dbtcmain.cpp

mysql-5.0.22.tar.gz源码包

  cmasterNodeId = readNodes->masterNodeId;

  con_lineNodes = 0;
  arrGuard(csystemnodes, MAX_NDB_NODES);
  guard0 = csystemnodes - 1;
  arrGuard(guard0, MAX_NDB_NODES);       // Check not zero nodes

  for (unsigned i = 1; i < MAX_NDB_NODES; i++) {
    jam();
    if (NodeBitmask::get(readNodes->allNodes, i)) {
      hostptr.i = i;
      ptrCheckGuard(hostptr, chostFilesize, hostRecord);

      if (NodeBitmask::get(readNodes->inactiveNodes, i)) {
        jam();
        hostptr.p->hostStatus = HS_DEAD;
      } else {
        jam();
        con_lineNodes++;
        hostptr.p->hostStatus = HS_ALIVE;
	c_alive_nodes.set(i);
      }//if
    }//if
  }//for
  ndbsttorry010Lab(signal);
}//Dbtc::execREAD_NODESCONF()

/*****************************************************************************/
/*                     A P I _ F A I L R E Q                                 */
// An API node has failed for some reason. We need to disconnect all API 
// connections to the API node. This also includes 
/*****************************************************************************/
void Dbtc::execAPI_FAILREQ(Signal* signal)
{
  /***************************************************************************
   * Set the block reference to return API_FAILCONF to. Set the number of api 
   * connects currently closing to one to indicate that we are still in the 
   * process of going through the api connect records. Thus checking for zero 
   * can only be true after all api connect records have been checked.
   **************************************************************************/
  jamEntry();  
  capiFailRef = signal->theData[1];
  arrGuard(signal->theData[0], MAX_NODES);
  capiConnectClosing[signal->theData[0]] = 1;
  handleFailedApiNode(signal, signal->theData[0], (UintR)0);
}

void
Dbtc::handleFailedApiNode(Signal* signal, 
                          UintR TapiFailedNode, 
                          UintR TapiConnectPtr)
{
  UintR TloopCount = 0;
  arrGuard(TapiFailedNode, MAX_NODES);
  apiConnectptr.i = TapiConnectPtr;
  do {
    ptrCheckGuard(apiConnectptr, capiConnectFilesize, apiConnectRecord);
    const UintR TapiNode = refToNode(apiConnectptr.p->ndbapiBlockref);
    if (TapiNode == TapiFailedNode) {
#ifdef VM_TRACE
      if (apiConnectptr.p->apiFailState != ZFALSE) {
        ndbout << "Error in previous API fail handling discovered" << endl
	       << "  apiConnectptr.i = " << apiConnectptr.i << endl
	       << "  apiConnectstate = " << apiConnectptr.p->apiConnectstate 
	       << endl
	       << "  ndbapiBlockref = " << hex
	       << apiConnectptr.p->ndbapiBlockref << endl
	       << "  apiNode = " << refToNode(apiConnectptr.p->ndbapiBlockref) 
	       << endl;
	if (apiConnectptr.p->lastTcConnect != RNIL){	  
	  jam();
	  tcConnectptr.i = apiConnectptr.p->lastTcConnect;
	  ptrCheckGuard(tcConnectptr, ctcConnectFilesize, tcConnectRecord);
	  ndbout << "  tcConnectptr.i = " << tcConnectptr.i << endl
		 << "  tcConnectstate = " << tcConnectptr.p->tcConnectstate 
		 << endl;
	}
      }//if
#endif
      
