Commit 15767832 authored by unknown's avatar unknown
Browse files

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

into  perch.ndb.mysql.com:/home/jonas/src/mysql-5.0-ndb
parents 76dd9aa2 6b041b49

ndb/src/kernel/blocks/ERROR_codes.txt

+3 −1
Original line number Diff line number Diff line
@@ -5,7 +5,7 @@ Next DBACC 3002 
Next DBTUP 4014

Next DBLQH 5043

Next DBDICT 6007

Next DBDIH 7177

Next DBDIH 7178

Next DBTC 8038

Next CMVMI 9000

Next BACKUP 10022
@@ -66,6 +66,8 @@ Delay GCP_SAVEREQ by 10 secs 
7030: Delay in GCP_PREPARE until node has completed a node failure

7031: Delay in GCP_PREPARE and die 3s later



7177: Delay copying of sysfileData in execCOPY_GCIREQ



ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:

-----------------------------------------------------------------

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

+2 −0
Original line number Diff line number Diff line
@@ -1048,6 +1048,8 @@ private: 
  void removeStoredReplica(FragmentstorePtr regFragptr,

                           ReplicaRecordPtr replicaPtr);

  void searchStoredReplicas(FragmentstorePtr regFragptr);

  bool setup_create_replica(FragmentstorePtr, CreateReplicaRecord*,

			    ConstPtr<ReplicaRecord>);

  void updateNodeInfo(FragmentstorePtr regFragptr);



//------------------------------------

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

+123 −77
Original line number Diff line number Diff line
@@ -627,21 +627,47 @@ void Dbdih::execCOPY_GCIREQ(Signal* signal) 
  ndbrequire(c_copyGCISlave.m_copyReason  == CopyGCIReq::IDLE);

  ndbrequire(c_copyGCISlave.m_expectedNextWord == tstart);

  ndbrequire(reason != CopyGCIReq::IDLE);

  bool isdone = (tstart + CopyGCIReq::DATA_SIZE) >= Sysfile::SYSFILE_SIZE32;



  if (ERROR_INSERTED(7177))

  {

    jam();



    if (signal->getLength() == 3)

    {

      jam();

      goto done;

    }

  }



  arrGuard(tstart + CopyGCIReq::DATA_SIZE, sizeof(sysfileData)/4);

  for(Uint32 i = 0; i<CopyGCIReq::DATA_SIZE; i++)

    cdata[tstart+i] = copyGCI->data[i];

  

  if ((tstart + CopyGCIReq::DATA_SIZE) >= Sysfile::SYSFILE_SIZE32) {

  if (ERROR_INSERTED(7177) && isMaster() && isdone)

  {

    sendSignalWithDelay(reference(), GSN_COPY_GCIREQ, signal, 1000, 3);

    return;

  }

  

done:  

  if (isdone)

  {

    jam();

    c_copyGCISlave.m_expectedNextWord = 0;

  } else {

  } 

  else 

  {

    jam();

    c_copyGCISlave.m_expectedNextWord += CopyGCIReq::DATA_SIZE;

    return;

  }//if

  }

  

  if (cmasterdihref != reference())

  {

    jam();

    memcpy(sysfileData, cdata, sizeof(sysfileData));

  }



  c_copyGCISlave.m_copyReason = reason;

  c_copyGCISlave.m_senderRef  = signal->senderBlockRef();
@@ -8447,14 +8473,30 @@ Dbdih::resetReplicaSr(TabRecordPtr tabPtr){ 


	  resetReplicaLcp(replicaPtr.p, newestRestorableGCI);



	  /* -----------------------------------------------------------------

	   *   LINK THE REPLICA INTO THE STORED REPLICA LIST. WE WILL USE THIS

	   *   NODE AS A STORED REPLICA.                                      

	   *   WE MUST FIRST LINK IT OUT OF THE LIST OF OLD STORED REPLICAS.  

	   * --------------------------------------------------------------- */

	  /**

	   * Make sure we can also find REDO for restoring replica...

	   */

	  {

	    CreateReplicaRecord createReplica;

	    ConstPtr<ReplicaRecord> constReplicaPtr;

	    constReplicaPtr.i = replicaPtr.i;

	    constReplicaPtr.p = replicaPtr.p;

	    if (setup_create_replica(fragPtr,

				     &createReplica, constReplicaPtr))

	    {

	      removeOldStoredReplica(fragPtr, replicaPtr);

	      linkStoredReplica(fragPtr, replicaPtr);

	    }

	    else

	    {

	      infoEvent("Forcing take-over of node %d due to unsufficient REDO"

			" for table %d fragment: %d",

			nodePtr.i, tabPtr.i, i);

	      

	      setNodeActiveStatus(nodePtr.i, 

				  Sysfile::NS_NotActive_NotTakenOver);

