ndb - bug#24717

  5.0 version
  Add some extra state to try to prevent sending committedread to node too early


ndb/src/kernel/blocks/dbdih/Dbdih.hpp:
  remove extra virtual-indirection
ndb/src/kernel/blocks/dbdih/DbdihInit.cpp:
  remove extra virtual-indirection
ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  remove extra virtual-indirection
ndb/src/kernel/blocks/dblqh/Dblqh.hpp:
  Add retry mechanism to send TCKEYREF
    (as extra safety)
ndb/src/kernel/blocks/dblqh/DblqhMain.cpp:
  Add retry mechanism to send TCKEYREF
    (as extra safety)
ndb/src/kernel/blocks/dbtc/Dbtc.hpp:
  Add extra state to prevent sending committed reads to node to early
    (before connection establishment)
ndb/src/kernel/blocks/dbtc/DbtcInit.cpp:
  Add extra state to prevent sending committed reads to node to early
    (before connection establishment)
ndb/src/kernel/blocks/dbtc/DbtcMain.cpp:
  Add extra state to prevent sending committed reads to node to early
    (before connection establishment)
ndb/src/kernel/vm/SimulatedBlock.cpp:
  remove extra virtual-indirection
ndb/src/kernel/vm/SimulatedBlock.hpp:
  remove extra virtual-indirection

10 files changed, 87 insertions, 19 deletions

diff --git a/ndb/src/kernel/blocks/dbdih/Dbdih.hpp b/ndb/src/kernel/blocks/dbdih/Dbdih.hpp
index 53e54e96d2b..f1b3897c76f 100644
--- a/ndb/src/kernel/blocks/dbdih/Dbdih.hpp
+++ b/ndb/src/kernel/blocks/dbdih/Dbdih.hpp
@@ -1602,7 +1602,7 @@ private:
   void startInfoReply(Signal *, Uint32 nodeId);
 
   // DIH specifics for execNODE_START_REP (sendDictUnlockOrd)
-  void exec_node_start_rep(Signal* signal);
+  void execNODE_START_REP(Signal* signal);
 
   /*
    * Lock master DICT.  Only current use is by starting node
diff --git a/ndb/src/kernel/blocks/dbdih/DbdihInit.cpp b/ndb/src/kernel/blocks/dbdih/DbdihInit.cpp
index 2b878034258..aee64687072 100644
--- a/ndb/src/kernel/blocks/dbdih/DbdihInit.cpp
+++ b/ndb/src/kernel/blocks/dbdih/DbdihInit.cpp
@@ -269,7 +269,8 @@ Dbdih::Dbdih(const class Configuration & config):
 
   addRecSignal(GSN_DICT_LOCK_CONF, &Dbdih::execDICT_LOCK_CONF);
   addRecSignal(GSN_DICT_LOCK_REF, &Dbdih::execDICT_LOCK_REF);
-
+  addRecSignal(GSN_NODE_START_REP, &Dbdih::execNODE_START_REP);
+  
   apiConnectRecord = 0;  
   connectRecord = 0;  
   fileRecord = 0;  
diff --git a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
index cf3779737c1..cd82b6fc425 100644
--- a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
+++ b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
@@ -1390,7 +1390,7 @@ void Dbdih::execNDB_STTOR(Signal* signal)
 }//Dbdih::execNDB_STTOR()
 
 void
-Dbdih::exec_node_start_rep(Signal* signal)
+Dbdih::execNODE_START_REP(Signal* signal)
 {
   /*
    * Send DICT_UNLOCK_ORD when this node is SL_STARTED.
diff --git a/ndb/src/kernel/blocks/dblqh/Dblqh.hpp b/ndb/src/kernel/blocks/dblqh/Dblqh.hpp
index 299cad16ec1..0691e52d0ee 100644
--- a/ndb/src/kernel/blocks/dblqh/Dblqh.hpp
+++ b/ndb/src/kernel/blocks/dblqh/Dblqh.hpp
@@ -236,6 +236,7 @@
 #define ZOPERATION_EVENT_REP 19
 #define ZPREP_DROP_TABLE 20
 #define ZENABLE_EXPAND_CHECK 21
+#define ZRETRY_TCKEYREF 22
 
