ndb - wl1760/bug#18216

  add two new start options that will decrease likelyhood of bug#18612
  push cntr-sp2 logic down into qmgr-sp1 to decrease likelyhood of bug#18612


ndb/include/kernel/signaldata/CmRegSignalData.hpp:
  Expand CmRegReq with lots of stuff
ndb/include/mgmapi/ndb_logevent.h:
  Add Start report during sp1
ndb/src/common/debugger/EventLogger.cpp:
  Add Start report during sp1
ndb/src/kernel/blocks/dbdih/DbdihInit.cpp:
  Init cntrref
ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  Fix small bugs related to partial initial start
ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp:
  Remove timeouts as they are handled in QMGR nowadays
ndb/src/kernel/blocks/qmgr/Qmgr.hpp:
  Push cntr sp2 logic down to QMGR to prevent the forming of multiple QMGR cluster
    that will lead to start problems (crashes...)
ndb/src/kernel/blocks/qmgr/QmgrInit.cpp:
  Push cntr sp2 logic down to QMGR to prevent the forming of multiple QMGR cluster
    that will lead to start problems (crashes...)
ndb/src/kernel/blocks/qmgr/QmgrMain.cpp:
  Push cntr sp2 logic down to QMGR to prevent the forming of multiple QMGR cluster
    that will lead to start problems (crashes...)
ndb/src/kernel/vm/Configuration.cpp:
  Add new flags
    --initial-start
    --nowait-nodes

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 {

ndb/include/mgmapi/ndb_logevent.h

@@ -166,10 +166,14 @@ extern "C" {
     /** NDB_MGM_EVENT_CATEGORY_BACKUP */
     NDB_LE_BackupCompleted = 56,
     /** NDB_MGM_EVENT_CATEGORY_BACKUP */
-    NDB_LE_BackupAborted = 57
+    NDB_LE_BackupAborted = 57,
 
     /* 58 used in 5.1  */
     /* 59 used */
+
+    /** NDB_MGM_EVENT_CATEGORY_STARTUP */
+    NDB_LE_StartReport = 60
+
     /* 60 unused */
     /* 61 unused */
     /* 62 unused */
@@ -625,6 +629,13 @@ 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

ndb/src/common/debugger/EventLogger.cpp

@@ -707,6 +707,90 @@ 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, 
@@ -755,6 +839,7 @@ 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 ),

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

@@ -71,6 +71,7 @@ void Dbdih::initData()
   cwaitLcpSr       = false;
   c_blockCommit    = false;
   c_blockCommitNo  = 1;
+  cntrlblockref    = RNIL;
 }//Dbdih::initData()
 
 void Dbdih::initRecords() 

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

@@ -11659,7 +11659,7 @@ 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 
@@ -11705,6 +11705,38 @@ 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()
 
 /**
@@ -12512,7 +12544,6 @@ void Dbdih::sendStartFragreq(Signal* signal,
 void Dbdih::setInitialActiveStatus()
 {
   NodeRecordPtr siaNodeptr;
-  Uint32 tsiaNodeActiveStatus;
   Uint32 tsiaNoActiveNodes;
 
   tsiaNoActiveNodes = csystemnodes - cnoHotSpare;
@@ -12520,39 +12551,34 @@ 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
-    } else {
-      jam();
-      siaNodeptr.p->activeStatus = Sysfile::NS_NotDefined;
-    }//if
-    switch (siaNodeptr.p->activeStatus) {
-    case Sysfile::NS_Active:
-      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;
+        if (siaNodeptr.p->nodeStatus == NodeRecord::ALIVE)
+	{
+	  jam();
+	  siaNodeptr.p->activeStatus = Sysfile::NS_Active;
+	} 
+	else
+	{
+	  siaNodeptr.p->activeStatus = Sysfile::NS_NotActive_NotTakenOver;
+	}
+      }
       break;
     default:
-      ndbrequire(false);
-      return;
+      jam();
+      siaNodeptr.p->activeStatus = Sysfile::NS_NotDefined;
       break;
-    }//switch
-    Sysfile::setNodeStatus(siaNodeptr.i, SYSFILE->nodeStatus,
-                           tsiaNodeActiveStatus);
+    }//if
+    Sysfile::setNodeStatus(siaNodeptr.i, 
+			   SYSFILE->nodeStatus,
+                           siaNodeptr.p->activeStatus);
   }//for
 }//Dbdih::setInitialActiveStatus()
 

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

@@ -820,17 +820,9 @@ 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(!allNodes){
-      jam();
-      return false;
-    }
+  if(c_start.m_waiting.equal(c_start.m_withoutLog))
+  {
     jam();
     srType = NodeState::ST_INITIAL_START;
     c_start.m_starting = c_start.m_withoutLog; // Used for starting...
@@ -860,10 +852,6 @@ Ndbcntr::trySystemRestart(Signal* signal){
 	ndbrequire(false); // All nodes -> partitioning, which is not allowed
       }
       
-      if(c_start.m_startPartitionedTimeout > now){
-	jam();
-	return false;
-      }
       break;
     }    
     

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

@@ -50,6 +50,7 @@
 #define ZAPI_HB_HANDLING 3
 #define ZTIMER_HANDLING 4
 #define ZARBIT_HANDLING 5
+#define ZSTART_FAILURE_LIMIT 6
 
 /* Error Codes ------------------------------*/
 #define ZERRTOOMANY 1101
@@ -113,8 +114,19 @@ public:
     
     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
@@ -125,7 +137,7 @@ public:
    * i.e. nodes that connect to use, when we already have elected president
    */
   NdbNodeBitmask c_readnodes_nodes;
-  
+
   Uint32 c_maxDynamicId;
   
   // Records
@@ -236,6 +248,9 @@ private:
   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);
 
@@ -252,6 +267,7 @@ private:
 
   // 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);
@@ -374,12 +390,12 @@ 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;

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

@@ -98,6 +98,9 @@ Qmgr::Qmgr(const class Configuration & conf)
 
   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()

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;
 }