ndb - bug#21271

  make each fragment use own LCP file, so that (s/n)r can use different LCP-no for different fragments

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

@@ -136,9 +136,11 @@ private:
    */
   static Uint32 v5_getLcpNo(const Uint32 fileNumber[]);
   static Uint32 v5_getTableId(const Uint32 fileNumber[]);
-  
+  static Uint32 v5_getFragmentId(const Uint32 fileNumber[]);
+
   static void v5_setLcpNo(Uint32 fileNumber[], Uint32 no);
   static void v5_setTableId(Uint32 fileNumber[], Uint32 no);
+  static void v5_setFragmentId(Uint32 fileNumber[], Uint32 no);
 };
 
 /**
@@ -318,5 +320,15 @@ void FsOpenReq::v5_setLcpNo(Uint32 fileNumber[], Uint32 val){
   fileNumber[1] = val;
 }
 
+inline 
+Uint32 FsOpenReq::v5_getFragmentId(const Uint32 fileNumber[]){
+  return fileNumber[2];
+}
+
+inline
+void FsOpenReq::v5_setFragmentId(Uint32 fileNumber[], Uint32 val){
+  fileNumber[2] = val;
+}
+
 #endif
 

storage/ndb/src/kernel/blocks/ERROR_codes.txt

@@ -3,7 +3,7 @@ Next NDBCNTR 1000
 Next NDBFS 2000
 Next DBACC 3002
 Next DBTUP 4024
-Next DBLQH 5043
+Next DBLQH 5045
 Next DBDICT 6007
 Next DBDIH 7177
 Next DBTC 8037

storage/ndb/src/kernel/blocks/backup/Backup.cpp

@@ -3796,20 +3796,28 @@ Backup::fragmentCompleted(Signal* signal, BackupFilePtr filePtr)
   BackupRecordPtr ptr;
   c_backupPool.getPtr(ptr, filePtr.p->backupPtr);
 
-  BackupFragmentConf * conf = (BackupFragmentConf*)signal->getDataPtrSend();
-  conf->backupId = ptr.p->backupId;
-  conf->backupPtr = ptr.i;
-  conf->tableId = filePtr.p->tableId;
-  conf->fragmentNo = filePtr.p->fragmentNo;
-  conf->noOfRecordsLow = (Uint32)(op.noOfRecords & 0xFFFFFFFF);
-  conf->noOfRecordsHigh = (Uint32)(op.noOfRecords >> 32);
-  conf->noOfBytesLow = (Uint32)(op.noOfBytes & 0xFFFFFFFF);
-  conf->noOfBytesHigh = (Uint32)(op.noOfBytes >> 32);
-  sendSignal(ptr.p->masterRef, GSN_BACKUP_FRAGMENT_CONF, signal,
-	     BackupFragmentConf::SignalLength, JBB);
-  
-  ptr.p->m_gsn = GSN_BACKUP_FRAGMENT_CONF;
-  ptr.p->slaveState.setState(STARTED);
+  if (ptr.p->is_lcp())
+  {
+    ptr.p->slaveState.setState(STOPPING);
+    filePtr.p->operation.dataBuffer.eof();
+  }
+  else
+  {
+    BackupFragmentConf * conf = (BackupFragmentConf*)signal->getDataPtrSend();
+    conf->backupId = ptr.p->backupId;
+    conf->backupPtr = ptr.i;
+    conf->tableId = filePtr.p->tableId;
+    conf->fragmentNo = filePtr.p->fragmentNo;
+    conf->noOfRecordsLow = (Uint32)(op.noOfRecords & 0xFFFFFFFF);
+    conf->noOfRecordsHigh = (Uint32)(op.noOfRecords >> 32);
+    conf->noOfBytesLow = (Uint32)(op.noOfBytes & 0xFFFFFFFF);
+    conf->noOfBytesHigh = (Uint32)(op.noOfBytes >> 32);
+    sendSignal(ptr.p->masterRef, GSN_BACKUP_FRAGMENT_CONF, signal,
+	       BackupFragmentConf::SignalLength, JBB);
+
+    ptr.p->m_gsn = GSN_BACKUP_FRAGMENT_CONF;
+    ptr.p->slaveState.setState(STARTED);
+  }
   return;
 }
 