      apiConnectptr.p->returnsignal = RS_NO_RETURN;
      /***********************************************************************/
      // The connected node is the failed node.
      /**********************************************************************/
      switch(apiConnectptr.p->apiConnectstate) {
      case CS_DISCONNECTED:
        /*********************************************************************/
        // These states do not need any special handling. 
        // Simply continue with the next.
        /*********************************************************************/
        jam();
        break;
      case CS_ABORTING:
        /*********************************************************************/
        // This could actually mean that the API connection is already 
        // ready to release if the abortState is IDLE.
        /*********************************************************************/
        if (apiConnectptr.p->abortState == AS_IDLE) {
          jam();
          releaseApiCon(signal, apiConnectptr.i);
        } else {
          jam();
          capiConnectClosing[TapiFailedNode]++;
          apiConnectptr.p->apiFailState = ZTRUE;
        }//if
        break;
      case CS_WAIT_ABORT_CONF:
      case CS_WAIT_COMMIT_CONF:
      case CS_START_COMMITTING:
      case CS_PREPARE_TO_COMMIT:
      case CS_COMMITTING:
      case CS_COMMIT_SENT:
        /*********************************************************************/
        // These states indicate that an abort process or commit process is 
        // already ongoing. We will set a state in the api record indicating 
        // that the API node has failed.
        // Also we will increase the number of outstanding api records to 
        // wait for before we can respond with API_FAILCONF.
        /*********************************************************************/
        jam();
        capiConnectClosing[TapiFailedNode]++;
        apiConnectptr.p->apiFailState = ZTRUE;
        break;
      case CS_START_SCAN:
        /*********************************************************************/
        // The api record was performing a scan operation. We need to check 
        // on the scan state. Since completing a scan process might involve
        // sending several signals we will increase the loop count by 64.
        /*********************************************************************/
        jam();

	apiConnectptr.p->apiFailState = ZTRUE;
	capiConnectClosing[TapiFailedNode]++;

	ScanRecordPtr scanPtr;
	scanPtr.i = apiConnectptr.p->apiScanRec;
	ptrCheckGuard(scanPtr, cscanrecFileSize, scanRecord);
	close_scan_req(signal, scanPtr, true);
	
        TloopCount += 64;
        break;
      case CS_CONNECTED:
        /*********************************************************************/
        // The api record is connected to failed node. We need to release the 
        // connection and set it in a disconnected state.
        /*********************************************************************/
        jam();
        releaseApiCon(signal, apiConnectptr.i);
        break;
      case CS_REC_COMMITTING:
      case CS_RECEIVING:
      case CS_STARTED:
        /*********************************************************************/
        // The api record was in the process of performing a transaction but 
        // had not yet sent all information. 
        // We need to initiate an ABORT since the API will not provide any 
        // more information. 
        // Since the abort can send many signals we will insert a real-time
        // break after checking this record.
        /*********************************************************************/
        jam();
        apiConnectptr.p->apiFailState = ZTRUE;
        capiConnectClosing[TapiFailedNode]++;
        abort010Lab(signal);
        TloopCount = 256;
        break;
      case CS_PREPARED:
        jam();
      case CS_REC_PREPARING:
        jam();
      case CS_START_PREPARING:
        jam();
        /*********************************************************************/
        // Not implemented yet.
        /*********************************************************************/
        systemErrorLab(signal, __LINE__);
        break;
      case CS_RESTART:
        jam();
      case CS_COMPLETING:
        jam();
      case CS_COMPLETE_SENT:
        jam();
      case CS_WAIT_COMPLETE_CONF:
        jam();
      case CS_FAIL_ABORTING:
        jam();
      case CS_FAIL_ABORTED:
        jam();
      case CS_FAIL_PREPARED:
        jam();
      case CS_FAIL_COMMITTING:
        jam();
      case CS_FAIL_COMMITTED:
        /*********************************************************************/
        // These states are only valid on copy and fail API connections.
        /*********************************************************************/
      default:
        jam();
        systemErrorLab(signal, __LINE__);
        break;
      }//switch
    } else {
      jam();
    }//if
    apiConnectptr.i++;
    if (apiConnectptr.i > ((capiConnectFilesize / 3) - 1)) {
      jam();
      /**
       * Finished with scanning connection record
       *
       * Now scan markers
       */
      removeMarkerForFailedAPI(signal, TapiFailedNode, 0);
      return;
    }//if
  } while (TloopCount++ < 256);
  signal->theData[0] = TcContinueB::ZHANDLE_FAILED_API_NODE;
  signal->theData[1] = TapiFailedNode;
  signal->theData[2] = apiConnectptr.i;
  sendSignal(cownref, GSN_CONTINUEB, signal, 3, JBB);
}//Dbtc::handleFailedApiNode()

