Merge tulin@bk-internal.mysql.com:/home/bk/mysql-5.0-ndb

into  poseidon.mysql.com:/home/tomas/mysql-5.0-ndb

ndb/src/kernel/blocks/ERROR_codes.txt

@@ -5,7 +5,7 @@ Next DBACC 3002
 Next DBTUP 4014
 Next DBLQH 5043
 Next DBDICT 6007
-Next DBDIH 7178
+Next DBDIH 7181
 Next DBTC 8039
 Next CMVMI 9000
 Next BACKUP 10022
@@ -71,6 +71,8 @@ Delay GCP_SAVEREQ by 10 secs
 
 7177: Delay copying of sysfileData in execCOPY_GCIREQ
 
+7180: Crash master during master-take-over in execMASTER_LCPCONF
+
 ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
 -----------------------------------------------------------------
 

ndb/src/kernel/blocks/dbdih/Dbdih.hpp

@@ -1366,6 +1366,7 @@ private:
   Uint32 csystemnodes;
   Uint32 currentgcp;
   Uint32 c_newest_restorable_gci;
+  Uint32 c_set_initial_start_flag;
 
   enum GcpMasterTakeOverState {
     GMTOS_IDLE = 0,

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

@@ -74,6 +74,7 @@ void Dbdih::initData()
   c_blockCommit    = false;
   c_blockCommitNo  = 1;
   cntrlblockref    = RNIL;
+  c_set_initial_start_flag = FALSE;
 }//Dbdih::initData()
 
 void Dbdih::initRecords() 

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

@@ -666,6 +666,12 @@ void Dbdih::execCOPY_GCIREQ(Signal* signal)
   {
     jam();
     memcpy(sysfileData, cdata, sizeof(sysfileData));
+
+    if (c_set_initial_start_flag)
+    {
+      jam();
+      Sysfile::setInitialStartOngoing(SYSFILE->systemRestartBits);
+    }
   }
 
   c_copyGCISlave.m_copyReason = reason;
@@ -1259,6 +1265,11 @@ void Dbdih::execNDB_STTOR(Signal* signal)
     // The permission is given by the master node in the alive set.  
     /*-----------------------------------------------------------------------*/
     createMutexes(signal, 0);
+    if (cstarttype == NodeState::ST_INITIAL_NODE_RESTART)
+    {
+      jam();
+      c_set_initial_start_flag = TRUE; // In sysfile...
+    }
     break;
     
   case ZNDB_SPH3:
@@ -4612,6 +4623,8 @@ void
 Dbdih::startLcpMasterTakeOver(Signal* signal, Uint32 nodeId){
   jam();
 
+  Uint32 oldNode = c_lcpMasterTakeOverState.failedNodeId;
+
   c_lcpMasterTakeOverState.minTableId = ~0;
   c_lcpMasterTakeOverState.minFragId = ~0;
   c_lcpMasterTakeOverState.failedNodeId = nodeId;
@@ -4630,7 +4643,20 @@ Dbdih::startLcpMasterTakeOver(Signal* signal, Uint32 nodeId){
     /**
      * Node failure during master take over...
      */
-    g_eventLogger.info("Nodefail during master take over");
+    g_eventLogger.info("Nodefail during master take over (old: %d)", oldNode);
+  }
+  
+  NodeRecordPtr nodePtr;
+  nodePtr.i = oldNode;
+  if (oldNode > 0 && oldNode < MAX_NDB_NODES)
+  {
+    jam();
+    ptrCheckGuard(nodePtr, MAX_NDB_NODES, nodeRecord);
+    if (nodePtr.p->m_nodefailSteps.get(NF_LCP_TAKE_OVER))
+    {
+      jam();
+      checkLocalNodefailComplete(signal, oldNode, NF_LCP_TAKE_OVER);
+    }
   }
   
   setLocalNodefailHandling(signal, nodeId, NF_LCP_TAKE_OVER);
@@ -5647,6 +5673,14 @@ void Dbdih::execMASTER_LCPREQ(Signal* signal)
   jamEntry();
   const BlockReference newMasterBlockref = req->masterRef;
 