	    }

	  }

	}

        default:

	  jam();
@@ -9498,6 +9540,7 @@ void Dbdih::calculateKeepGciLab(Signal* signal, Uint32 tableId, Uint32 fragId) 
  FragmentstorePtr fragPtr;

  getFragstore(tabPtr.p, fragId, fragPtr);

  checkKeepGci(tabPtr, fragId, fragPtr.p, fragPtr.p->storedReplicas);

  checkKeepGci(tabPtr, fragId, fragPtr.p, fragPtr.p->oldStoredReplicas);

  fragId++;

  if (fragId >= tabPtr.p->totalfragments) {

    jam();
@@ -12493,37 +12536,14 @@ void Dbdih::removeTooNewCrashedReplicas(ReplicaRecordPtr rtnReplicaPtr) 
/*               CHECKPOINT WITHOUT NEEDING ANY EXTRA LOGGING FACILITIES.*/

/*               A MAXIMUM OF FOUR NODES IS RETRIEVED.                   */

/*************************************************************************/

void Dbdih::searchStoredReplicas(FragmentstorePtr fragPtr) 

bool

Dbdih::setup_create_replica(FragmentstorePtr fragPtr,

			    CreateReplicaRecord* createReplicaPtrP,

			    ConstPtr<ReplicaRecord> replicaPtr)

{

  Uint32 nextReplicaPtrI;

  ConstPtr<ReplicaRecord> replicaPtr;

  createReplicaPtrP->dataNodeId = replicaPtr.p->procNode;

  createReplicaPtrP->replicaRec = replicaPtr.i;



  replicaPtr.i = fragPtr.p->storedReplicas;

  while (replicaPtr.i != RNIL) {

    jam();

    ptrCheckGuard(replicaPtr, creplicaFileSize, replicaRecord);

    nextReplicaPtrI = replicaPtr.p->nextReplica;

    NodeRecordPtr nodePtr;

    nodePtr.i = replicaPtr.p->procNode;

    ptrCheckGuard(nodePtr, MAX_NDB_NODES, nodeRecord);

    if (nodePtr.p->nodeStatus == NodeRecord::ALIVE) {

      jam();

      switch (nodePtr.p->activeStatus) {

      case Sysfile::NS_Active:

      case Sysfile::NS_ActiveMissed_1:

      case Sysfile::NS_ActiveMissed_2:{

	/* ----------------------------------------------------------------- */

	/*   INITIALISE THE CREATE REPLICA STRUCTURE THAT IS USED FOR SENDING*/

	/*   TO LQH START_FRAGREQ.                                           */

	/*   SET THE DATA NODE WHERE THE LOCAL CHECKPOINT IS FOUND. ALSO     */

	/*   SET A REFERENCE TO THE REPLICA POINTER OF THAT.                 */

	/* ----------------------------------------------------------------- */

	CreateReplicaRecordPtr createReplicaPtr;

	createReplicaPtr.i = cnoOfCreateReplicas;

	ptrCheckGuard(createReplicaPtr, 4, createReplicaRecord);

	cnoOfCreateReplicas++;

	createReplicaPtr.p->dataNodeId = replicaPtr.p->procNode;

	createReplicaPtr.p->replicaRec = replicaPtr.i;

  /* ----------------------------------------------------------------- */

  /*   WE NEED TO SEARCH FOR A PROPER LOCAL CHECKPOINT TO USE FOR THE  */

  /*   SYSTEM RESTART.                                                 */
@@ -12535,7 +12555,8 @@ void Dbdih::searchStoredReplicas(FragmentstorePtr fragPtr) 
			     stopGci,

			     startGci,

			     startLcpNo);

	if (!result) {

  if (!result) 

  {

    jam();

    /* --------------------------------------------------------------- */

    /* WE COULD NOT FIND ANY LOCAL CHECKPOINT. THE FRAGMENT THUS DO NOT*/
@@ -12547,22 +12568,20 @@ void Dbdih::searchStoredReplicas(FragmentstorePtr fragPtr) 
    /* TO INDICATE THAT NO LOCAL CHECKPOINT IS TO BE USED WE SET THE   */

    /* LOCAL CHECKPOINT TO ZNIL.                                       */

    /* --------------------------------------------------------------- */

	  createReplicaPtr.p->lcpNo = ZNIL;

	} else {

    createReplicaPtrP->lcpNo = ZNIL;

  } 

  else 

  {

    jam();