void
Dbtc::removeMarkerForFailedAPI(Signal* signal, 
                               Uint32 nodeId, 
                               Uint32 startBucket)
{
  TcFailRecordPtr node_fail_ptr;
  node_fail_ptr.i = 0;
  ptrAss(node_fail_ptr, tcFailRecord);
  if(node_fail_ptr.p->failStatus != FS_IDLE) {
    jam();
    DEBUG("Restarting removeMarkerForFailedAPI");
    /**
     * TC take-over in progress
     *   needs to restart as this
     *   creates new markers
     */
    signal->theData[0] = TcContinueB::ZHANDLE_FAILED_API_NODE_REMOVE_MARKERS;
    signal->theData[1] = nodeId;
    signal->theData[2] = 0;
    sendSignalWithDelay(cownref, GSN_CONTINUEB, signal, 500, 3);
    return;
  }

  CommitAckMarkerIterator iter;
  m_commitAckMarkerHash.next(startBucket, iter);
  
  const Uint32 RT_BREAK = 256;
  for(Uint32 i = 0; i<RT_BREAK || iter.bucket == startBucket; i++){ 
    jam();
    
    if(iter.curr.i == RNIL){
      jam();
      /**
       * Done with iteration
       */
      capiConnectClosing[nodeId]--;
      if (capiConnectClosing[nodeId] == 0) {
        jam();
        /********************************************************************/
        // No outstanding ABORT or COMMIT's of this failed API node. 
        // We can respond with API_FAILCONF
        /********************************************************************/
        signal->theData[0] = nodeId;
        signal->theData[1] = cownref;
        sendSignal(capiFailRef, GSN_API_FAILCONF, signal, 2, JBB);
      }
      return;
    }
    
    if(iter.curr.p->apiNodeId == nodeId){
      jam();
      
      /**
       * Check so that the record is not still in use
       *
       */
      ApiConnectRecordPtr apiConnectPtr;
      apiConnectPtr.i = iter.curr.p->apiConnectPtr;
      ptrCheckGuard(apiConnectPtr, capiConnectFilesize, apiConnectRecord);
      if(apiConnectPtr.p->commitAckMarker == iter.curr.i){
	jam();
        /**
         * The record is still active
         *
         * Don't remove it, but continueb instead
         */
	break;
      }
      sendRemoveMarkers(signal, iter.curr.p);
      m_commitAckMarkerHash.release(iter.curr);
      
      break;
    } 
    m_commitAckMarkerHash.next(iter);
  }
  
  signal->theData[0] = TcContinueB::ZHANDLE_FAILED_API_NODE_REMOVE_MARKERS;
  signal->theData[1] = nodeId;
  signal->theData[2] = iter.bucket;
  sendSignal(cownref, GSN_CONTINUEB, signal, 3, JBB);
}

void Dbtc::handleApiFailState(Signal* signal, UintR TapiConnectptr)
{
  ApiConnectRecordPtr TlocalApiConnectptr;
  UintR TfailedApiNode;

  TlocalApiConnectptr.i = TapiConnectptr;
  ptrCheckGuard(TlocalApiConnectptr, capiConnectFilesize, apiConnectRecord);
  TfailedApiNode = refToNode(TlocalApiConnectptr.p->ndbapiBlockref);
  arrGuard(TfailedApiNode, MAX_NODES);
  capiConnectClosing[TfailedApiNode]--;
  releaseApiCon(signal, TapiConnectptr);
  TlocalApiConnectptr.p->apiFailState = ZFALSE;
  if (capiConnectClosing[TfailedApiNode] == 0) {
    jam();
    signal->theData[0] = TfailedApiNode;
    signal->theData[1] = cownref;
    sendSignal(capiFailRef, GSN_API_FAILCONF, signal, 2, JBB);
  }//if
}//Dbtc::handleApiFailState()

