ndb - bug#25329

  Fix most obvious error in dict/suma error handling of events

storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp

@@ -10151,6 +10151,7 @@ void Dbdict::execSUB_START_REF(Signal* signal)
     SubStartRef* ref = (SubStartRef*) signal->getDataPtrSend();
     ref->senderRef = reference();
     ref->senderData = subbPtr.p->m_senderData;
+    ref->errorCode = err;
     sendSignal(subbPtr.p->m_senderRef, GSN_SUB_START_REF,
 	       signal, SubStartRef::SignalLength2, JBB);
     c_opSubEvent.release(subbPtr);
@@ -10213,6 +10214,7 @@ void Dbdict::execSUB_START_CONF(Signal* signal)
 #ifdef EVENT_PH3_DEBUG
   ndbout_c("DBDICT(Coordinator) got GSN_SUB_START_CONF = (%d)", subbPtr.i);
 #endif
+  subbPtr.p->m_sub_start_conf = *conf;
   subbPtr.p->m_reqTracker.reportConf(c_counterMgr, refToNode(senderRef));
   completeSubStartReq(signal,subbPtr.i,0);
 }
@@ -10252,6 +10254,9 @@ void Dbdict::completeSubStartReq(Signal* signal,
 #ifdef EVENT_DEBUG
   ndbout_c("SUB_START_CONF");
 #endif
+  
+  SubStartConf* conf = (SubStartConf*)signal->getDataPtrSend();
+  * conf = subbPtr.p->m_sub_start_conf;
   sendSignal(subbPtr.p->m_senderRef, GSN_SUB_START_CONF,
 	     signal, SubStartConf::SignalLength, JBB);
   c_opSubEvent.release(subbPtr);
@@ -10373,6 +10378,7 @@ void Dbdict::execSUB_STOP_REF(Signal* signal)
     SubStopRef* ref = (SubStopRef*) signal->getDataPtrSend();
     ref->senderRef = reference();
     ref->senderData = subbPtr.p->m_senderData;
+    ref->errorCode = err;
     sendSignal(subbPtr.p->m_senderRef, GSN_SUB_STOP_REF,
 	       signal, SubStopRef::SignalLength, JBB);
     c_opSubEvent.release(subbPtr);
@@ -10425,6 +10431,7 @@ void Dbdict::execSUB_STOP_CONF(Signal* signal)
    * Coordinator
    */
   ndbrequire(refToBlock(senderRef) == DBDICT);
+  subbPtr.p->m_sub_stop_conf = *conf;
   subbPtr.p->m_reqTracker.reportConf(c_counterMgr, refToNode(senderRef));
   completeSubStopReq(signal,subbPtr.i,0);
 }
@@ -10465,6 +10472,8 @@ void Dbdict::completeSubStopReq(Signal* signal,
 #ifdef EVENT_DEBUG
   ndbout_c("SUB_STOP_CONF");
 #endif
+  SubStopConf* conf = (SubStopConf*)signal->getDataPtrSend();
+  * conf = subbPtr.p->m_sub_stop_conf;
   sendSignal(subbPtr.p->m_senderRef, GSN_SUB_STOP_CONF,
 	     signal, SubStopConf::SignalLength, JBB);
   c_opSubEvent.release(subbPtr);
@@ -10713,6 +10722,7 @@ Dbdict::execSUB_REMOVE_REF(Signal* signal)
       SubRemoveRef* ref = (SubRemoveRef*) signal->getDataPtrSend();
       ref->senderRef = reference();
       ref->senderData = subbPtr.p->m_senderData;
+      ref->errorCode = err;
       sendSignal(subbPtr.p->m_senderRef, GSN_SUB_REMOVE_REF,
 		 signal, SubRemoveRef::SignalLength, JBB);
     }

storage/ndb/src/kernel/blocks/dbdict/Dbdict.hpp

@@ -52,6 +52,7 @@
 #include <signaldata/DropTrig.hpp>
 #include <signaldata/AlterTrig.hpp>
 #include <signaldata/DictLock.hpp>
+#include <signaldata/SumaImpl.hpp>
 #include "SchemaFile.hpp"
 #include <blocks/mutexes.hpp>
 #include <SafeCounter.hpp>
@@ -1632,6 +1633,10 @@ private:
     Uint32 m_senderRef;
     Uint32 m_senderData;
     Uint32 m_errorCode;
+    union {
+      SubStartConf m_sub_start_conf;
+      SubStopConf m_sub_stop_conf;
+    };
     RequestTracker m_reqTracker;
   };
   typedef Ptr<OpSubEvent> OpSubEventPtr;

storage/ndb/src/kernel/blocks/suma/Suma.cpp

@@ -230,7 +230,6 @@ Suma::execREAD_CONFIG_REQ(Signal* signal)
  
   c_startup.m_wait_handover= false; 
   c_failedApiNodes.clear();
-  c_startup.m_restart_server_node_id = 0; // Server for my NR
 