+  if (newMasterBlockref != cmasterdihref)
+  {
+    jam();
+    ndbout_c("resending GSN_MASTER_LCPREQ");
+    sendSignalWithDelay(reference(), GSN_MASTER_LCPREQ, signal,
+			signal->getLength(), 50);
+    return;
+  }
   Uint32 failedNodeId = req->failedNodeId;
 
   /**
@@ -5947,6 +5981,8 @@ void Dbdih::execMASTER_LCPCONF(Signal* signal)
   ptrCheckGuard(nodePtr, MAX_NDB_NODES, nodeRecord);
   nodePtr.p->lcpStateAtTakeOver = lcpState;
 
+  CRASH_INSERTION(7180);
+  
 #ifdef VM_TRACE
   g_eventLogger.info("MASTER_LCPCONF");
   printMASTER_LCP_CONF(stdout, &signal->theData[0], 0, 0);
@@ -10259,6 +10295,17 @@ Dbdih::sendLCP_COMPLETE_REP(Signal* signal){
   
   sendSignal(c_lcpState.m_masterLcpDihRef, GSN_LCP_COMPLETE_REP, signal, 
 	     LcpCompleteRep::SignalLength, JBB);
+
+  /**
+   * Say that an initial node restart does not need to be redone
+   *   once node has been part of first LCP
+   */
+  if (c_set_initial_start_flag &&
+      c_lcpState.m_participatingLQH.get(getOwnNodeId()))
+  {
+    jam();
+    c_set_initial_start_flag = FALSE;
+  }
 }
 
 /*-------------------------------------------------------------------------- */

ndb/src/kernel/blocks/dblqh/DblqhMain.cpp

@@ -11672,7 +11672,8 @@ void Dblqh::execGCP_SAVEREQ(Signal* signal)
     return;
   }
 
-  if(getNodeState().getNodeRestartInProgress()){
+  if(getNodeState().getNodeRestartInProgress() && cstartRecReq == ZFALSE)
+  {
     GCPSaveRef * const saveRef = (GCPSaveRef*)&signal->theData[0];
     saveRef->dihPtr = dihPtr;
     saveRef->nodeId = getOwnNodeId();

ndb/test/include/NdbRestarter.hpp

@@ -61,6 +61,8 @@ public:
   int dumpStateAllNodes(int * _args, int _num_args);
 
   int getMasterNodeId();
+  int getNextMasterNodeId(int nodeId);
+  int getNodeGroup(int nodeId);
   int getRandomNodeSameNodeGroup(int nodeId, int randomNumber);
   int getRandomNodeOtherNodeGroup(int nodeId, int randomNumber);
   int getRandomNotMasterNodeId(int randomNumber);

ndb/test/ndbapi/testBitfield.cpp

@@ -8,6 +8,15 @@
 static const char* _dbname = "TEST_DB";
 static int g_loops = 7;
 
+
+NDB_STD_OPTS_VARS;
+
+static struct my_option my_long_options[] =
+{
+  NDB_STD_OPTS("ndb_desc"),
+  { 0, 0, 0, 0, 0, 0, GET_NO_ARG, NO_ARG, 0, 0, 0, 0, 0, 0}
+};
+
 static void usage()
 {
   ndb_std_print_version();
@@ -36,8 +45,9 @@ main(int argc, char** argv){
   load_defaults("my",load_default_groups,&argc,&argv);
   int ho_error;
 
-  argc--;
-  argv++;
+  if ((ho_error=handle_options(&argc, &argv, my_long_options,
+			       ndb_std_get_one_option)))
+    return NDBT_ProgramExit(NDBT_WRONGARGS);
 
   Ndb_cluster_connection con(opt_connect_str);
   if(con.connect(12, 5, 1))

ndb/test/ndbapi/testNodeRestart.cpp

@@ -1044,6 +1044,85 @@ int runBug25554(NDBT_Context* ctx, NDBT_Step* step){
   return NDBT_OK;
 }
 