@@ -4719,29 +4727,8 @@ Backup::execLCP_PREPARE_REQ(Signal* signal)
   BackupRecordPtr ptr;
   c_backupPool.getPtr(ptr, req.backupPtr);
 
-  bool first= true;
   TablePtr tabPtr;
-  if(ptr.p->tables.first(tabPtr) && tabPtr.p->tableId != req.tableId)
-  {
-    jam();
-    first= false;
-    tabPtr.p->attributes.release();
-    tabPtr.p->fragments.release();
-    ptr.p->tables.release();
-    ptr.p->errorCode = 0;
-  } 
-
-  if(ptr.p->tables.first(tabPtr) && ptr.p->errorCode == 0)
-  {
-    jam();
-    FragmentPtr fragPtr;
-    tabPtr.p->fragments.getPtr(fragPtr, 0);
-    fragPtr.p->fragmentId = req.fragmentId;
-    
-    lcp_open_file_done(signal, ptr);
-    return;
-  } 
-  else if(ptr.p->errorCode == 0)
+  if(ptr.p->errorCode == 0)
   {
     jam();
     FragmentPtr fragPtr;
@@ -4759,6 +4746,9 @@ Backup::execLCP_PREPARE_REQ(Signal* signal)
     fragPtr.p->scanned = 0;
     fragPtr.p->scanning = 0;
     fragPtr.p->tableId = req.tableId;
+
+    ptr.p->backupId= req.backupId;
+    lcp_open_file(signal, ptr);
   } 
   else
   {
@@ -4767,43 +4757,38 @@ Backup::execLCP_PREPARE_REQ(Signal* signal)
     tabPtr.p->fragments.getPtr(fragPtr, 0);
     fragPtr.p->fragmentId = req.fragmentId;
     defineBackupRef(signal, ptr, ptr.p->errorCode);
-    return;
-  }
-  
-  if(first)
-  {
-    jam();
-    // start file thread
-    ptr.p->backupId= req.backupId;
-    lcp_open_file(signal, ptr);
-    return;
-  }
-  else
-  {
-    jam();
-    ndbrequire(ptr.p->backupId == req.backupId);
   }
-  
-  /**
-   * Close previous file
-   */
-  jam();
-  BackupFilePtr filePtr;
-  c_backupFilePool.getPtr(filePtr, ptr.p->dataFilePtr);
-  filePtr.p->operation.dataBuffer.eof();
 }
 
 void
 Backup::lcp_close_file_conf(Signal* signal, BackupRecordPtr ptr)
 {
-  if(!ptr.p->tables.isEmpty())
-  {
-    jam();
-    lcp_open_file(signal, ptr);
-    return;
-  }
+  jam();
+  
+  TablePtr tabPtr;
+  ndbrequire(ptr.p->tables.first(tabPtr));
+  Uint32 tableId = tabPtr.p->tableId;
+
+  FragmentPtr fragPtr;
+  tabPtr.p->fragments.getPtr(fragPtr, 0);
+  Uint32 fragmentId = fragPtr.p->fragmentId;
+  
+  tabPtr.p->attributes.release();
+  tabPtr.p->fragments.release();
+  ptr.p->tables.release();
+  ptr.p->errorCode = 0;
   