   ReadConfigConf * conf = (ReadConfigConf*)signal->getDataPtrSend();
   conf->senderRef = reference();
@@ -261,6 +260,14 @@ Suma::execSTTOR(Signal* signal) {
 
   if(startphase == 5)
   {
+    if (ERROR_INSERTED(13029)) /* Hold startphase 5 */
+    {
+      sendSignalWithDelay(SUMA_REF, GSN_STTOR, signal,
+                          30, signal->getLength());
+      DBUG_VOID_RETURN;
+    }
+
+    c_startup.m_restart_server_node_id = 0;    
     getNodeGroupMembers(signal);
     if (typeOfStart == NodeState::ST_NODE_RESTART ||
 	typeOfStart == NodeState::ST_INITIAL_NODE_RESTART)
@@ -373,6 +380,8 @@ Suma::execSUMA_START_ME_REF(Signal* signal)
 
   infoEvent("Suma: node %d refused %d", 
 	    c_startup.m_restart_server_node_id, ref->errorCode);
+
+  c_startup.m_restart_server_node_id++;
   send_start_me_req(signal);
 }
 
@@ -887,6 +896,22 @@ Suma::execDUMP_STATE_ORD(Signal* signal){
 		ptr->m_buffer_head.m_page_id);
     }
   }  
+
+  if (tCase == 8006)
+  {
+    SET_ERROR_INSERT_VALUE(13029);
+  }
+
+  if (tCase == 8007)
+  {
+    c_startup.m_restart_server_node_id = MAX_NDB_NODES + 1;
+    SET_ERROR_INSERT_VALUE(13029);
+  }
+
+  if (tCase == 8008)
+  {
+    CLEAR_ERROR_INSERT_VALUE;
+  }
 }
 
 /*************************************************************
@@ -1092,14 +1117,14 @@ Suma::execSUB_CREATE_REQ(Signal* signal)
     }
   } else {
     if (c_startup.m_restart_server_node_id && 
-        refToNode(subRef) != c_startup.m_restart_server_node_id)
+        subRef != calcSumaBlockRef(c_startup.m_restart_server_node_id))
     {
       /**
        * only allow "restart_server" Suma's to come through 
        * for restart purposes
        */
       jam();
-      sendSubStartRef(signal, 1405);
+      sendSubCreateRef(signal, 1415);
       DBUG_VOID_RETURN;
     }
     // Check that id/key is unique
@@ -2232,14 +2257,17 @@ Suma::execSUB_START_REQ(Signal* signal){
   key.m_subscriptionKey       = req->subscriptionKey;
 
   if (c_startup.m_restart_server_node_id && 
-      refToNode(senderRef) != c_startup.m_restart_server_node_id)
+      senderRef != calcSumaBlockRef(c_startup.m_restart_server_node_id))
   {
     /**
      * only allow "restart_server" Suma's to come through 
      * for restart purposes
      */
     jam();
-    sendSubStartRef(signal, 1405);
+    Uint32 err = c_startup.m_restart_server_node_id != RNIL ? 1405 : 
+      SubStartRef::NF_FakeErrorREF;
+    
+    sendSubStartRef(signal, err);
     DBUG_VOID_RETURN;
   }
   
@@ -2454,25 +2482,28 @@ Suma::execSUB_STOP_REQ(Signal* signal){
     DBUG_VOID_RETURN;
   }
 
-  if(!c_subscriptions.find(subPtr, key)){
-    jam();
-    DBUG_PRINT("error", ("not found"));
-    sendSubStopRef(signal, 1407);
-    DBUG_VOID_RETURN;
-  }
-  
   if (c_startup.m_restart_server_node_id && 
-      refToNode(senderRef) != c_startup.m_restart_server_node_id)
+      senderRef != calcSumaBlockRef(c_startup.m_restart_server_node_id))
   {
     /**
      * only allow "restart_server" Suma's to come through 
      * for restart purposes
      */
     jam();
-    sendSubStopRef(signal, 1405);
+    Uint32 err = c_startup.m_restart_server_node_id != RNIL ? 1405 : 
+      SubStopRef::NF_FakeErrorREF;
+    
+    sendSubStopRef(signal, err);
     DBUG_VOID_RETURN;
   }
 
+  if(!c_subscriptions.find(subPtr, key)){
+    jam();
+    DBUG_PRINT("error", ("not found"));
+    sendSubStopRef(signal, 1407);
+    DBUG_VOID_RETURN;
+  }
+  
   if (subPtr.p->m_state == Subscription::LOCKED) {
     jam();
     DBUG_PRINT("error", ("locked"));

storage/ndb/src/kernel/blocks/suma/SumaInit.cpp

@@ -122,6 +122,8 @@ Suma::Suma(Block_context& ctx) :
   
   addRecSignal(GSN_SUB_GCP_COMPLETE_REP, 
 	       &Suma::execSUB_GCP_COMPLETE_REP);
+
+  c_startup.m_restart_server_node_id = RNIL; // Server for my NR
 }
 
 Suma::~Suma()