    /* --------------------------------------------------------------- */

    /* WE FOUND A PROPER LOCAL CHECKPOINT TO RESTART FROM.             */

    /* SET LOCAL CHECKPOINT ID AND LOCAL CHECKPOINT NUMBER.            */

    /* --------------------------------------------------------------- */

	  createReplicaPtr.p->lcpNo = startLcpNo;

    createReplicaPtrP->lcpNo = startLcpNo;

    arrGuard(startLcpNo, MAX_LCP_STORED);

	  createReplicaPtr.p->createLcpId = replicaPtr.p->lcpId[startLcpNo];

    createReplicaPtrP->createLcpId = replicaPtr.p->lcpId[startLcpNo];

  }//if

  

	if(ERROR_INSERTED(7073) || ERROR_INSERTED(7074)){

	  jam();

	  nodePtr.p->nodeStatus = NodeRecord::DEAD;

	}

  

  /* ----------------------------------------------------------------- */

  /*   WE HAVE EITHER FOUND A LOCAL CHECKPOINT OR WE ARE PLANNING TO   */
@@ -12570,21 +12589,48 @@ void Dbdih::searchStoredReplicas(FragmentstorePtr fragPtr) 
  /*   CASES WE NEED TO FIND A SET OF LOGS THAT CAN EXECUTE SUCH THAT  */

  /*   WE RECOVER TO THE SYSTEM RESTART GLOBAL CHECKPOINT.             */

  /* -_--------------------------------------------------------------- */

	if (!findLogNodes(createReplicaPtr.p, fragPtr, startGci, stopGci)) {

	  jam();

	  /* --------------------------------------------------------------- */

	  /* WE WERE NOT ABLE TO FIND ANY WAY OF RESTORING THIS REPLICA.     */

	  /* THIS IS A POTENTIAL SYSTEM ERROR.                               */

	  /* --------------------------------------------------------------- */

	  cnoOfCreateReplicas--;

	  return;

	}//if

  return findLogNodes(createReplicaPtrP, fragPtr, startGci, stopGci);

}			    



	if(ERROR_INSERTED(7073) || ERROR_INSERTED(7074)){

void Dbdih::searchStoredReplicas(FragmentstorePtr fragPtr) 

{

  Uint32 nextReplicaPtrI;

  Ptr<ReplicaRecord> replicaPtr;



  replicaPtr.i = fragPtr.p->storedReplicas;

  while (replicaPtr.i != RNIL) {

    jam();

	  nodePtr.p->nodeStatus = NodeRecord::ALIVE;

	}

    ptrCheckGuard(replicaPtr, creplicaFileSize, replicaRecord);

    nextReplicaPtrI = replicaPtr.p->nextReplica;

    ConstPtr<ReplicaRecord> constReplicaPtr;

    constReplicaPtr.i = replicaPtr.i;

    constReplicaPtr.p = replicaPtr.p;

    NodeRecordPtr nodePtr;

    nodePtr.i = replicaPtr.p->procNode;

    ptrCheckGuard(nodePtr, MAX_NDB_NODES, nodeRecord);

    if (nodePtr.p->nodeStatus == NodeRecord::ALIVE) {

      jam();

      switch (nodePtr.p->activeStatus) {

      case Sysfile::NS_Active:

      case Sysfile::NS_ActiveMissed_1:

      case Sysfile::NS_ActiveMissed_2:{

	/* ----------------------------------------------------------------- */

	/*   INITIALISE THE CREATE REPLICA STRUCTURE THAT IS USED FOR SENDING*/

	/*   TO LQH START_FRAGREQ.                                           */

	/*   SET THE DATA NODE WHERE THE LOCAL CHECKPOINT IS FOUND. ALSO     */

	/*   SET A REFERENCE TO THE REPLICA POINTER OF THAT.                 */

	/* ----------------------------------------------------------------- */

	CreateReplicaRecordPtr createReplicaPtr;

	createReplicaPtr.i = cnoOfCreateReplicas;

	ptrCheckGuard(createReplicaPtr, 4, createReplicaRecord);

	cnoOfCreateReplicas++;

	

	/**

	 * Should have been checked in resetReplicaSr

	 */

	ndbrequire(setup_create_replica(fragPtr,

					createReplicaPtr.p, 

					constReplicaPtr));

	break;

      }

      default:

ndb/src/ndbapi/NdbScanOperation.cpp

+2 −2
Original line number Diff line number Diff line
@@ -680,7 +680,7 @@ void NdbScanOperation::close(bool forceSend, bool releaseOp) 
  theNdbCon = NULL;

  m_transConnection = NULL;



  if (tTransCon) 

  if (tTransCon && releaseOp) 

  {

    NdbIndexScanOperation* tOp = (NdbIndexScanOperation*)this;
@@ -695,7 +695,7 @@ void NdbScanOperation::close(bool forceSend, bool releaseOp) 
					&tTransCon->m_theLastScanOperation,

					tOp);

    }

    else if (releaseOp)

    else

    {

      ret = tTransCon->releaseScanOperation(&tTransCon->m_firstExecutedScanOp,

					    0, tOp);