-  lcp_send_end_lcp_conf(signal, ptr);
+  BackupFragmentConf * conf = (BackupFragmentConf*)signal->getDataPtrSend();
+  conf->backupId = ptr.p->backupId;
+  conf->backupPtr = ptr.i;
+  conf->tableId = tableId;
+  conf->fragmentNo = fragmentId;
+  conf->noOfRecordsLow = 0;
+  conf->noOfRecordsHigh = 0;
+  conf->noOfBytesLow = 0;
+  conf->noOfBytesHigh = 0;
+  sendSignal(ptr.p->masterRef, GSN_BACKUP_FRAGMENT_CONF, signal,
+	     BackupFragmentConf::SignalLength, JBB);
 }
 
 void
@@ -4840,6 +4825,7 @@ Backup::lcp_open_file(Signal* signal, BackupRecordPtr ptr)
   FsOpenReq::setSuffix(req->fileNumber, FsOpenReq::S_DATA);
   FsOpenReq::v5_setLcpNo(req->fileNumber, fragPtr.p->lcp_no);
   FsOpenReq::v5_setTableId(req->fileNumber, tabPtr.p->tableId);
+  FsOpenReq::v5_setFragmentId(req->fileNumber, fragPtr.p->fragmentId);
   sendSignal(NDBFS_REF, GSN_FSOPENREQ, signal, FsOpenReq::SignalLength, JBA);
 }
 
@@ -4872,38 +4858,15 @@ Backup::execEND_LCPREQ(Signal* signal)
   c_backupPool.getPtr(ptr, req->backupPtr);
   ndbrequire(ptr.p->backupId == req->backupId);
 
-  ptr.p->slaveState.setState(STOPPING);
-
-  TablePtr tabPtr;
-  if(ptr.p->tables.first(tabPtr))
-  {
-    tabPtr.p->attributes.release();
-    tabPtr.p->fragments.release();
-    ptr.p->tables.release();
-  
-    BackupFilePtr filePtr;
-    c_backupFilePool.getPtr(filePtr, ptr.p->dataFilePtr);
-    filePtr.p->operation.dataBuffer.eof();
-    return;
-  } 
-  
-  lcp_send_end_lcp_conf(signal, ptr);
-}
-
-void
-Backup::lcp_send_end_lcp_conf(Signal* signal, BackupRecordPtr ptr)
-{
-  EndLcpConf* conf= (EndLcpConf*)signal->getDataPtr();
-
-  conf->senderData = ptr.p->clientData;
-  conf->senderRef = reference();
-  
   ptr.p->errorCode = 0;
   ptr.p->slaveState.setState(CLEANING);
   ptr.p->slaveState.setState(INITIAL);
   ptr.p->slaveState.setState(DEFINING);
   ptr.p->slaveState.setState(DEFINED);
 
+  EndLcpConf* conf= (EndLcpConf*)signal->getDataPtr();
+  conf->senderData = ptr.p->clientData;
+  conf->senderRef = reference();
   sendSignal(ptr.p->masterRef, GSN_END_LCPCONF,
 	     signal, EndLcpConf::SignalLength, JBB);
 }

storage/ndb/src/kernel/blocks/backup/Backup.hpp

@@ -645,7 +645,6 @@ public:
   void lcp_open_file(Signal* signal, BackupRecordPtr ptr);
   void lcp_open_file_done(Signal*, BackupRecordPtr);
   void lcp_close_file_conf(Signal* signal, BackupRecordPtr);
-  void lcp_send_end_lcp_conf(Signal* signal, BackupRecordPtr);
 
   bool ready_to_write(bool ready, Uint32 sz, bool eof, BackupFile *fileP);
 };

storage/ndb/src/kernel/blocks/dblqh/Dblqh.hpp

@@ -2794,10 +2794,10 @@ private:
 /*THIS VARIABLE IS THE HEAD OF A LINKED LIST OF FRAGMENTS WAITING TO BE      */
 /*RESTORED FROM DISK.                                                        */
 /* ------------------------------------------------------------------------- */
-  DLList<Fragrecord> c_lcp_waiting_fragments;  // StartFragReq'ed
-  DLList<Fragrecord> c_lcp_restoring_fragments; // Restoring as we speek
-  DLList<Fragrecord> c_lcp_complete_fragments;  // Restored
-  DLList<Fragrecord> c_redo_complete_fragments; // Redo'ed
+  DLFifoList<Fragrecord> c_lcp_waiting_fragments;  // StartFragReq'ed
+  DLFifoList<Fragrecord> c_lcp_restoring_fragments; // Restoring as we speek
+  DLFifoList<Fragrecord> c_lcp_complete_fragments;  // Restored
+  DLFifoList<Fragrecord> c_redo_complete_fragments; // Redo'ed
   
 /* ------------------------------------------------------------------------- */
 /*USED DURING SYSTEM RESTART, INDICATES THE OLDEST GCI THAT CAN BE RESTARTED */

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

@@ -427,17 +427,20 @@ void Dblqh::execCONTINUEB(Signal* signal)
       signal->theData[0] = fragptr.p->tabRef;
       signal->theData[1] = fragptr.p->fragId;
       sendSignal(DBACC_REF, GSN_EXPANDCHECK2, signal, 2, JBB);
+      Ptr<Fragrecord> save = fragptr;
 
       c_redo_complete_fragments.next(fragptr);
       signal->theData[0] = ZENABLE_EXPAND_CHECK;
       signal->theData[1] = fragptr.i;
       sendSignal(DBLQH_REF, GSN_CONTINUEB, signal, 2, JBB);	
+
+      c_redo_complete_fragments.remove(save);
       return;
     }
     else
     {
       jam();
-      c_redo_complete_fragments.remove();
+      ndbrequire(c_redo_complete_fragments.isEmpty());
       StartRecConf * conf = (StartRecConf*)signal->getDataPtrSend();
       conf->startingNodeId = getOwnNodeId();
       sendSignal(cmasterDihBlockref, GSN_START_RECCONF, signal, 
@@ -11269,8 +11272,22 @@ void Dblqh::execLCP_PREPARE_CONF(Signal* signal)
     else
 #endif
     {
-      sendSignal(BACKUP_REF, GSN_BACKUP_FRAGMENT_REQ, signal, 
-		 BackupFragmentReq::SignalLength, JBB);
+      if (ERROR_INSERTED(5044) && 
+	  (fragptr.p->tabRef == c_error_insert_table_id) && 
+	  fragptr.p->fragId) // Not first frag
+      {
+	/**
+	 * Force CRASH_INSERTION in 10s
+	 */
+	ndbout_c("table: %d frag: %d", fragptr.p->tabRef, fragptr.p->fragId);
+	SET_ERROR_INSERT_VALUE(5027);
+	sendSignalWithDelay(reference(), GSN_START_RECREQ, signal, 10000, 1);
+      }
+      else
+      {
+	sendSignal(BACKUP_REF, GSN_BACKUP_FRAGMENT_REQ, signal, 
+		   BackupFragmentReq::SignalLength, JBB);
+      }
     }
   }
 }
@@ -13745,7 +13762,7 @@ void Dblqh::execSTART_FRAGREQ(Signal* signal)
     fragptr.p->newestGci = cnewestGci;
   }//if
   
-  if (lcpNo == ZNIL || fragptr.i != tabptr.p->fragrec[0])
+  if (lcpNo == ZNIL)
   {
     jam();
     /**
@@ -18470,10 +18487,17 @@ Dblqh::execDUMP_STATE_ORD(Signal* signal)
     }
   }
 
-  if (dumpState->args[0] == DumpStateOrd::LqhErrorInsert5042 && signal->getLength() == 2)
+  if (dumpState->args[0] == DumpStateOrd::LqhErrorInsert5042 && (signal->getLength() >= 2))
   {
     c_error_insert_table_id = dumpState->args[1];
-    SET_ERROR_INSERT_VALUE(5042);
+    if (signal->getLength() == 2)
+    {
+      SET_ERROR_INSERT_VALUE(5042);
+    }
+    else
+    {
+      SET_ERROR_INSERT_VALUE(dumpState->args[2]);
+    }
   }
 