/****************************************************************************
 *                                T C S E I Z E R E Q
 * THE APPLICATION SENDS A REQUEST TO SEIZE A CONNECT RECORD TO CARRY OUT A 
 * TRANSACTION
 * TC BLOCK TAKE OUT A CONNECT RECORD FROM THE FREE LIST AND ESTABLISHES ALL 
 * NECESSARY CONNECTION BEFORE REPLYING TO THE APPLICATION BLOCK   
 ****************************************************************************/
void Dbtc::execTCSEIZEREQ(Signal* signal) 
{
  UintR tapiPointer;
  BlockReference tapiBlockref;       /* SENDER BLOCK REFERENCE*/
  
  jamEntry();
  tapiPointer = signal->theData[0]; /* REQUEST SENDERS CONNECT RECORD POINTER*/
  tapiBlockref = signal->theData[1]; /* SENDERS BLOCK REFERENCE*/
  
  const NodeState::StartLevel sl = 
    (NodeState::StartLevel)getNodeState().startLevel;

  const NodeId senderNodeId = refToNode(tapiBlockref);
  const bool local = senderNodeId == getOwnNodeId() || senderNodeId == 0;
  
  if(!(senderNodeId == getNodeState().getSingleUserApi()) &&
      !getNodeState().getSingleUserMode()) {
    if(!(sl==NodeState::SL_SINGLEUSER && 
	 senderNodeId == getNodeState().getSingleUserApi())) {
      if (!(sl == NodeState::SL_STARTED ||
	    (sl == NodeState::SL_STARTING && local == true))) {
	jam();
	
	Uint32 errCode;
	if(!(sl == NodeState::SL_SINGLEUSER && local))
	  {
	    switch(sl){
	    case NodeState::SL_STARTING:
	      errCode = ZSYSTEM_NOT_STARTED_ERROR;
	      break;
	    case NodeState::SL_STOPPING_1:
	    case NodeState::SL_STOPPING_2:
	    case NodeState::SL_STOPPING_3:
	    case NodeState::SL_STOPPING_4:
	      if(getNodeState().stopping.systemShutdown)
		errCode = ZCLUSTER_SHUTDOWN_IN_PROGRESS;
	      else
		errCode = ZNODE_SHUTDOWN_IN_PROGRESS;
	      break;
	    case NodeState::SL_SINGLEUSER:
	      errCode = ZCLUSTER_IN_SINGLEUSER_MODE;
	      break;
	    default:
	      errCode = ZWRONG_STATE;
	      break;
	    }
	    signal->theData[0] = tapiPointer;
	    signal->theData[1] = errCode;
	    sendSignal(tapiBlockref, GSN_TCSEIZEREF, signal, 2, JBB);
	    return;
	  }//if (!(sl == SL_SINGLEUSER))
      } //if
    }
  } 
  
  seizeApiConnect(signal);
  if (terrorCode == ZOK) {
    jam();
    apiConnectptr.p->ndbapiConnect = tapiPointer;
    apiConnectptr.p->ndbapiBlockref = tapiBlockref;
    signal->theData[0] = apiConnectptr.p->ndbapiConnect;
    signal->theData[1] = apiConnectptr.i;
    sendSignal(tapiBlockref, GSN_TCSEIZECONF, signal, 2, JBB);
    return;
  }

  signal->theData[0] = tapiPointer;
  signal->theData[1] = terrorCode;
  sendSignal(tapiBlockref, GSN_TCSEIZEREF, signal, 2, JBB);
}//Dbtc::execTCSEIZEREQ()

/****************************************************************************/
/*                    T C R E L E A S E Q                                   */
/*                  REQUEST TO RELEASE A CONNECT RECORD                     */
/****************************************************************************/
void Dbtc::execTCRELEASEREQ(Signal* signal) 
{
  UintR tapiPointer;
  BlockReference tapiBlockref;     /* SENDER BLOCK REFERENCE*/

  jamEntry();
  tapiPointer = signal->theData[0]; /* REQUEST SENDERS CONNECT RECORD POINTER*/
  tapiBlockref = signal->theData[1];/* SENDERS BLOCK REFERENCE*/
  tuserpointer = signal->theData[2];
  if (tapiPointer >= capiConnectFilesize) {
    jam();
    signal->theData[0] = tuserpointer;
    signal->theData[1] = ZINVALID_CONNECTION;
    signal->theData[2] = __LINE__;
    sendSignal(tapiBlockref, GSN_TCRELEASEREF, signal, 3, JBB);
    return;
  } else {
    jam();
    apiConnectptr.i = tapiPointer;
  }//if
  ptrAss(apiConnectptr, apiConnectRecord);
  if (apiConnectptr.p->apiConnectstate == CS_DISCONNECTED) {
    jam();
    signal->theData[0] = tuserpointer;