+int
+runBug26457(NDBT_Context* ctx, NDBT_Step* step)
+{
+  NdbRestarter res;
+  if (res.getNumDbNodes() < 4)
+    return NDBT_OK;
+
+  int loops = ctx->getNumLoops();
+  while (loops --)
+  {
+retry:
+    int master = res.getMasterNodeId();
+    int next = res.getNextMasterNodeId(master);
+
+    ndbout_c("master: %d next: %d", master, next);
+
+    if (res.getNodeGroup(master) == res.getNodeGroup(next))
+    {
+      res.restartOneDbNode(next, false, false, true);
+      if (res.waitClusterStarted())
+	return NDBT_FAILED;
+      goto retry;
+    }
+
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 2 };
+    
+    if (res.dumpStateOneNode(next, val2, 2))
+      return NDBT_FAILED;
+    
+    if (res.insertErrorInNode(next, 7180))
+      return NDBT_FAILED;
+    
+    res.restartOneDbNode(master, false, false, true);
+    if (res.waitClusterStarted())
+      return NDBT_FAILED;
+  }
+  
+  return NDBT_OK;
+}
+
+int 
+runBug26481(NDBT_Context* ctx, NDBT_Step* step)
+{
+  
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter res;
+  
+  int node = res.getRandomNotMasterNodeId(rand());
+  ndbout_c("node: %d", node);
+  if (res.restartOneDbNode(node, true, true, true))
+    return NDBT_FAILED;
+
+  if (res.waitNodesNoStart(&node, 1))
+    return NDBT_FAILED;
+
+  int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  if (res.dumpStateOneNode(node, val2, 2))
+    return NDBT_FAILED;
+
+  if (res.insertErrorInNode(node, 7018))
+    return NDBT_FAILED;
+
+  if (res.startNodes(&node, 1))
+    return NDBT_FAILED;
+
+  res.waitNodesStartPhase(&node, 1, 3);
+  
+  if (res.waitNodesNoStart(&node, 1))
+    return NDBT_FAILED;
+
+  res.startNodes(&node, 1);
+  
+  if (res.waitClusterStarted())
+    return NDBT_FAILED;
+  
+  return NDBT_OK;
+}
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -1366,6 +1445,12 @@ TESTCASE("Bug25364", ""){
 TESTCASE("Bug25554", ""){
   INITIALIZER(runBug25554);
 }
+TESTCASE("Bug26457", ""){
+  INITIALIZER(runBug26457);
+}
+TESTCASE("Bug26481", ""){
+  INITIALIZER(runBug26481);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -477,6 +477,14 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug25554 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug26457 T1
+
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug26481 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench

ndb/test/src/NdbRestarter.cpp

@@ -127,6 +127,68 @@ NdbRestarter::getMasterNodeId(){
   return node;
 }
 
+int
+NdbRestarter::getNodeGroup(int nodeId){
+  if (!isConnected())
+    return -1;
+  
+  if (getStatus() != 0)
+    return -1;
+  
+  for(size_t i = 0; i < ndbNodes.size(); i++)
+  {
+    if(ndbNodes[i].node_id == nodeId)
+    {
+      return ndbNodes[i].node_group;
+    }
+  }
+  
+  return -1;
+}
+
+int
+NdbRestarter::getNextMasterNodeId(int nodeId){
+  if (!isConnected())
+    return -1;
+  
+  if (getStatus() != 0)
+    return -1;
+  
+  size_t i;
+  for(i = 0; i < ndbNodes.size(); i++)
+  {
+    if(ndbNodes[i].node_id == nodeId)
+    {
+      break;
+    }
+  }
+  assert(i < ndbNodes.size());
+  if (i == ndbNodes.size())
+    return -1;
+
+  int dynid = ndbNodes[i].dynamic_id;
+  int minid = dynid;
+  for (i = 0; i<ndbNodes.size(); i++)
+    if (ndbNodes[i].dynamic_id > minid)
+      minid = ndbNodes[i].dynamic_id;
+  
+  for (i = 0; i<ndbNodes.size(); i++)
+    if (ndbNodes[i].dynamic_id > dynid && 
+	ndbNodes[i].dynamic_id < minid)
+    {
+      minid = ndbNodes[i].dynamic_id;
+    }
+  
+  if (minid != ~0)
+  {
+    for (i = 0; i<ndbNodes.size(); i++)
+      if (ndbNodes[i].dynamic_id == minid)
+	return ndbNodes[i].node_id;
+  }
+  
+  return getMasterNodeId();
+}
+
 int
 NdbRestarter::getRandomNotMasterNodeId(int rand){
   int master = getMasterNodeId();