   TcConnectionrec *regTcConnectionrec = tcConnectionrec;

storage/ndb/src/kernel/blocks/ndbfs/Filename.cpp

@@ -149,7 +149,8 @@ Filename::set(Filename::NameSpec& spec,
   {
     Uint32 tableId = FsOpenReq::v5_getTableId(filenumber);
     Uint32 lcpNo = FsOpenReq::v5_getLcpNo(filenumber);
-    BaseString::snprintf(buf, sizeof(buf), "LCP/%d/T%d", lcpNo, tableId);
+    Uint32 fragId = FsOpenReq::v5_getFragmentId(filenumber);
+    BaseString::snprintf(buf, sizeof(buf), "LCP/%d/T%dF%d", lcpNo, tableId, fragId);
     strcat(theName, buf);
     break;
   }

storage/ndb/src/kernel/blocks/restore.cpp

@@ -264,7 +264,7 @@ Restore::init_file(const RestoreLcpReq* req, FilePtr file_ptr)
   file_ptr.p->m_status = File::FIRST_READ;
   
   file_ptr.p->m_table_id = req->tableId;
-  file_ptr.p->m_fragment_id = RNIL;
+  file_ptr.p->m_fragment_id = req->fragmentId;
   file_ptr.p->m_table_version = RNIL;
 
   file_ptr.p->m_bytes_left = 0; // Bytes read from FS
@@ -361,6 +361,7 @@ Restore::open_file(Signal* signal, FilePtr file_ptr, Uint32 lcpNo)
   FsOpenReq::setSuffix(req->fileNumber, FsOpenReq::S_DATA);
   FsOpenReq::v5_setLcpNo(req->fileNumber, lcpNo);
   FsOpenReq::v5_setTableId(req->fileNumber, file_ptr.p->m_table_id);
+  FsOpenReq::v5_setFragmentId(req->fileNumber, file_ptr.p->m_fragment_id);
   sendSignal(NDBFS_REF, GSN_FSOPENREQ, signal, FsOpenReq::SignalLength, JBA);
 }
 
@@ -475,6 +476,11 @@ Restore::restore_next(Signal* signal, FilePtr file_ptr)
        */
       ndbout_c("records: %d len: %x left: %d", 
 	       status & File::READING_RECORDS, 4*len, left);
+      
+      if (unlikely((status & File:: FILE_THREAD_RUNNING) == 0))
+      {
+	ndbrequire(false);
+      }
       len= 0;
       break;
     }

storage/ndb/test/ndbapi/testNodeRestart.cpp

@@ -931,6 +931,41 @@ retry:
   return NDBT_OK;
 }
 
+int 
+runBug21271(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  HugoOperations hugoOps(*ctx->getTab());
+  Ndb* pNdb = GETNDB(step);
+  
+  const int masterNode = restarter.getMasterNodeId();
+  const int nodeId = restarter.getRandomNodeSameNodeGroup(masterNode, rand());
+
+  int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  if (restarter.dumpStateOneNode(nodeId, val2, 2))
+    return NDBT_FAILED;
+  
+  Uint32 tableId = ctx->getTab()->getTableId();
+  int dump[] = { DumpStateOrd::LqhErrorInsert5042, 0, 5044 };
+  dump[1] = tableId;
+
+  if (restarter.dumpStateOneNode(nodeId, dump, 3))
+    return NDBT_FAILED;
+  
+  restarter.waitNodesNoStart(&nodeId, 1);
+  ctx->stopTest();
+
+  restarter.startNodes(&nodeId, 1);
+
+  if (restarter.waitClusterStarted() != 0)
+    return NDBT_FAILED;
+
+  return NDBT_OK;
+  return NDBT_OK;
+}
+
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -1244,6 +1279,13 @@ TESTCASE("Bug20185",
   STEP(runBug20185);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug21271",
+	 ""){
+  INITIALIZER(runLoadTable);
+  STEP(runBug21271);
+  STEP(runPkUpdateUntilStopped);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -489,6 +489,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug20185 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug21271 T6
+
 #
 # DICT TESTS
 max-time: 1500