 /* ------------------------------------------------------------------------- */
 /*        NODE STATE DURING SYSTEM RESTART, VARIABLES CNODES_SR_STATE        */
@@ -2276,6 +2277,7 @@ private:
   void releaseScanrec(Signal* signal);
   void seizeScanrec(Signal* signal);
   Uint32 sendKeyinfo20(Signal* signal, ScanRecord *, TcConnectionrec *);
+  void sendTCKEYREF(Signal*, Uint32 dst, Uint32 route, Uint32 cnt);
   void sendScanFragConf(Signal* signal, Uint32 scanCompleted);
   void initCopyrec(Signal* signal);
   void initCopyTc(Signal* signal);
diff --git a/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp b/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
index 4739450884c..3d93ce3fc31 100644
--- a/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
+++ b/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
@@ -464,6 +464,22 @@ void Dblqh::execCONTINUEB(Signal* signal)
       return;
     }
   }
+  case ZRETRY_TCKEYREF:
+  {
+    jam();
+    Uint32 cnt = signal->theData[1];
+    Uint32 ref = signal->theData[2];
+    if (cnt < (10 * 60 * 5))
+    {
+      jam();
+      /**
+       * Only retry for 5 minutes...then hope that API has handled it..somehow
+       */
+      memmove(signal->theData, signal->theData+3, 4*TcKeyRef::SignalLength);
+      sendTCKEYREF(signal, ref, 0, cnt);
+    }
+    return;
+  }
   default:
     ndbrequire(false);
     break;
@@ -2370,7 +2386,7 @@ void Dblqh::noFreeRecordLab(Signal* signal,
     tcKeyRef->transId[0] = transid1;
     tcKeyRef->transId[1] = transid2;
     tcKeyRef->errorCode = errCode;
-    sendSignal(apiRef, GSN_TCKEYREF, signal, TcKeyRef::SignalLength, JBB);
+    sendTCKEYREF(signal, apiRef, signal->getSendersBlockRef(), 0);
   } else {
     jam();
 
@@ -6576,8 +6592,7 @@ void Dblqh::continueAfterLogAbortWriteLab(Signal* signal)
     tcKeyRef->transId[0] = regTcPtr->transid[0];
     tcKeyRef->transId[1] = regTcPtr->transid[1];
     tcKeyRef->errorCode = regTcPtr->errorCode;
-    sendSignal(regTcPtr->applRef, 
-               GSN_TCKEYREF, signal, TcKeyRef::SignalLength, JBB);
+    sendTCKEYREF(signal, regTcPtr->applRef, regTcPtr->clientBlockref, 0);
     cleanUp(signal);
     return;
   }//if
@@ -6612,6 +6627,29 @@ void Dblqh::continueAfterLogAbortWriteLab(Signal* signal)
   cleanUp(signal);
 }//Dblqh::continueAfterLogAbortWriteLab()
 
+void
+Dblqh::sendTCKEYREF(Signal* signal, Uint32 ref, Uint32 routeRef, Uint32 cnt)
+{
+  const Uint32 nodeId = refToNode(ref);
+  const bool connectedToNode = getNodeInfo(nodeId).m_connected;
+  
+  if (likely(connectedToNode))
+  {
+    jam();
+    sendSignal(ref, GSN_TCKEYREF, signal, TcKeyRef::SignalLength, JBB);
+  }
+  else
+  {
+    jam();
+    memmove(signal->theData + 3, signal->theData, 4*TcKeyRef::SignalLength);
+    signal->theData[0] = ZRETRY_TCKEYREF;
+    signal->theData[1] = cnt + 1;
+    signal->theData[2] = ref;
+    sendSignalWithDelay(reference(), GSN_CONTINUEB, signal, 100,
+			TcKeyRef::SignalLength + 3);
+  }
+}
+
 /* ########################################################################## 
  * #######                       MODULE TO HANDLE TC FAILURE          ####### 
  *
diff --git a/ndb/src/kernel/blocks/dbtc/Dbtc.hpp b/ndb/src/kernel/blocks/dbtc/Dbtc.hpp
index bf6ce7129ba..792586d7baf 100644
--- a/ndb/src/kernel/blocks/dbtc/Dbtc.hpp
+++ b/ndb/src/kernel/blocks/dbtc/Dbtc.hpp
@@ -942,7 +942,8 @@ public:
       NF_CHECK_SCAN        = 0x2,
       NF_CHECK_TRANSACTION = 0x4,
       NF_CHECK_DROP_TAB    = 0x8,
-      NF_NODE_FAIL_BITS    = 0xF // All bits...
+      NF_NODE_FAIL_BITS    = 0xF, // All bits...
+      NF_STARTED           = 0x10
     };
     Uint32 m_nf_bits;
     NdbNodeBitmask m_lqh_trans_conf;
@@ -1319,6 +1320,7 @@ private:
   void execCOMMITCONF(Signal* signal);
   void execABORTCONF(Signal* signal);
   void execNODE_FAILREP(Signal* signal);
+  void execNODE_START_REP(Signal* signal);
   void execINCL_NODEREQ(Signal* signal);
   void execTIME_SIGNAL(Signal* signal);
   void execAPI_FAILREQ(Signal* signal);
diff --git a/ndb/src/kernel/blocks/dbtc/DbtcInit.cpp b/ndb/src/kernel/blocks/dbtc/DbtcInit.cpp
index 7bd308119fc..098373fb3de 100644
--- a/ndb/src/kernel/blocks/dbtc/DbtcInit.cpp
+++ b/ndb/src/kernel/blocks/dbtc/DbtcInit.cpp
@@ -290,7 +290,8 @@ Dbtc::Dbtc(const class Configuration & conf):
   addRecSignal(GSN_WAIT_DROP_TAB_CONF, &Dbtc::execWAIT_DROP_TAB_CONF);
   
   addRecSignal(GSN_ALTER_TAB_REQ, &Dbtc::execALTER_TAB_REQ);
-
+  addRecSignal(GSN_NODE_START_REP, &Dbtc::execNODE_START_REP, true);
+  
   cacheRecord = 0;
   apiConnectRecord = 0;
   tcConnectRecord = 0;
diff --git a/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp b/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
index 9edfe7f79bd..a436316f7c9 100644
--- a/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
+++ b/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
@@ -3068,7 +3068,28 @@ void Dbtc::tckeyreq050Lab(Signal* signal)
 	  }//if
 	}//for
       }
-    }//if
+
+      if (regTcPtr->tcNodedata[0] != getOwnNodeId())
+      {
+	jam();
+	for (Uint32 i = 0; i < tnoOfBackup + 1; i++)
+	{
+	  HostRecordPtr hostPtr;
+	  hostPtr.i = regTcPtr->tcNodedata[i];
+	  ptrCheckGuard(hostPtr, chostFilesize, hostRecord);
+	  if (hostPtr.p->m_nf_bits & HostRecord::NF_STARTED)
+	  {
+	    jam();
+	    if (i != 0)
+	    {
+	      jam();
+	      regTcPtr->tcNodedata[0] = hostPtr.i;
+	    }
+	    break;
+	  }
+	}
+      }//if
+    }
     jam();
     regTcPtr->lastReplicaNo = 0;
     regTcPtr->noOfNodes = 1;
@@ -7003,6 +7024,19 @@ void Dbtc::execNODE_FAILREP(Signal* signal)
 }//Dbtc::execNODE_FAILREP()
 
 void
+Dbtc::execNODE_START_REP(Signal* signal)
+{
+  Uint32 nodeId = signal->theData[0];
+  hostptr.i = nodeId;
+  ptrCheckGuard(hostptr, chostFilesize, hostRecord);
+  if (hostptr.p->m_nf_bits == 0)
+  {
+    jam();
+    hostptr.p->m_nf_bits |= HostRecord::NF_STARTED;
+  }
+}
+
+void
 Dbtc::checkNodeFailComplete(Signal* signal, 
 			    Uint32 failedNodeId,
 			    Uint32 bit)
diff --git a/ndb/src/kernel/vm/SimulatedBlock.cpp b/ndb/src/kernel/vm/SimulatedBlock.cpp
index b4787209d55..bbf13528c5c 100644
--- a/ndb/src/kernel/vm/SimulatedBlock.cpp
+++ b/ndb/src/kernel/vm/SimulatedBlock.cpp
@@ -917,15 +917,6 @@ SimulatedBlock::execCONTINUE_FRAGMENTED(Signal * signal){
 void
 SimulatedBlock::execNODE_START_REP(Signal* signal)
 {
-  // common stuff for all blocks
-
-  // block specific stuff by virtual method override (default empty)
-  exec_node_start_rep(signal);
-}
-
-void
-SimulatedBlock::exec_node_start_rep(Signal* signal)
-{
 }
 
 #ifdef VM_TRACE_TIME
diff --git a/ndb/src/kernel/vm/SimulatedBlock.hpp b/ndb/src/kernel/vm/SimulatedBlock.hpp
index 4a3620a00ab..f7ca4ecbf38 100644
--- a/ndb/src/kernel/vm/SimulatedBlock.hpp
+++ b/ndb/src/kernel/vm/SimulatedBlock.hpp
@@ -424,7 +424,6 @@ private:
   void execSIGNAL_DROPPED_REP(Signal* signal);
   void execCONTINUE_FRAGMENTED(Signal* signal);
   void execNODE_START_REP(Signal* signal);
-  virtual void exec_node_start_rep(Signal* signal);
 
   Uint32 c_fragmentIdCounter;
   ArrayPool<FragmentInfo> c_fragmentInfoPool;