ndb - bug#18781: close a tiny window (re-commit, try to by-pass merge jam)

ndb/src/kernel/blocks/dbdict/DictLock.txt

@@ -85,10 +85,14 @@ DIH/s
     START_MECONF
 DIH/s
 
-* sp7 - release DICT lock
+* (copy data, omitted)
 
-DIH/s
-    DICT_UNLOCK_ORD
-        DICT/m
+* SL_STARTED - release DICT lock
+
+CNTR/s
+    NODE_START_REP
+        DIH/s
+            DICT_UNLOCK_ORD
+                DICT/m
 
 # vim: set et sw=4:

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

@@ -1599,6 +1599,9 @@ private:
    */
   void startInfoReply(Signal *, Uint32 nodeId);
 
+  // DIH specifics for execNODE_START_REP (sendDictUnlockOrd)
+  void exec_node_start_rep(Signal* signal);
+
   /*
    * Lock master DICT.  Only current use is by starting node
    * during NR.  A pool of slave records is convenient anyway.

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

@@ -1356,24 +1356,6 @@ void Dbdih::execNDB_STTOR(Signal* signal)
     }
     ndbrequire(false);
     break;
-  case ZNDB_SPH7:
-    jam();
-    switch (typestart) {
-    case NodeState::ST_INITIAL_START:
-    case NodeState::ST_SYSTEM_RESTART:
-      jam();
-      ndbsttorry10Lab(signal, __LINE__);
-      return;
-    case NodeState::ST_NODE_RESTART:
-    case NodeState::ST_INITIAL_NODE_RESTART:
-      jam();
-      sendDictUnlockOrd(signal, c_dictLockSlavePtrI_nodeRestart);
-      c_dictLockSlavePtrI_nodeRestart = RNIL;
-      ndbsttorry10Lab(signal, __LINE__);
-      return;
-    }
-    ndbrequire(false);
-    break;
   default:
     jam();
     ndbsttorry10Lab(signal, __LINE__);
@@ -1381,6 +1363,27 @@ void Dbdih::execNDB_STTOR(Signal* signal)
   }//switch
 }//Dbdih::execNDB_STTOR()
 
+void
+Dbdih::exec_node_start_rep(Signal* signal)
+{
+  /*
+   * Send DICT_UNLOCK_ORD when this node is SL_STARTED.
+   *
+   * Sending it before (sp 7) conflicts with code which assumes
+   * SL_STARTING means we are in copy phase of NR.
+   *
+   * NodeState::starting.restartType is not supposed to be used
+   * when SL_STARTED.  Also it seems NODE_START_REP can arrive twice.
+   *
+   * For these reasons there are no consistency checks and
+   * we rely on c_dictLockSlavePtrI_nodeRestart alone.
+   */
+  if (c_dictLockSlavePtrI_nodeRestart != RNIL) {
+    sendDictUnlockOrd(signal, c_dictLockSlavePtrI_nodeRestart);
+    c_dictLockSlavePtrI_nodeRestart = RNIL;
+  }
+}
+
 void
 Dbdih::createMutexes(Signal * signal, Uint32 count){
   Callback c = { safe_cast(&Dbdih::createMutex_done), count };
@@ -1605,6 +1608,7 @@ void Dbdih::nodeRestartPh2Lab(Signal* signal)
 void Dbdih::recvDictLockConf_nodeRestart(Signal* signal, Uint32 data, Uint32 ret)
 {
   ndbrequire(c_dictLockSlavePtrI_nodeRestart == RNIL);
+  ndbrequire(data != RNIL);
   c_dictLockSlavePtrI_nodeRestart = data;
 
   nodeRestartPh2Lab2(signal);

ndb/src/kernel/vm/SimulatedBlock.cpp

@@ -916,6 +916,15 @@ 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)
 {
 }
 

ndb/src/kernel/vm/SimulatedBlock.hpp

@@ -424,6 +424,7 @@ 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;

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

@@ -500,6 +500,10 @@ max-time: 1500
 cmd: testDict
 args: -n TemporaryTables T1 T6 T7 T8 
 
+max-time: 1500
+cmd: testDict
+args: -n Restart_NR2 T1
+
 #
 # TEST NDBAPI
 #