ndb - bug#18414

  Fix timeout during ABORT when ZABORT_TIMEOUT_BREAK is outstanding

ndb/src/kernel/blocks/ERROR_codes.txt

@@ -226,6 +226,8 @@ Delay execution of COMPLETECONF signal 2 seconds to generate time-out.
 8045: (ABORTCONF only as part of take-over)
 Delay execution of ABORTCONF signal 2 seconds to generate time-out.
 
+8050: Send ZABORT_TIMEOUT_BREAK delayed
+
 ERROR CODES FOR TESTING TIME-OUT HANDLING IN DBTC
 -------------------------------------------------
 

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

@@ -5982,10 +5982,6 @@ void Dbdih::MASTER_LCPhandling(Signal* signal, Uint32 failedNodeId)
   signal->theData[0] = 7012;
   execDUMP_STATE_ORD(signal);
 
-  signal->theData[0] = 7015;
-  signal->theData[1] = 0;
-  execDUMP_STATE_ORD(signal);
-
   c_lcpMasterTakeOverState.set(LMTOS_IDLE, __LINE__);
 
   checkLocalNodefailComplete(signal, failedNodePtr.i, NF_LCP_TAKE_OVER);

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

@@ -6386,6 +6386,7 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
     return;
   }
   
+  bool found = false;
   OperationState tmp[16];
   
   Uint32 TloopCount = 0;
@@ -6393,7 +6394,31 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
     jam();
     if (tcConnectptr.i == RNIL) {
       jam();
-      if (Tcheck == 0) {
+
+#ifdef VM_TRACE
+      ndbout_c("found: %d Tcheck: %d apiConnectptr.p->counter: %d",
+	       found, Tcheck, apiConnectptr.p->counter);
+#endif
+      if (found || apiConnectptr.p->counter)
+      {
+	jam();
+	/**
+	 * We sent atleast one ABORT/ABORTED
+	 *   or ZABORT_TIMEOUT_BREAK is in job buffer
+	 *   wait for reception...
+	 */
+	return;
+      }
+      
+      if (Tcheck == 1)
+      {
+	jam();
+	releaseAbortResources(signal);
+	return;
+      }
+      
+      if (Tcheck == 0)
+      {
         jam();
 	/*------------------------------------------------------------------
 	 * All nodes had already reported ABORTED for all tcConnect records.
@@ -6402,9 +6427,11 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
 	 *------------------------------------------------------------------*/
 	char buf[96]; buf[0] = 0;
 	char buf2[96];
-	BaseString::snprintf(buf, sizeof(buf), "TC %d: %d ops:",
-		 __LINE__, apiConnectptr.i);
-	for(Uint32 i = 0; i<TloopCount; i++){
+	BaseString::snprintf(buf, sizeof(buf), "TC %d: %d counter: %d ops:",
+			     __LINE__, apiConnectptr.i,
+			     apiConnectptr.p->counter);
+	for(Uint32 i = 0; i<TloopCount; i++)
+	{
 	  BaseString::snprintf(buf2, sizeof(buf2), "%s %d", buf, tmp[i]);
 	  BaseString::snprintf(buf, sizeof(buf), buf2);
 	}
@@ -6412,7 +6439,9 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
 	ndbout_c(buf);
 	ndbrequire(false);
 	releaseAbortResources(signal);
+	return;
       }
+      
       return;
     }//if
     TloopCount++;
@@ -6427,7 +6456,16 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
       signal->theData[0] = TcContinueB::ZABORT_TIMEOUT_BREAK;
       signal->theData[1] = tcConnectptr.i;
       signal->theData[2] = apiConnectptr.i;      
-      sendSignal(cownref, GSN_CONTINUEB, signal, 3, JBB);
+      if (ERROR_INSERTED(8050))
+      {
+	ndbout_c("sending ZABORT_TIMEOUT_BREAK delayed (%d %d)", 
+		 Tcheck, apiConnectptr.p->counter);
+	sendSignalWithDelay(cownref, GSN_CONTINUEB, signal, 2000, 3);
+      }
+      else
+      {
+	sendSignal(cownref, GSN_CONTINUEB, signal, 3, JBB);
+      }
       return;
     }//if
     ptrCheckGuard(tcConnectptr, ctcConnectFilesize, tcConnectRecord);
@@ -6450,7 +6488,7 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
         jam();
         if (tcConnectptr.p->tcNodedata[Ti] != 0) {
           TloopCount += 31;
-          Tcheck = 1;
+	  found = true;
           hostptr.i = tcConnectptr.p->tcNodedata[Ti];
           ptrCheckGuard(hostptr, chostFilesize, hostRecord);
           if (hostptr.p->hostStatus == HS_ALIVE) {
@@ -7007,8 +7045,6 @@ void Dbtc::execTAKE_OVERTCCONF(Signal* signal)
   hostptr.i = tfailedNodeId;
   ptrCheckGuard(hostptr, chostFilesize, hostRecord);
 
