ndb - bug#29331 (51)

    Add better handling of GCP Stop
    Only kill "offending" node


storage/ndb/src/kernel/blocks/ERROR_codes.txt:
  add new error codes
storage/ndb/src/kernel/blocks/dbdih/Dbdih.hpp:
  add better GCP stop handling
storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  add better GCP stop handling

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

@@ -5,7 +5,7 @@ Next DBACC 3002
 Next DBTUP 4029
 Next DBLQH 5045
 Next DBDICT 6007
-Next DBDIH 7183
+Next DBDIH 7186
 Next DBTC 8040
 Next CMVMI 9000
 Next BACKUP 10038
@@ -75,6 +75,10 @@ Delay GCP_SAVEREQ by 10 secs
 
 7180: Crash master during master-take-over in execMASTER_LCPCONF
 
+7184: Crash before starting next GCP after a node failure
+
+7185: Dont reply to COPY_GCI_REQ where reason == GCP
+
 ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
 -----------------------------------------------------------------
 

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

@@ -899,7 +899,7 @@ private:
   void ndbsttorry10Lab(Signal *, Uint32 _line);
   void createMutexes(Signal* signal, Uint32 no);
   void createMutex_done(Signal* signal, Uint32 no, Uint32 retVal);
-  void crashSystemAtGcpStop(Signal *);
+  void crashSystemAtGcpStop(Signal *, bool);
   void sendFirstDictfragsreq(Signal *, TabRecordPtr regTabPtr);
   void addtabrefuseLab(Signal *, ConnectRecordPtr regConnectPtr, Uint32 errorCode);
   void GCP_SAVEhandling(Signal *, Uint32 nodeId);

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

@@ -747,6 +747,13 @@ void Dbdih::execCOPY_GCIREQ(Signal* signal)
   }
   ndbrequire(ok);
   
+ 
+  if (ERROR_INSERTED(7185) && reason==CopyGCIReq::GLOBAL_CHECKPOINT)
+  {
+    jam();
+    return;
+  }
+
   /* ----------------------------------------------------------------------- */
   /*     WE START BY TRYING TO OPEN THE FIRST RESTORABLE GCI FILE.           */
   /* ----------------------------------------------------------------------- */
@@ -4071,6 +4078,11 @@ void Dbdih::execNODE_FAILREP(Signal* signal)
     CLEAR_ERROR_INSERT_VALUE;
   }
 
+  if (ERROR_INSERTED(7184))
+  {
+    SET_ERROR_INSERT_VALUE(7000);
+  }
+
   /*-------------------------------------------------------------------------*/
   // The first step is to convert from a bit mask to an array of failed nodes.
   /*-------------------------------------------------------------------------*/
@@ -7745,7 +7757,7 @@ void Dbdih::checkGcpStopLab(Signal* signal)
           g_eventLogger.error("System crash due to GCP Stop in state = %u",
                               (Uint32) cgcpStatus);
 #endif
-          crashSystemAtGcpStop(signal);
+          crashSystemAtGcpStop(signal, false);
           return;
         }//if
       } else {
@@ -7759,7 +7771,7 @@ void Dbdih::checkGcpStopLab(Signal* signal)
             g_eventLogger.error("System crash due to GCP Stop in state = %u",
                                 (Uint32) cgcpStatus);
 #endif
-	    crashSystemAtGcpStop(signal);
+	    crashSystemAtGcpStop(signal, false);
             return;
           }//if
         } else {
@@ -11117,37 +11129,128 @@ void Dbdih::tableCloseLab(Signal* signal, FileRecordPtr filePtr)
  * GCP stop detected, 
  * send SYSTEM_ERROR to all other alive nodes
  */
-void Dbdih::crashSystemAtGcpStop(Signal* signal)
+void Dbdih::crashSystemAtGcpStop(Signal* signal, bool local)
 {
+  if (local)
+    goto dolocal;
+
   switch(cgcpStatus){
+  case GCP_PREPARE_SENT:
+  {
+    jam();
+    /**
+     * We're waiting for a GCP PREPARE CONF
+     */
+    infoEvent("Detected GCP stop(%d)...sending kill to %s", 
+              cgcpStatus, c_GCP_PREPARE_Counter.getText());
+    ndbout_c("Detected GCP stop(%d)...sending kill to %s", 
+             cgcpStatus, c_GCP_PREPARE_Counter.getText());
+    
+    {
+      NodeReceiverGroup rg(DBDIH, c_GCP_PREPARE_Counter);
+      signal->theData[0] = 7022;
+      sendSignal(rg, GSN_DUMP_STATE_ORD, signal, 1, JBA);
+    }
+    
+    {
+      NodeReceiverGroup rg(NDBCNTR, c_GCP_PREPARE_Counter);
+      SystemError * const sysErr = (SystemError*)&signal->theData[0];
+      sysErr->errorCode = SystemError::GCPStopDetected;
+      sysErr->errorRef = reference();
+      sysErr->data1 = cgcpStatus;
+      sysErr->data2 = cgcpOrderBlocked;
+      sendSignal(rg, GSN_SYSTEM_ERROR, signal, 
+                 SystemError::SignalLength, JBA);
+    }
+    ndbrequire(!c_GCP_PREPARE_Counter.done());
+    return;
+  }
+  case GCP_COMMIT_SENT:
+  {
+    jam();
+    /**
+     * We're waiting for a GCP_NODEFINISH
+     */
+    infoEvent("Detected GCP stop(%d)...sending kill to %s", 
+	      cgcpStatus, c_GCP_COMMIT_Counter.getText());
+    ndbout_c("Detected GCP stop(%d)...sending kill to %s", 
+	     cgcpStatus, c_GCP_COMMIT_Counter.getText());
+    
+    {
+      NodeReceiverGroup rg(DBDIH, c_GCP_COMMIT_Counter);
+      signal->theData[0] = 7022;
+      sendSignal(rg, GSN_DUMP_STATE_ORD, signal, 1, JBA);
+    }
+
+    {
+      NodeReceiverGroup rg(NDBCNTR, c_GCP_COMMIT_Counter);
+      SystemError * const sysErr = (SystemError*)&signal->theData[0];
+      sysErr->errorCode = SystemError::GCPStopDetected;
+      sysErr->errorRef = reference();
+      sysErr->data1 = cgcpStatus;
+      sysErr->data2 = cgcpOrderBlocked;
+      sendSignal(rg, GSN_SYSTEM_ERROR, signal, 
+                 SystemError::SignalLength, JBA);
+    }
+    ndbrequire(!c_GCP_COMMIT_Counter.done());
+    return;
+  }
   case GCP_NODE_FINISHED:
   {
+    jam();
     /**
      * We're waiting for a GCP save conf
      */
-    ndbrequire(!c_GCP_SAVEREQ_Counter.done());
     NodeReceiverGroup rg(DBLQH, c_GCP_SAVEREQ_Counter);
     signal->theData[0] = 2305;
     sendSignal(rg, GSN_DUMP_STATE_ORD, signal, 1, JBB);
     
-    infoEvent("Detected GCP stop...sending kill to %s", 
-	      c_GCP_SAVEREQ_Counter.getText());
-    g_eventLogger.error("Detected GCP stop...sending kill to %s", 
-                        c_GCP_SAVEREQ_Counter.getText());
+    infoEvent("Detected GCP stop(%d)...sending kill to %s", 
+              cgcpStatus, c_GCP_SAVEREQ_Counter.getText());
+    ndbout_c("Detected GCP stop(%d)...sending kill to %s", 
+	     cgcpStatus, c_GCP_SAVEREQ_Counter.getText());
+    ndbrequire(!c_GCP_SAVEREQ_Counter.done());
     return;
   }
   case GCP_SAVE_LQH_FINISHED:
-    g_eventLogger.error("m_copyReason: %d m_waiting: %d",
+  {
+    jam();
+    /**
+     * We're waiting for a COPY_GCICONF
+     */
+    infoEvent("Detected GCP stop(%d)...sending kill to %s", 
+	      cgcpStatus, c_COPY_GCIREQ_Counter.getText());
+    ndbout_c("Detected GCP stop(%d)...sending kill to %s", 
+	     cgcpStatus, c_COPY_GCIREQ_Counter.getText());
+
+    {
+      NodeReceiverGroup rg(DBDIH, c_COPY_GCIREQ_Counter);
+      signal->theData[0] = 7022;
+      sendSignal(rg, GSN_DUMP_STATE_ORD, signal, 1, JBA);
+    }
+    
+    {
+      NodeReceiverGroup rg(NDBCNTR, c_COPY_GCIREQ_Counter);
+      SystemError * const sysErr = (SystemError*)&signal->theData[0];
+      sysErr->errorCode = SystemError::GCPStopDetected;
+      sysErr->errorRef = reference();
+      sysErr->data1 = cgcpStatus;
+      sysErr->data2 = cgcpOrderBlocked;
+      sendSignal(rg, GSN_SYSTEM_ERROR, signal, 
+                 SystemError::SignalLength, JBA);
+    }
+    ndbrequire(!c_COPY_GCIREQ_Counter.done());
+    return;
+  }
+  case GCP_READY: (void)1;
+  }
+
+dolocal:  
+  ndbout_c("m_copyReason: %d m_waiting: %d",
            c_copyGCIMaster.m_copyReason,
            c_copyGCIMaster.m_waiting);
-    break;
-  case GCP_READY: // shut up lint
-  case GCP_PREPARE_SENT:
-  case GCP_COMMIT_SENT:
-    break;
-  }
   
-  g_eventLogger.error("c_copyGCISlave: sender{Data, Ref} %d %x reason: %d nextWord: %d",
+  ndbout_c("c_copyGCISlave: sender{Data, Ref} %d %x reason: %d nextWord: %d",
 	   c_copyGCISlave.m_senderData,
 	   c_copyGCISlave.m_senderRef,
 	   c_copyGCISlave.m_copyReason,
@@ -11202,6 +11305,9 @@ void Dbdih::crashSystemAtGcpStop(Signal* signal)
 	   c_TCGETOPSIZEREQ_Counter.getText());
   ndbout_c("c_UPDATE_TOREQ_Counter = %s", c_UPDATE_TOREQ_Counter.getText());
 
+  if (local == false)
+  {
+    jam();
     NodeRecordPtr nodePtr;
     for (nodePtr.i = 1; nodePtr.i < MAX_NDB_NODES; nodePtr.i++) {
       jam();
@@ -11219,6 +11325,19 @@ void Dbdih::crashSystemAtGcpStop(Signal* signal)
                    SystemError::SignalLength, JBA);
       }//if
     }//for
+  }
+  else
+  {
+    jam();
+    SystemError * const sysErr = (SystemError*)&signal->theData[0];
+    sysErr->errorCode = SystemError::GCPStopDetected;
+    sysErr->errorRef = reference();
+    sysErr->data1 = cgcpStatus;
+    sysErr->data2 = cgcpOrderBlocked;
+    EXECUTE_DIRECT(NDBCNTR, GSN_SYSTEM_ERROR, 
+                   signal, SystemError::SignalLength);
+    ndbrequire(false);
+  }
   return;
 }//Dbdih::crashSystemAtGcpStop()
 
@@ -14304,6 +14423,12 @@ Dbdih::execDUMP_STATE_ORD(Signal* signal)
       infoEvent(buf);
     }
   }
+
+  if (arg == 7022)
+  {
+    jam();
+    crashSystemAtGcpStop(signal, true);
+  }
 }//Dbdih::execDUMP_STATE_ORD()
 
 void