ndb - bug#20185

  Dont be too aggressive in Dbtc::nodeFailCheckTransaction
  let it timeout by 1, so that it does not assert that it has waited too long
  old impl. set timeotu value to 0, making timeout = (ctcTimer - 0) which could be quite big.


ndb/src/kernel/blocks/ERROR_codes.txt:
  error codes
ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  2 new error inserts
  7030 - delay in GCP_PREPARE until checkLocalNodefailComplete is true
  7031 - delay in GCP_PREPARE and die
ndb/src/kernel/blocks/dbtc/DbtcMain.cpp:
  Dont set api con timer to 0, as this might trigger asserion in timeoutfound lab if state == PREPARE_TO_COMMIT
ndb/test/ndbapi/testNodeRestart.cpp:
  testcase
ndb/test/run-test/daily-basic-tests.txt:
  autotest

ndb/src/kernel/blocks/ERROR_codes.txt

@@ -63,6 +63,9 @@ Delay GCP_SAVEREQ by 10 secs
 
 7165: Delay INCL_NODE_REQ in starting node yeilding error in GCP_PREPARE
 
+7030: Delay in GCP_PREPARE until node has completed a node failure
+7031: Delay in GCP_PREPARE and die 3s later
+
 ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
 -----------------------------------------------------------------
 

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

@@ -5417,6 +5417,12 @@ Dbdih::checkLocalNodefailComplete(Signal* signal, Uint32 failedNodeId,
     return;
   }
 
+  if (ERROR_INSERTED(7030))
+  {
+    ndbout_c("Reenable GCP_PREPARE");
+    CLEAR_ERROR_INSERT_VALUE;
+  }
+  
   NFCompleteRep * const nf = (NFCompleteRep *)&signal->theData[0];
   nf->blockNo = DBDIH;
   nf->nodeId = cownNodeId;
@@ -7459,6 +7465,16 @@ void Dbdih::execGCP_PREPARE(Signal* signal)
 {
   jamEntry();
   CRASH_INSERTION(7005);
+
+  if (ERROR_INSERTED(7030))
+  {
+    cgckptflag = true;
+    ndbout_c("Delayed GCP_PREPARE 5s");
+    sendSignalWithDelay(reference(), GSN_GCP_PREPARE, signal, 5000,
+			signal->getLength());
+    return;
+  }
+  
   Uint32 masterNodeId = signal->theData[0];
   Uint32 gci = signal->theData[1];
   BlockReference retRef = calcDihBlockRef(masterNodeId);
@@ -7471,6 +7487,14 @@ void Dbdih::execGCP_PREPARE(Signal* signal)
   cgcpParticipantState = GCP_PARTICIPANT_PREPARE_RECEIVED;
   cnewgcp = gci;
 
+  if (ERROR_INSERTED(7031))
+  {
+    ndbout_c("Crashing delayed in GCP_PREPARE 3s");
+    signal->theData[0] = 9999;
+    sendSignalWithDelay(CMVMI_REF, GSN_NDB_TAMPER, signal, 3000, 1);
+    return;
+  }
+  
   signal->theData[0] = cownNodeId;
   signal->theData[1] = gci;  
   sendSignal(retRef, GSN_GCP_PREPARECONF, signal, 2, JBA);

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

@@ -7002,18 +7002,20 @@ Dbtc::nodeFailCheckTransactions(Signal* signal,
 {
   jam();
   Ptr<ApiConnectRecord> transPtr;
+  Uint32 TtcTimer = ctcTimer;
+  Uint32 TapplTimeout = c_appl_timeout_value;
   for (transPtr.i = transPtrI; transPtr.i < capiConnectFilesize; transPtr.i++)
   {
     ptrCheckGuard(transPtr, capiConnectFilesize, apiConnectRecord); 
     if (transPtr.p->m_transaction_nodes.get(failedNodeId))
     {
       jam();
+
       // Force timeout regardless of state      
-      Uint32 save = c_appl_timeout_value;
       c_appl_timeout_value = 1;
-      setApiConTimer(transPtr.i, 0, __LINE__);
+      setApiConTimer(transPtr.i, TtcTimer - 2, __LINE__);
       timeOutFoundLab(signal, transPtr.i, ZNODEFAIL_BEFORE_COMMIT);
-      c_appl_timeout_value = save;
+      c_appl_timeout_value = TapplTimeout;
     }
     
     // Send CONTINUEB to continue later

ndb/test/ndbapi/testNodeRestart.cpp

@@ -868,6 +868,56 @@ runBug18612SR(NDBT_Context* ctx, NDBT_Step* step){
   return NDBT_OK;
 }
 
+int runBug20185(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  HugoOperations hugoOps(*ctx->getTab());
+  Ndb* pNdb = GETNDB(step);
+  
+  int dump[] = { 7090, 20 } ;
+  if (restarter.dumpStateAllNodes(dump, 2))
+    return NDBT_FAILED;
+  
+  NdbSleep_MilliSleep(3000);
+
+  if(hugoOps.startTransaction(pNdb) != 0)
+    return NDBT_FAILED;
+  
+  if(hugoOps.pkUpdateRecord(pNdb, 1, 1) != 0)
+    return NDBT_FAILED;
+  
+  if (hugoOps.execute_NoCommit(pNdb) != 0)
+    return NDBT_FAILED;
+  
+  int nodeId;
+  const int node = hugoOps.getTransaction()->getConnectedNodeId();
+  do {
+    nodeId = restarter.getDbNodeId(rand() % restarter.getNumDbNodes());
+  } while (nodeId == node);
+  
+  if (restarter.insertErrorInAllNodes(7030))
+    return NDBT_FAILED;
+  
+  if (restarter.insertErrorInNode(nodeId, 7031))
+    return NDBT_FAILED;
+  
+  NdbSleep_MilliSleep(500);
+  
+  if (hugoOps.execute_Commit(pNdb) == 0)
+    return NDBT_FAILED;
+
+  NdbSleep_MilliSleep(3000);
+
+  restarter.waitClusterStarted();
+  
+  if (restarter.dumpStateAllNodes(dump, 1))
+    return NDBT_FAILED;
+  
+  return NDBT_OK;
+}
+
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -1175,6 +1225,12 @@ TESTCASE("Bug18612SR",
   STEP(runBug18612SR);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug20185",
+	 ""){
+  INITIALIZER(runLoadTable);
+  STEP(runBug20185);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -470,6 +470,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug18612SR T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug20185 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench