define ZSTART_FAILURE_LIMIT 6
 
 /* Error Codes ------------------------------*/
 #define ZERRTOOMANY 1101
@@ -104,18 +106,42 @@ public:
   };
 
   struct StartRecord {
-    void reset(){ m_startKey++; m_startNode = 0;}
+    void reset(){ 
+      m_startKey++; 
+      m_startNode = 0; 
+      m_gsn = RNIL; 
+      m_nodes.clearWaitingFor();
+    }
     Uint32 m_startKey;
     Uint32 m_startNode;
     Uint64 m_startTimeout;
     
     Uint32 m_gsn;
     SignalCounter m_nodes;
-  } c_start;
+    Uint32 m_latest_gci;
 
+    Uint32 m_start_type;
+    NdbNodeBitmask m_skip_nodes;
+    NdbNodeBitmask m_starting_nodes;
+    NdbNodeBitmask m_starting_nodes_w_log;
+
+    Uint16 m_president_candidate;
+    Uint32 m_president_candidate_gci;
+    Uint16 m_regReqReqSent;
+    Uint16 m_regReqReqRecv;
+  } c_start;
+  
   NdbNodeBitmask c_definedNodes; // DB nodes in config
   NdbNodeBitmask c_clusterNodes; // DB nodes in cluster
   NodeBitmask c_connectedNodes;  // All kinds of connected nodes
+
+  /**
+   * Nodes which we're checking for partitioned cluster
+   *
+   * i.e. nodes that connect to use, when we already have elected president
+   */
+  NdbNodeBitmask c_readnodes_nodes;
+
   Uint32 c_maxDynamicId;
   
   // Records
@@ -208,6 +234,7 @@ private:
   void execPRES_TOCONF(Signal* signal);
   void execDISCONNECT_REP(Signal* signal);
   void execSYSTEM_ERROR(Signal* signal);
+  void execSTOP_REQ(Signal* signal);
 
   // Received signals
   void execDUMP_STATE_ORD(Signal* signal);
@@ -222,7 +249,12 @@ private:
   void execREAD_NODESREQ(Signal* signal);
   void execSET_VAR_REQ(Signal* signal);
 
+  void execREAD_NODESREF(Signal* signal);
+  void execREAD_NODESCONF(Signal* signal);
 
+  void execDIH_RESTARTREF(Signal* signal);
+  void execDIH_RESTARTCONF(Signal* signal);
+  
   void execAPI_VERSION_REQ(Signal* signal);
   void execAPI_BROADCAST_REP(Signal* signal);
 
@@ -244,6 +276,9 @@ private:
   void execARBIT_STOPREP(Signal* signal);
 
   // Statement blocks
+  void check_readnodes_reply(Signal* signal, Uint32 nodeId, Uint32 gsn);
+  Uint32 check_startup(Signal* signal);
+
   void node_failed(Signal* signal, Uint16 aFailedNode);
   void checkStartInterface(Signal* signal);
   void failReport(Signal* signal,
@@ -261,8 +296,9 @@ private:
 
   // Generated statement blocks
   void startphase1(Signal* signal);
-  void electionWon();
+  void electionWon(Signal* signal);
   void cmInfoconf010Lab(Signal* signal);
+  
   void apiHbHandlingLab(Signal* signal);
   void timerHandlingLab(Signal* signal);
   void hbReceivedLab(Signal* signal);
@@ -364,12 +400,12 @@ private:
   /* Status flags ----------------------------------*/
 
   Uint32 c_restartPartialTimeout;
+  Uint32 c_restartPartionedTimeout;
+  Uint32 c_restartFailureTimeout;
+  Uint64 c_start_election_time;
 
   Uint16 creadyDistCom;
-  Uint16 c_regReqReqSent;
+
-  Uint16 c_regReqReqRecv;
-  Uint64 c_stopElectionTime;
-  Uint16 cpresidentCandidate;
   Uint16 cdelayRegreq;
   Uint16 cpresidentAlive;
   Uint16 cnoFailedNodes;
@@ -397,7 +433,7 @@ private:
   Uint16 cfailedNodes[MAX_NDB_NODES];
   Uint16 cprepFailedNodes[MAX_NDB_NODES];
   Uint16 ccommitFailedNodes[MAX_NDB_NODES];
-
+  
   struct OpAllocNodeIdReq {
     RequestTracker m_tracker;
     AllocNodeIdReq m_req;
@@ -406,6 +442,9 @@ private:
   };
 
   struct OpAllocNodeIdReq opAllocNodeIdReq;
+  
+  StopReq c_stopReq;
+  bool check_multi_node_shutdown(Signal* signal);
 };
 
 #endif

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

@@ -35,9 +35,8 @@ void Qmgr::initData()
 
   Uint32 hbDBAPI = 500;
   setHbApiDelay(hbDBAPI);
-
-  c_connectedNodes.clear();
   c_connectedNodes.set(getOwnNodeId());
+  c_stopReq.senderRef = 0;
 }//Qmgr::initData()
 
 void Qmgr::initRecords() 
@@ -52,6 +51,7 @@ Qmgr::Qmgr(Block_context& ctx)
 
   // Transit signals
   addRecSignal(GSN_DUMP_STATE_ORD, &Qmgr::execDUMP_STATE_ORD);
+  addRecSignal(GSN_STOP_REQ, &Qmgr::execSTOP_REQ);
   addRecSignal(GSN_DEBUG_SIG, &Qmgr::execDEBUG_SIG);
   addRecSignal(GSN_CONTINUEB, &Qmgr::execCONTINUEB);
   addRecSignal(GSN_CM_HEARTBEAT, &Qmgr::execCM_HEARTBEAT);
@@ -101,6 +101,12 @@ Qmgr::Qmgr(Block_context& ctx)
   addRecSignal(GSN_ARBIT_CHOOSEREF, &Qmgr::execARBIT_CHOOSEREF);
   addRecSignal(GSN_ARBIT_STOPREP, &Qmgr::execARBIT_STOPREP);
 
+  addRecSignal(GSN_READ_NODESREF, &Qmgr::execREAD_NODESREF);
+  addRecSignal(GSN_READ_NODESCONF, &Qmgr::execREAD_NODESCONF);
+
+  addRecSignal(GSN_DIH_RESTARTREF, &Qmgr::execDIH_RESTARTREF);
+  addRecSignal(GSN_DIH_RESTARTCONF, &Qmgr::execDIH_RESTARTCONF);
+  
   initData();
 }//Qmgr::Qmgr()
 

storage/ndb/src/kernel/vm/Configuration.cpp

@@ -55,6 +55,12 @@ enum ndbd_options {
 NDB_STD_OPTS_VARS;
 // XXX should be my_bool ???
 static int _daemon, _no_daemon, _foreground,  _initial, _no_start;
+static int _initialstart;
+static const char* _nowait_nodes;
+
+extern Uint32 g_start_type;
+extern NdbNodeBitmask g_nowait_nodes;
+
 /**
  * Arguments to NDB process
  */ 
@@ -82,6 +88,14 @@ static struct my_option my_long_options[] =
     " (implies --nodaemon)",
     (gptr*) &_foreground, (gptr*) &_foreground, 0,
     GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0 },
+  { "nowait-nodes", NO_ARG, 
+    "Nodes that will not be waited for during start",
+    (gptr*) &_nowait_nodes, (gptr*) &_nowait_nodes, 0,
+    GET_STR, REQUIRED_ARG, 0, 0, 0, 0, 0, 0 },
+  { "initial-start", NO_ARG, 
+    "Perform initial start",
+    (gptr*) &_initialstart, (gptr*) &_initialstart, 0,
+    GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0 },
   { 0, 0, 0, 0, 0, 0, GET_NO_ARG, NO_ARG, 0, 0, 0, 0, 0, 0}
 };
 static void short_usage_sub(void)
@@ -150,6 +164,37 @@ Configuration::init(int argc, char** argv)
   
   globalData.ownId= 0;
 
+  if (_nowait_nodes)
+  {
+    BaseString str(_nowait_nodes);
+    Vector<BaseString> arr;
+    str.split(arr, ",");
+    for (Uint32 i = 0; i<arr.size(); i++)
+    {
+      char *endptr = 0;
+      long val = strtol(arr[i].c_str(), &endptr, 10);
+      if (*endptr)
+      {
+	ndbout_c("Unable to parse nowait-nodes argument: %s : %s", 
+		 arr[i].c_str(), _nowait_nodes);
+	exit(-1);
+      }
+      if (! (val > 0 && val < MAX_NDB_NODES))
+      {
+	ndbout_c("Invalid nodeid specified in nowait-nodes: %d : %s", 
+		 val, _nowait_nodes);
+	exit(-1);
+      }
+      g_nowait_nodes.set(val);
+    }
+  }
+
+  if (_initialstart)
+  {
+    _initialStart = true;
+    g_start_type |= (1 << NodeState::ST_INITIAL_START);
+  }
+  
   return true;
 }
 

storage/ndb/src/mgmclient/CommandInterpreter.cpp

@@ -18,6 +18,7 @@
 #include <my_sys.h>
 #include <Vector.hpp>
 #include <mgmapi.h>
+#include <util/BaseString.hpp>
 
 class MgmtSrvr;
 
@@ -63,6 +64,9 @@ private:
    */
   void analyseAfterFirstToken(int processId, char* allAfterFirstTokenCstr);
 
+  void executeCommand(Vector<BaseString> &command_list,
+                      unsigned command_pos,
+                      int *node_ids, int no_of_nodes);
   /**
    *   Parse the block specification part of the LOG* commands,
    *   things after LOG*: [BLOCK = {ALL|<blockName>+}]
@@ -97,10 +101,14 @@ private:
 
 public:
   void executeStop(int processId, const char* parameters, bool all);
+  void executeStop(Vector<BaseString> &command_list, unsigned command_pos,
+                   int *node_ids, int no_of_nodes);
   void executeEnterSingleUser(char* parameters);
   void executeExitSingleUser(char* parameters);
   void executeStart(int processId, const char* parameters, bool all);
   void executeRestart(int processId, const char* parameters, bool all);
+  void executeRestart(Vector<BaseString> &command_list, unsigned command_pos,
+                      int *node_ids, int no_of_nodes);
   void executeLogLevel(int processId, const char* parameters, bool all);
   void executeError(int processId, const char* parameters, bool all);
   void executeLog(int processId, const char* parameters, bool all);
@@ -583,9 +591,16 @@ CommandInterpreter::execute_impl(const char *_line)
     }
   } while (do_continue);
   // if there is anything in the line proceed
+  Vector<BaseString> command_list;
+  {
+    BaseString tmp(line);
+    tmp.split(command_list);
+    for (unsigned i= 0; i < command_list.size();)
+      command_list[i].c_str()[0] ? i++ : (command_list.erase(i),0);
+  }
   char* firstToken = strtok(line, " ");
   char* allAfterFirstToken = strtok(NULL, "");
-  
+
   if (strcasecmp(firstToken, "HELP") == 0 ||
       strcasecmp(firstToken, "?") == 0) {
     executeHelp(allAfterFirstToken);
@@ -656,22 +671,45 @@ CommandInterpreter::execute_impl(const char *_line)
     analyseAfterFirstToken(-1, allAfterFirstToken);
   } else {
     /**
-     * First token should be a digit, node ID
+     * First tokens should be digits, node ID's
      */
-    int nodeId;
+    int node_ids[MAX_NODES];
-
+    unsigned pos;
-    if (! convert(firstToken, nodeId)) {
+    for (pos= 0; pos < command_list.size(); pos++)
+    {
+      int node_id;
+      if (convert(command_list[pos].c_str(), node_id))
+      {
+        if (node_id <= 0) {
+          ndbout << "Invalid node ID: " << command_list[pos].c_str()
+                 << "." << endl;
+          DBUG_RETURN(true);
+        }
+        node_ids[pos]= node_id;
+        continue;
+      }
+      break;
+    }
+    int no_of_nodes= pos;
+    if (no_of_nodes == 0)
+    {
+      /* No digit found */
       invalid_command(_line);
       DBUG_RETURN(true);
     }
-
+    if (pos == command_list.size())
-    if (nodeId <= 0) {
+    {
-      ndbout << "Invalid node ID: " << firstToken << "." << endl;
+      /* No command found */
+      invalid_command(_line);
       DBUG_RETURN(true);
     }
-    
+    if (no_of_nodes == 1)
-    analyseAfterFirstToken(nodeId, allAfterFirstToken);
+    {
-    
+      analyseAfterFirstToken(node_ids[0], allAfterFirstToken);
+      DBUG_RETURN(true);
+    }
+    executeCommand(command_list, pos, node_ids, no_of_nodes);
+    DBUG_RETURN(true);
   }
   DBUG_RETURN(true);
 }
@@ -741,6 +779,27 @@ CommandInterpreter::analyseAfterFirstToken(int processId,
   ndbout << endl;
 }
 
+void
+CommandInterpreter::executeCommand(Vector<BaseString> &command_list,
+                                   unsigned command_pos,
+                                   int *node_ids, int no_of_nodes)
+{
+  const char *cmd= command_list[command_pos].c_str();
+  if (strcasecmp("STOP", cmd) == 0)
+  {
+    executeStop(command_list, command_pos+1, node_ids, no_of_nodes);
+    return;
+  }
+  if (strcasecmp("RESTART", cmd) == 0)
+  {
+    executeRestart(command_list, command_pos+1, node_ids, no_of_nodes);
+    return;
+  }
+  ndbout_c("Invalid command: '%s' after multi node id list. "
+           "Expected STOP or RESTART.", cmd);
+  return;
+}
+
 /**
  * Get next nodeid larger than the give node_id. node_id will be
  * set to the next node_id in the list. node_id should be set
@@ -1326,24 +1385,60 @@ CommandInterpreter::executeClusterLog(char* parameters)
 //*****************************************************************************
 
 void
-CommandInterpreter::executeStop(int processId, const char *, bool all) 
+CommandInterpreter::executeStop(int processId, const char *parameters,
+                                bool all) 
 {
-  int result = 0;
+  Vector<BaseString> command_list;
-  if(all) {
+  if (parameters)
-    result = ndb_mgm_stop(m_mgmsrv, 0, 0);
+  {
-  } else {
+    BaseString tmp(parameters);
-    result = ndb_mgm_stop(m_mgmsrv, 1, &processId);
+    tmp.split(command_list);
+    for (unsigned i= 0; i < command_list.size();)
+      command_list[i].c_str()[0] ? i++ : (command_list.erase(i),0);
   }
-  if (result < 0) {
+  if (all)
-    ndbout << "Shutdown failed." << endl;
+    executeStop(command_list, 0, 0, 0);
-    printError();
+  else
-  } else
+    executeStop(command_list, 0, &processId, 1);
+}
+
+void
+CommandInterpreter::executeStop(Vector<BaseString> &command_list,
+                                unsigned command_pos,
+                                int *node_ids, int no_of_nodes)
+{
+  int abort= 0;
+  for (; command_pos < command_list.size(); command_pos++)
+  {
+    const char *item= command_list[command_pos].c_str();
+    if (strcasecmp(item, "-A") == 0)
     {
-      if(all)
+      abort= 1;
-	ndbout << "NDB Cluster has shutdown." << endl;
+      continue;
-      else
-	ndbout << "Node " << processId << " has shutdown." << endl;
     }
+    ndbout_c("Invalid option: %s. Expecting -A after STOP",
+             item);
+    return;
+  }
+
+  int result= ndb_mgm_stop2(m_mgmsrv, no_of_nodes, node_ids, abort);
+  if (result < 0)
+  {
+    ndbout_c("Shutdown failed.");
+    printError();
+  }
+  else
+  {
+    if (node_ids == 0)
+      ndbout_c("NDB Cluster has shutdown.");
+    else
+    {
+      ndbout << "Node";
+      for (int i= 0; i < no_of_nodes; i++)
+        ndbout << " " << node_ids[i];
+      ndbout_c(" has shutdown.");
+    }
+  }
 }
 
 void
@@ -1409,47 +1504,74 @@ CommandInterpreter::executeStart(int processId, const char* parameters,
 
 void
 CommandInterpreter::executeRestart(int processId, const char* parameters,
-				   bool all) 
+				   bool all)
+{
+  Vector<BaseString> command_list;
+  if (parameters)
+  {
+    BaseString tmp(parameters);
+    tmp.split(command_list);
+    for (unsigned i= 0; i < command_list.size();)
+      command_list[i].c_str()[0] ? i++ : (command_list.erase(i),0);
+  }
+  if (all)
+    executeRestart(command_list, 0, 0, 0);
+  else
+    executeRestart(command_list, 0, &processId, 1);
+}
+
+void
+CommandInterpreter::executeRestart(Vector<BaseString> &command_list,
+                                   unsigned command_pos,
+                                   int *node_ids, int no_of_nodes)
 {
   int result;
-  int nostart = 0;
+  int nostart= 0;
-  int initialstart = 0;
+  int initialstart= 0;
-  int abort = 0;
+  int abort= 0;
 
-  if(parameters != 0 && strlen(parameters) != 0){
+  for (; command_pos < command_list.size(); command_pos++)
-    char * tmpString = my_strdup(parameters,MYF(MY_WME));
+  {
-    My_auto_ptr<char> ap1(tmpString);
+    const char *item= command_list[command_pos].c_str();
-    char * tmpPtr = 0;
+    if (strcasecmp(item, "-N") == 0)
-    char * item = strtok_r(tmpString, " ", &tmpPtr);
+    {
-    while(item != NULL){
+      nostart= 1;
-      if(strcasecmp(item, "-N") == 0)
+      continue;
-	nostart = 1;
-      if(strcasecmp(item, "-I") == 0)
-	initialstart = 1;
-      if(strcasecmp(item, "-A") == 0)
-	abort = 1;
-      item = strtok_r(NULL, " ", &tmpPtr);
     }
+    if (strcasecmp(item, "-I") == 0)
+    {
+      initialstart= 1;
+      continue;
+    }
+    if (strcasecmp(item, "-A") == 0)
+    {
+      abort= 1;
+      continue;
+    }
+    ndbout_c("Invalid option: %s. Expecting -A,-N or -I after RESTART",
+             item);
+    return;
   }
 
-  if(all) {
+  result= ndb_mgm_restart2(m_mgmsrv, no_of_nodes, node_ids,
-    result = ndb_mgm_restart2(m_mgmsrv, 0, NULL, initialstart, nostart, abort);
+                           initialstart, nostart, abort);
-  } else {
-    int v[1];
-    v[0] = processId;
-    result = ndb_mgm_restart2(m_mgmsrv, 1, v, initialstart, nostart, abort);
-  }
   
   if (result <= 0) {
-    ndbout.println("Restart failed.", result);
+    ndbout_c("Restart failed.");
     printError();
-  } else
+  }
+  else
+  {
+    if (node_ids == 0)
+      ndbout_c("NDB Cluster is being restarted.");
+    else
     {
-      if(all)
+      ndbout << "Node";
-	ndbout << "NDB Cluster is being restarted." << endl;
+      for (int i= 0; i < no_of_nodes; i++)
-      else
+        ndbout << " " << node_ids[i];
-	ndbout_c("Node %d is being restarted.", processId);
+      ndbout_c(" is being restarted");
     }
+  }
 }
 
 void

storage/ndb/src/mgmsrv/MgmtSrvr.cpp

@@ -294,6 +294,8 @@ static ErrorItem errorTable[] =
   {MgmtSrvr::SYSTEM_SHUTDOWN_IN_PROGRESS, "System shutdown in progress" },
   {MgmtSrvr::NODE_SHUTDOWN_WOULD_CAUSE_SYSTEM_CRASH,
    "Node shutdown would cause system crash" },
+  {MgmtSrvr::UNSUPPORTED_NODE_SHUTDOWN,
+   "Unsupported multi node shutdown. Abort option required." },
   {MgmtSrvr::NODE_NOT_API_NODE, "The specified node is not an API node." },
   {MgmtSrvr::OPERATION_NOT_ALLOWED_START_STOP, 
    "Operation not allowed while nodes are starting or stopping."},
@@ -312,6 +314,9 @@ int MgmtSrvr::translateStopRef(Uint32 errCode)
   case StopRef::NodeShutdownWouldCauseSystemCrash:
     return NODE_SHUTDOWN_WOULD_CAUSE_SYSTEM_CRASH;
     break;
+  case StopRef::UnsupportedNodeShutdown:
+    return UNSUPPORTED_NODE_SHUTDOWN;
+    break;
   }
   return 4999;
 }
@@ -386,8 +391,9 @@ MgmtSrvr::MgmtSrvr(SocketServer *socket_server,
   _ownReference(0),
   theSignalIdleList(NULL),
   theWaitState(WAIT_SUBSCRIBE_CONF),
+  m_local_mgm_handle(0),
   m_event_listner(this),
-  m_local_mgm_handle(0)
+  m_master_node(0)
 {
     
   DBUG_ENTER("MgmtSrvr::MgmtSrvr");
@@ -672,23 +678,16 @@ MgmtSrvr::~MgmtSrvr()
 
 int MgmtSrvr::okToSendTo(NodeId nodeId, bool unCond) 
 {
-  if(nodeId == 0)
+  if(nodeId == 0 || getNodeType(nodeId) != NDB_MGM_NODE_TYPE_NDB)
-    return 0;
-
-  if (getNodeType(nodeId) != NDB_MGM_NODE_TYPE_NDB)
     return WRONG_PROCESS_TYPE;
-  
   // Check if we have contact with it
   if(unCond){
     if(theFacade->theClusterMgr->getNodeInfo(nodeId).connected)
       return 0;
-    return NO_CONTACT_WITH_PROCESS;
   }
-  if (theFacade->get_node_alive(nodeId) == 0) {
+  else if (theFacade->get_node_alive(nodeId) == true)
-    return NO_CONTACT_WITH_PROCESS;
-  } else {
     return 0;
-  }
+  return NO_CONTACT_WITH_PROCESS;
 }
 
 void report_unknown_signal(SimpleSignal *signal)
@@ -930,7 +929,7 @@ int MgmtSrvr::sendStopMgmd(NodeId nodeId,
  * distributed communication up.
  */
 
-int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
+int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids,
 			   NodeBitmask &stoppedNodes,
 			   Uint32 singleUserNodeId,
 			   bool abort,
@@ -940,6 +939,12 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
 			   bool initialStart)
 {
   int error = 0;
+  DBUG_ENTER("MgmtSrvr::sendSTOP_REQ");
+  DBUG_PRINT("enter", ("no of nodes: %d  singleUseNodeId: %d  "
+                       "abort: %d  stop: %d  restart: %d  "
+                       "nostart: %d  initialStart: %d",
+                       node_ids.size(), singleUserNodeId,
+                       abort, stop, restart, nostart, initialStart));
 
   stoppedNodes.clear();
 
@@ -977,36 +982,49 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
 
   // send the signals
   NodeBitmask nodes;
-  if (nodeId)
+  NodeId nodeId= 0;
+  int use_master_node= 0;
+  int do_send= 0;
+  int do_stop_self= 0;
+  NdbNodeBitmask nodes_to_stop;
   {
-    if(nodeId==getOwnNodeId())
+    for (unsigned i= 0; i < node_ids.size(); i++)
     {
-      if(restart)
+      nodeId= node_ids[i];
-        g_RestartServer= true;
+      if (getNodeType(nodeId) != NDB_MGM_NODE_TYPE_MGM)
-      g_StopServer= true;
+        nodes_to_stop.set(nodeId);
-      return 0;
+      else if (nodeId != getOwnNodeId())
+      {
+        error= sendStopMgmd(nodeId, abort, stop, restart,
+                            nostart, initialStart);
+        if (error == 0)
+          stoppedNodes.set(nodeId);
+      }
+      else
+        do_stop_self= 1;;
     }
-    if(getNodeType(nodeId) == NDB_MGM_NODE_TYPE_NDB)
+  }
+  int no_of_nodes_to_stop= nodes_to_stop.count();
+  if (node_ids.size())
+  {
+    if (no_of_nodes_to_stop)
     {
-      int r;
+      do_send= 1;
-      if((r= okToSendTo(nodeId, true)) != 0)
+      if (no_of_nodes_to_stop == 1)
-        return r;
+      {
-      if (ss.sendSignal(nodeId, &ssig) != SEND_OK)
+        nodeId= nodes_to_stop.find(0);
-	return SEND_OR_RECEIVE_FAILED;
+      }
+      else // multi node stop, send to master
+      {
+        use_master_node= 1;
+        nodes_to_stop.copyto(NdbNodeBitmask::Size, stopReq->nodes);
+        StopReq::setStopNodes(stopReq->requestInfo, 1);
+      }
     }
-    else if(getNodeType(nodeId) == NDB_MGM_NODE_TYPE_MGM)
-    {
-      error= sendStopMgmd(nodeId, abort, stop, restart, nostart, initialStart);
-      if(error==0)
-        stoppedNodes.set(nodeId);
-      return error;
-    }
-    else
-      return WRONG_PROCESS_TYPE;
-    nodes.set(nodeId);
   }
   else
   {
+    nodeId= 0;
     while(getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_NDB))
     {
       if(okToSendTo(nodeId, true) == 0)
@@ -1027,8 +1045,30 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
   }
 
   // now wait for the replies
-  while (!nodes.isclear())
+  while (!nodes.isclear() || do_send)
   {
+    if (do_send)
+    {
+      int r;
+      assert(nodes.count() == 0);
+      if (use_master_node)
+        nodeId= m_master_node;
+      if ((r= okToSendTo(nodeId, true)) != 0)
+      {
+        bool next;
+        if (!use_master_node)
+          DBUG_RETURN(r);
+        m_master_node= nodeId= 0;
+        while((next= getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_NDB)) == true &&
+              (r= okToSendTo(nodeId, true)) != 0);
+        if (!next)
+          DBUG_RETURN(NO_CONTACT_WITH_DB_NODES);
+      }
+      if (ss.sendSignal(nodeId, &ssig) != SEND_OK)
+        DBUG_RETURN(SEND_OR_RECEIVE_FAILED);
+      nodes.set(nodeId);
+      do_send= 0;
+    }
     SimpleSignal *signal = ss.waitFor();
     int gsn = signal->readSignalNumber();
     switch (gsn) {
@@ -1040,6 +1080,13 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
 #endif
       assert(nodes.get(nodeId));
       nodes.clear(nodeId);
+      if (ref->errorCode == StopRef::MultiNodeShutdownNotMaster)
+      {
+        assert(use_master_node);
+        m_master_node= ref->masterNodeId;
+        do_send= 1;
+        continue;
+      }
       error = translateStopRef(ref->errorCode);
       break;
     }
@@ -1050,9 +1097,16 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
       ndbout_c("Node %d single user mode", nodeId);
 #endif
       assert(nodes.get(nodeId));
-      assert(singleUserNodeId != 0);
+      if (singleUserNodeId != 0)
+      {
+        stoppedNodes.set(nodeId);
+      }
+      else
+      {
+        assert(no_of_nodes_to_stop > 1);
+        stoppedNodes.bitOR(nodes_to_stop);
+      }
       nodes.clear(nodeId);
-      stoppedNodes.set(nodeId);
       break;
     }
     case GSN_NF_COMPLETEREP:{
@@ -1091,17 +1145,24 @@ int MgmtSrvr::sendSTOP_REQ(NodeId nodeId,
 #ifdef VM_TRACE
       ndbout_c("Unknown signal %d", gsn);
 #endif
-      return SEND_OR_RECEIVE_FAILED;
+      DBUG_RETURN(SEND_OR_RECEIVE_FAILED);
     }
   }
-  return error;
+  if (!error && do_stop_self)
+  {
+    if (restart)
+      g_RestartServer= true;
+    g_StopServer= true;
+  }
+  DBUG_RETURN(error);
 }
 
 /*
- * Stop one node
+ * Stop one nodes
  */
 
-int MgmtSrvr::stopNode(int nodeId, bool abort)
+int MgmtSrvr::stopNodes(const Vector<NodeId> &node_ids,
+                        int *stopCount, bool abort)
 {
   if (!abort)
   {
@@ -1116,14 +1177,17 @@ int MgmtSrvr::stopNode(int nodeId, bool abort)
     }
   }
   NodeBitmask nodes;
-  return sendSTOP_REQ(nodeId,
+  int ret= sendSTOP_REQ(node_ids,
-		      nodes,
+                        nodes,
-		      0,
+                        0,
-		      abort,
+                        abort,
-		      false,
+                        false,
-		      false,
+                        false,
-		      false,
+                        false,
-		      false);
+                        false);
+  if (stopCount)
+    *stopCount= nodes.count();
+  return ret;
 }
 
 /*
@@ -1133,7 +1197,8 @@ int MgmtSrvr::stopNode(int nodeId, bool abort)
 int MgmtSrvr::stop(int * stopCount, bool abort)
 {
   NodeBitmask nodes;
-  int ret = sendSTOP_REQ(0,
+  Vector<NodeId> node_ids;
+  int ret = sendSTOP_REQ(node_ids,
 			 nodes,
 			 0,
 			 abort,
@@ -1164,7 +1229,8 @@ int MgmtSrvr::enterSingleUser(int * stopCount, Uint32 singleUserNodeId)
       return OPERATION_NOT_ALLOWED_START_STOP;
   }
   NodeBitmask nodes;
-  int ret = sendSTOP_REQ(0,
+  Vector<NodeId> node_ids;
+  int ret = sendSTOP_REQ(node_ids,
 			 nodes,
 			 singleUserNodeId,
 			 false,
@@ -1181,18 +1247,22 @@ int MgmtSrvr::enterSingleUser(int * stopCount, Uint32 singleUserNodeId)
  * Perform node restart
  */
 
-int MgmtSrvr::restartNode(int nodeId, bool nostart, bool initialStart, 
+int MgmtSrvr::restartNodes(const Vector<NodeId> &node_ids,
-			  bool abort)
+                           int * stopCount, bool nostart,
+                           bool initialStart, bool abort)
 {
   NodeBitmask nodes;
-  return sendSTOP_REQ(nodeId,
+  int ret= sendSTOP_REQ(node_ids,
-		      nodes,
+                        nodes,
-		      0,
+                        0,
-		      abort,
+                        abort,
-		      false,
+                        false,
-		      true,
+                        true,
-		      nostart,
+                        nostart,
-		      initialStart);
+                        initialStart);
+  if (stopCount)
+    *stopCount = nodes.count();
+  return ret;
 }
 
 /*
@@ -1203,7 +1273,8 @@ int MgmtSrvr::restart(bool nostart, bool initialStart,
 		      bool abort, int * stopCount )
 {
   NodeBitmask nodes;
-  int ret = sendSTOP_REQ(0,
+  Vector<NodeId> node_ids;
+  int ret = sendSTOP_REQ(node_ids,
 			 nodes,
 			 0,
 			 abort,
@@ -2241,12 +2312,16 @@ MgmtSrvr::startBackup(Uint32& backupId, int waitCompleted)
   SignalSender ss(theFacade);
   ss.lock(); // lock will be released on exit
 
-  bool next;
+  NodeId nodeId = m_master_node;
-  NodeId nodeId = 0;
+  if (okToSendTo(nodeId, false) != 0)
-  while((next = getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_NDB)) == true &&
+  {
-	theFacade->get_node_alive(nodeId) == false);
+    bool next;
-  
+    nodeId = m_master_node = 0;
-  if(!next) return NO_CONTACT_WITH_DB_NODES;
+    while((next = getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_NDB)) == true &&
+          okToSendTo(nodeId, false) != 0);
+    if(!next)
+      return NO_CONTACT_WITH_DB_NODES;
+  }
 
   SimpleSignal ssig;
   BackupReq* req = CAST_PTR(BackupReq, ssig.getDataPtrSend());
@@ -2314,7 +2389,7 @@ MgmtSrvr::startBackup(Uint32& backupId, int waitCompleted)
       const BackupRef * const ref = 
 	CAST_CONSTPTR(BackupRef, signal->getDataPtr());
       if(ref->errorCode == BackupRef::IAmNotMaster){
-	nodeId = refToNode(ref->masterRef);
+	m_master_node = nodeId = refToNode(ref->masterRef);
 #ifdef VM_TRACE
 	ndbout_c("I'm not master resending to %d", nodeId);
 #endif

storage/ndb/src/mgmsrv/MgmtSrvr.hpp

@@ -176,6 +176,7 @@ public:
   STATIC_CONST( NODE_SHUTDOWN_WOULD_CAUSE_SYSTEM_CRASH = 5028 );
 
   STATIC_CONST( NO_CONTACT_WITH_DB_NODES = 5030 );
+  STATIC_CONST( UNSUPPORTED_NODE_SHUTDOWN = 5031 );
 
   STATIC_CONST( NODE_NOT_API_NODE = 5062 );
   STATIC_CONST( OPERATION_NOT_ALLOWED_START_STOP = 5063 );
@@ -252,7 +253,7 @@ public:
    *   @param   processId: Id of the DB process to stop
    *   @return  0 if succeeded, otherwise: as stated above, plus:
    */
-  int stopNode(int nodeId, bool abort = false);
+  int stopNodes(const Vector<NodeId> &node_ids, int *stopCount, bool abort);
 
   /**
    *   Stop the system
@@ -286,11 +287,12 @@ public:
  int start(int processId);
 
   /**
-   *   Restart a node
+   *   Restart nodes
    *   @param processId: Id of the DB process to start
    */
-  int restartNode(int processId, bool nostart, bool initialStart, 
+  int restartNodes(const Vector<NodeId> &node_ids,
-		  bool abort = false);
+                   int *stopCount, bool nostart,
+                   bool initialStart, bool abort);
   
   /**
    *   Restart the system
@@ -489,7 +491,7 @@ private:
                    bool nostart,
                    bool initialStart);
 
-  int sendSTOP_REQ(NodeId nodeId,
+  int sendSTOP_REQ(const Vector<NodeId> &node_ids,
 		   NodeBitmask &stoppedNodes,
 		   Uint32 singleUserNodeId,
 		   bool abort,
@@ -649,6 +651,8 @@ private:
   friend class Ndb_mgmd_event_service;
   Ndb_mgmd_event_service m_event_listner;
   
+  NodeId m_master_node;
+
   /**
    * Handles the thread wich upon a 'Node is started' event will
    * set the node's previous loglevel settings.

storage/ndb/src/mgmsrv/Services.cpp

@@ -866,14 +866,11 @@ MgmApiSession::restart(Parser<MgmApiSession>::Context &,
   }
 
   int restarted = 0;
-  int result = 0;
+  int result= m_mgmsrv.restartNodes(nodes,
-
+                                    &restarted,
-  for(size_t i = 0; i < nodes.size(); i++)
+                                    nostart != 0,
-    if((result = m_mgmsrv.restartNode(nodes[i],
+                                    initialstart != 0,
-				      nostart != 0,
+                                    abort != 0);
-				      initialstart != 0,
-				      abort != 0)) == 0)
-      restarted++;
   
   m_output->println("restart reply");
   if(result != 0){
@@ -998,7 +995,12 @@ MgmApiSession::stop(Parser<MgmApiSession>::Context &,
 
   args.get("node", (const char **)&nodes_str);
   if(nodes_str == NULL)
+  {
+    m_output->println("stop reply");
+    m_output->println("result: empty node list");
+    m_output->println("");
     return;
+  }
   args.get("abort", &abort);
 
   char *p, *last;
@@ -1008,29 +1010,10 @@ MgmApiSession::stop(Parser<MgmApiSession>::Context &,
     nodes.push_back(atoi(p));
   }
 
-  int stop_self= 0;
+  int stopped= 0;
-  size_t i;
+  int result= 0;
-
+  if (nodes.size())
-  for(i=0; i < nodes.size(); i++) {
+    result= m_mgmsrv.stopNodes(nodes, &stopped, abort != 0);
-    if (nodes[i] == m_mgmsrv.getOwnNodeId()) {
-      stop_self= 1;
-      if (i != nodes.size()-1) {
-	m_output->println("stop reply");
-	m_output->println("result: server must be stopped last");
-	m_output->println("");
-	return;
-      }
-    }
-  }
-
-  int stopped = 0, result = 0;
-  
-  for(i=0; i < nodes.size(); i++)
-    if (nodes[i] != m_mgmsrv.getOwnNodeId()) {
-      if((result = m_mgmsrv.stopNode(nodes[i], abort != 0)) == 0)
-	stopped++;
-    } else
-      stopped++;
 
   m_output->println("stop reply");
   if(result != 0)
@@ -1039,9 +1022,6 @@ MgmApiSession::stop(Parser<MgmApiSession>::Context &,
     m_output->println("result: Ok");
   m_output->println("stopped: %d", stopped);
   m_output->println("");
-
-  if (stop_self)
-    g_StopServer= true;
 }
 
 

storage/ndb/src/ndbapi/NdbDictionaryImpl.cpp

@@ -1789,9 +1789,14 @@ NdbDictInterface::getTable(const BaseString& name, bool fullyQualifiedNames)
 
   // Copy name to m_buffer to get a word sized buffer
   m_buffer.clear();
-  m_buffer.grow(namelen_words*4);
+  m_buffer.grow(namelen_words*4+4);
   m_buffer.append(name.c_str(), namelen);
 
+#ifndef IGNORE_VALGRIND_WARNINGS
+  Uint32 pad = 0;
+  m_buffer.append(&pad, 4);
+#endif
+  
   LinearSectionPtr ptr[1];
   ptr[0].p= (Uint32*)m_buffer.get_data();
   ptr[0].sz= namelen_words;

storage/ndb/src/ndbapi/NdbLinHash.hpp

@@ -427,19 +427,26 @@ NdbLinHash<C>::getNext(NdbElement_t<C> * curr){
     return curr->next;
   
   int dir = 0, seg = 0;
-
+  int counts;
-  if(curr != 0){
+  if(curr != 0)
+  {
     getBucket(curr->hash, &dir, &seg);
+    counts = seg + 1;
+  }
+  else
+  {
+    counts = 0;
   }
   
   for(int countd = dir; countd < DIRECTORYSIZE;countd++ ){
     if (directory[countd] != 0) {
-      for(int counts = seg + 1; counts < SEGMENTSIZE; counts++ ){
+      for(; counts < SEGMENTSIZE; counts++ ){
 	if (directory[countd]->elements[counts] != 0) {
 	  return directory[countd]->elements[counts];
 	}   
       }
     }
+    counts = 0;
   }
 
   return 0;

storage/ndb/test/ndbapi/testNodeRestart.cpp

@@ -22,7 +22,7 @@
 #include <NdbRestarts.hpp>
 #include <Vector.hpp>
 #include <signaldata/DumpStateOrd.hpp>
-
+#include <Bitmask.hpp>
 
 int runLoadTable(NDBT_Context* ctx, NDBT_Step* step){
 
@@ -669,6 +669,206 @@ err:
   return NDBT_FAILED;    
 }
 
+int 
+runBug18612(NDBT_Context* ctx, NDBT_Step* step){
+
+  // Assume two replicas
+  NdbRestarter restarter;
+  if (restarter.getNumDbNodes() < 2)
+  {
+    ctx->stopTest();
+    return NDBT_OK;
+  }
+
+  Uint32 cnt = restarter.getNumDbNodes();
+
+  for(int loop = 0; loop < ctx->getNumLoops(); loop++)
+  {
+    int partition0[256];
+    int partition1[256];
+    bzero(partition0, sizeof(partition0));
+    bzero(partition1, sizeof(partition1));
+    Bitmask<4> nodesmask;
+    
+    Uint32 node1 = restarter.getDbNodeId(rand()%cnt);
+    for (Uint32 i = 0; i<cnt/2; i++)
+    {
+      do { 
+	int tmp = restarter.getRandomNodeOtherNodeGroup(node1, rand());
+	if (tmp == -1)
+	  break;
+	node1 = tmp;
+      } while(nodesmask.get(node1));
+      
+      partition0[i] = node1;
+      partition1[i] = restarter.getRandomNodeSameNodeGroup(node1, rand());
+      
+      ndbout_c("nodes %d %d", node1, partition1[i]);
+      
+      assert(!nodesmask.get(node1));
+      assert(!nodesmask.get(partition1[i]));
+      nodesmask.set(node1);
+      nodesmask.set(partition1[i]);
+    } 
+    
+    ndbout_c("done");
+
+    int dump[255];
+    dump[0] = DumpStateOrd::NdbcntrStopNodes;
+    memcpy(dump + 1, partition0, sizeof(int)*cnt/2);
+    
+    Uint32 master = restarter.getMasterNodeId();
+    
+    if (restarter.dumpStateOneNode(master, dump, 1+cnt/2))
+      return NDBT_FAILED;
+    
+    if (restarter.waitNodesNoStart(partition0, cnt/2))
+      return NDBT_FAILED;
+
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+    
+    if (restarter.dumpStateAllNodes(val2, 2))
+      return NDBT_FAILED;
+    
+    if (restarter.insertErrorInAllNodes(932))
+      return NDBT_FAILED;
+
+    dump[0] = 9000;
+    memcpy(dump + 1, partition0, sizeof(int)*cnt/2);    
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateOneNode(partition1[i], dump, 1+cnt/2))
+	return NDBT_FAILED;
+
+    dump[0] = 9000;
+    memcpy(dump + 1, partition1, sizeof(int)*cnt/2);    
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateOneNode(partition0[i], dump, 1+cnt/2))
+	return NDBT_FAILED;
+    
+    if (restarter.startNodes(partition0, cnt/2))
+      return NDBT_FAILED;
+    
+    if (restarter.waitNodesStartPhase(partition0, cnt/2, 2))
+      return NDBT_FAILED;
+    
+    dump[0] = 9001;
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateAllNodes(dump, 2))
+	return NDBT_FAILED;
+
+    if (restarter.waitNodesNoStart(partition0, cnt/2))
+      return NDBT_FAILED;
+    
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.restartOneDbNode(partition0[i], true, true, true))
+	return NDBT_FAILED;
+    
+    if (restarter.waitNodesNoStart(partition0, cnt/2))
+      return NDBT_FAILED;
+    
+    if (restarter.startAll())
+      return NDBT_FAILED;
+
+    if (restarter.waitClusterStarted())
+      return NDBT_FAILED;
+  }
+  return NDBT_OK;
+}
+
+int 
+runBug18612SR(NDBT_Context* ctx, NDBT_Step* step){
+
+  // Assume two replicas
+  NdbRestarter restarter;
+  if (restarter.getNumDbNodes() < 2)
+  {
+    ctx->stopTest();
+    return NDBT_OK;
+  }
+
+  Uint32 cnt = restarter.getNumDbNodes();
+
+  for(int loop = 0; loop < ctx->getNumLoops(); loop++)
+  {
+    int partition0[256];
+    int partition1[256];
+    bzero(partition0, sizeof(partition0));
+    bzero(partition1, sizeof(partition1));
+    Bitmask<4> nodesmask;
+    
+    Uint32 node1 = restarter.getDbNodeId(rand()%cnt);
+    for (Uint32 i = 0; i<cnt/2; i++)
+    {
+      do { 
+	int tmp = restarter.getRandomNodeOtherNodeGroup(node1, rand());
+	if (tmp == -1)
+	  break;
+	node1 = tmp;
+      } while(nodesmask.get(node1));
+      
+      partition0[i] = node1;
+      partition1[i] = restarter.getRandomNodeSameNodeGroup(node1, rand());
+      
+      ndbout_c("nodes %d %d", node1, partition1[i]);
+      
+      assert(!nodesmask.get(node1));
+      assert(!nodesmask.get(partition1[i]));
+      nodesmask.set(node1);
+      nodesmask.set(partition1[i]);
+    } 
+    
+    ndbout_c("done");
+
+    if (restarter.restartAll(false, true, false))
+      return NDBT_FAILED;
+
+    int dump[255];
+    dump[0] = 9000;
+    memcpy(dump + 1, partition0, sizeof(int)*cnt/2);    
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateOneNode(partition1[i], dump, 1+cnt/2))
+	return NDBT_FAILED;
+
+    dump[0] = 9000;
+    memcpy(dump + 1, partition1, sizeof(int)*cnt/2);    
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateOneNode(partition0[i], dump, 1+cnt/2))
+	return NDBT_FAILED;
+
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+    
+    if (restarter.dumpStateAllNodes(val2, 2))
+      return NDBT_FAILED;
+    
+    if (restarter.insertErrorInAllNodes(932))
+      return NDBT_FAILED;
+    
+    if (restarter.startAll())
+      return NDBT_FAILED;
+    
+    if (restarter.waitClusterStartPhase(2))
+      return NDBT_FAILED;
+    
+    dump[0] = 9001;
+    for (Uint32 i = 0; i<cnt/2; i++)
+      if (restarter.dumpStateAllNodes(dump, 2))
+	return NDBT_FAILED;
+
+    if (restarter.waitClusterNoStart(30))
+      if (restarter.waitNodesNoStart(partition0, cnt/2, 10))
+	if (restarter.waitNodesNoStart(partition1, cnt/2, 10))
+	  return NDBT_FAILED;
+    
+    if (restarter.startAll())
+      return NDBT_FAILED;
+    
+    if (restarter.waitClusterStarted())
+      return NDBT_FAILED;
+  }
+  return NDBT_OK;
+}
+
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -963,6 +1163,18 @@ TESTCASE("Bug18414",
   STEP(runBug18414);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug18612",
+	 "Test bug with partitioned clusters"){
+  INITIALIZER(runLoadTable);
+  STEP(runBug18612);
+  FINALIZER(runClearTable);
+}
+TESTCASE("Bug18612SR",
+	 "Test bug with partitioned clusters"){
+  INITIALIZER(runLoadTable);
+  STEP(runBug18612SR);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

storage/ndb/test/run-test/daily-basic-tests.txt

@@ -453,10 +453,18 @@ args: -n Bug16772 T1
 #cmd: testSystemRestart
 #args: -n Bug18385 T1
 #
-max-time: 500
+max-time: 1000
 cmd: testNodeRestart
 args: -n Bug18414 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug18612 T1
+
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug18612SR T1
+
 #
 # DICT TESTS
 max-time: 1500

storage/ndb/test/src/NdbRestarts.cpp

@@ -445,8 +445,7 @@ int twoNodeFailure(NdbRestarter& _restarter,
 	 << ") secs " << endl;
   NdbSleep_SecSleep(seconds);
 
-  randomId = (rand() % _restarter.getNumDbNodes());
+  nodeId = _restarter.getRandomNodeOtherNodeGroup(nodeId, rand());
-  nodeId = _restarter.getDbNodeId(randomId);  
   g_info << _restart->m_name << ": node = "<< nodeId << endl;
 
   CHECK(_restarter.insertErrorInNode(nodeId, 9999) == 0,