-  ndbout_c("received execTAKE_OVERTCCONF(%d) from %x (%x)",
-	   tfailedNodeId, signal->getSendersBlockRef(), reference());
   if (signal->getSendersBlockRef() != reference())
   {
     jam();

ndb/test/ndbapi/testNodeRestart.cpp

@@ -581,6 +581,73 @@ runBug16772(NDBT_Context* ctx, NDBT_Step* step){
   return ret ? NDBT_OK : NDBT_FAILED;
 }
 
+int 
+runBug18414(NDBT_Context* ctx, NDBT_Step* step){
+
+  NdbRestarter restarter;
+  if (restarter.getNumDbNodes() < 2)
+  {
+    ctx->stopTest();
+    return NDBT_OK;
+  }
+
+  Ndb* pNdb = GETNDB(step);
+  HugoOperations hugoOps(*ctx->getTab());
+  HugoTransactions hugoTrans(*ctx->getTab());
+  int loop = 0;
+  do 
+  {
+    if(hugoOps.startTransaction(pNdb) != 0)
+      goto err;
+    
+    if(hugoOps.pkUpdateRecord(pNdb, 0, 128, rand()) != 0)
+      goto err;
+    
+    if(hugoOps.execute_NoCommit(pNdb) != 0)
+      goto err;
+
+    int node1 = hugoOps.getTransaction()->getConnectedNodeId();
+    int node2 = restarter.getRandomNodeSameNodeGroup(node1, rand());
+    
+    if (node1 == -1 || node2 == -1)
+      break;
+    
+    if (loop & 1)
+    {
+      if (restarter.insertErrorInNode(node1, 8050))
+	goto err;
+    }
+    
+    if (restarter.insertErrorInNode(node2, 5003))
+      goto err;
+    
+    int res= hugoOps.execute_Rollback(pNdb);
+  
+    if (restarter.waitNodesNoStart(&node2, 1) != 0)
+      goto err;
+    
+    if (restarter.insertErrorInAllNodes(0))
+      goto err;
+    
+    if (restarter.startNodes(&node2, 1) != 0)
+      goto err;
+    
+    if (restarter.waitClusterStarted() != 0)
+      goto err;
+    
+    if (hugoTrans.scanUpdateRecords(pNdb, 128) != 0)
+      goto err;
+
+    hugoOps.closeTransaction(pNdb);
+    
+  } while(++loop < 5);
+  
+  return NDBT_OK;
+  
+err:
+  hugoOps.closeTransaction(pNdb);
+  return NDBT_FAILED;    
+}
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -870,6 +937,12 @@ TESTCASE("Bug16772",
 	 "Test bug with restarting before NF handling is complete"){
   STEP(runBug16772);
 }
+TESTCASE("Bug18414",
+	 "Test bug with NF during NR"){
+  INITIALIZER(runLoadTable);
+  STEP(runBug18414);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

ndb/test/ndbapi/testTimeout.cpp

@@ -173,8 +173,11 @@ int runTimeoutTrans(NDBT_Context* ctx, NDBT_Step* step){
       NdbSleep_MilliSleep(sleep);
       
       // Expect that transaction has timed-out
-      CHECK(hugoOps.execute_Commit(pNdb) == 237); 
-
+      int ret = hugoOps.execute_Commit(pNdb);
+      CHECK(ret != 0);
+      NdbError err = pNdb->getNdbError(ret);
+      CHECK(err.classification == NdbError::TimeoutExpired);
+      
     } while(false);
 
     hugoOps.closeTransaction(pNdb);

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

@@ -458,6 +458,10 @@ max-time: 500
 cmd: testSystemRestart
 args: -n Bug18385 T1
 
+max-time: 500
+cmd: testNodeRestart
+args: -n Bug18414 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench