ndb - bug#20185

  second try - handle CS_PREPARE_TO_COMMIT explicitly

storage/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp

@@ -7093,15 +7093,20 @@ Dbtc::nodeFailCheckTransactions(Signal* signal,
   for (transPtr.i = transPtrI; transPtr.i < capiConnectFilesize; transPtr.i++)
   {
     ptrCheckGuard(transPtr, capiConnectFilesize, apiConnectRecord); 
+    Uiint32 state = transPtr.p->apiConnectstate;
     if (transPtr.p->m_transaction_nodes.get(failedNodeId))
     {
       jam();
-
-      // Force timeout regardless of state      
-      c_appl_timeout_value = 1;
-      setApiConTimer(transPtr.i, TtcTimer - 2, __LINE__);
-      timeOutFoundLab(signal, transPtr.i, ZNODEFAIL_BEFORE_COMMIT);
-      c_appl_timeout_value = TapplTimeout;
+      
+      // avoid assertion in timeoutfoundlab
+      if (state != CS_PREPARE_TO_COMMIT)
+      {
+	// Force timeout regardless of state      
+	c_appl_timeout_value = 1;
+	setApiConTimer(transPtr.i, TtcTimer - 2, __LINE__);
+	timeOutFoundLab(signal, transPtr.i, ZNODEFAIL_BEFORE_COMMIT);
+	c_appl_timeout_value = TapplTimeout;
+      }
     }
     
     // Send CONTINUEB to continue later