ndb - bug#18352

Use variable waitfor_response_timeout (depending on TransactionDeadLockTimeout)
  When getting 4012, set NeedAbort and ReleaseOnClose


ndb/src/ndbapi/NdbConnection.cpp:
  Use variable for WAITFOR_RESPONSE_TIMEOUT
ndb/src/ndbapi/Ndbif.cpp:
  Use variable timeout for waitfor, 
    when receiving 4012, set NeedAbort and ReleaseOnClose
ndb/src/ndbapi/TransporterFacade.cpp:
  Init wait_for_response_timoue as max TRANSACTION_DEADLOCK_TIMEOUT
ndb/src/ndbapi/TransporterFacade.hpp:
  Init wait_for_response_timoue as max TRANSACTION_DEADLOCK_TIMEOUT
ndb/test/ndbapi/testTimeout.cpp:
  Add testcase for 4012
ndb/test/run-test/daily-basic-tests.txt:
  Add testcase for 4012

ndb/src/ndbapi/NdbConnection.cpp

@@ -450,12 +450,12 @@ NdbConnection::executeNoBlobs(ExecType aTypeOfExec,
 //------------------------------------------------------------------------
   Ndb* tNdb = theNdb;
 
+  Uint32 timeout = TransporterFacade::instance()->m_waitfor_timeout;
   m_waitForReply = false;
   executeAsynchPrepare(aTypeOfExec, NULL, NULL, abortOption);
   if (m_waitForReply){
     while (1) {
-      int noOfComp = tNdb->sendPollNdb((3 * WAITFOR_RESPONSE_TIMEOUT),
+      int noOfComp = tNdb->sendPollNdb(3 * timeout, 1, forceSend);
-                                       1, forceSend);
       if (noOfComp == 0) {
         /** 
          * This timeout situation can occur if NDB crashes.

ndb/src/ndbapi/Ndbif.cpp

@@ -954,23 +954,25 @@ Ndb::pollCompleted(NdbConnection** aCopyArray)
 void
 Ndb::check_send_timeout()
 {
+  Uint32 timeout = TransporterFacade::instance()->m_waitfor_timeout;
   NDB_TICKS current_time = NdbTick_CurrentMillisecond();
   if (current_time - the_last_check_time > 1000) {
     the_last_check_time = current_time;
     Uint32 no_of_sent = theNoOfSentTransactions;
     for (Uint32 i = 0; i < no_of_sent; i++) {
       NdbConnection* a_con = theSentTransactionsArray[i];
-      if ((current_time - a_con->theStartTransTime) >
+      if ((current_time - a_con->theStartTransTime) > timeout)
-          WAITFOR_RESPONSE_TIMEOUT) {
+      {
 #ifdef VM_TRACE
         a_con->printState();
 	Uint32 t1 = a_con->theTransactionId;
 	Uint32 t2 = a_con->theTransactionId >> 32;
-	ndbout_c("[%.8x %.8x]", t1, t2);
+	ndbout_c("4012 [%.8x %.8x]", t1, t2);
-	abort();
+	//abort();
 #endif
+        a_con->theReleaseOnClose = true;
         a_con->setOperationErrorCodeAbort(4012);
-        a_con->theCommitStatus = NdbConnection::Aborted;
+	a_con->theCommitStatus = NdbConnection::NeedAbort;
         a_con->theCompletionStatus = NdbConnection::CompletedFailure;
         a_con->handleExecuteCompletion();
         remove_sent_list(i);

ndb/src/ndbapi/TransporterFacade.cpp

@@ -567,6 +567,20 @@ TransporterFacade::init(Uint32 nodeId, const ndb_mgm_configuration* props)
   }
 #endif
   
+  Uint32 timeout = 120000;
+  iter.first();
+  for (iter.first(); iter.valid(); iter.next())
+  {
+    Uint32 tmp1 = 0, tmp2 = 0;
+    iter.get(CFG_DB_TRANSACTION_CHECK_INTERVAL, &tmp1);
+    iter.get(CFG_DB_TRANSACTION_DEADLOCK_TIMEOUT, &tmp2);
+    tmp1 += tmp2;
+    if (tmp1 > timeout)
+      timeout = tmp1;
+  }
+  m_waitfor_timeout = timeout;
+  ndbout_c("Using waitfor: %d", timeout);
+  
   if (!theTransporterRegistry->start_service(m_socket_server)){
     ndbout_c("Unable to start theTransporterRegistry->start_service");
     DBUG_RETURN(false);

ndb/src/ndbapi/TransporterFacade.hpp

@@ -172,6 +172,7 @@ private:
    */
 public:
   STATIC_CONST( MAX_NO_THREADS = 4711 );
+  Uint32 m_waitfor_timeout; // in milli seconds...
 private:
 
   struct ThreadData {

ndb/test/ndbapi/testTimeout.cpp

@@ -24,6 +24,7 @@
 
 #define TIMEOUT (Uint32)3000
 Uint32 g_org_timeout = 3000;
+Uint32 g_org_deadlock = 3000;
 
 int
 setTransactionTimeout(NDBT_Context* ctx, NDBT_Step* step){
@@ -59,6 +60,60 @@ resetTransactionTimeout(NDBT_Context* ctx, NDBT_Step* step){
   return NDBT_OK;
 }
 
+int
+setDeadlockTimeout(NDBT_Context* ctx, NDBT_Step* step){
+  NdbRestarter restarter;
+  int timeout = ctx->getProperty("TransactionDeadlockTimeout", TIMEOUT);
+  
+  NdbConfig conf(GETNDB(step)->getNodeId()+1);
+  unsigned int nodeId = conf.getMasterNodeId();
+  if (!conf.getProperty(nodeId,
+			NODE_TYPE_DB, 
+			CFG_DB_TRANSACTION_DEADLOCK_TIMEOUT,
+			&g_org_deadlock))
+    return NDBT_FAILED;
+  
+  g_err << "Setting timeout: " << timeout << endl;
+  int val[] = { DumpStateOrd::TcSetTransactionTimeout, timeout };
+  if(restarter.dumpStateAllNodes(val, 2) != 0){
+    return NDBT_FAILED;
+  }
+  
+  return NDBT_OK;
+}
+
+int
+getDeadlockTimeout(NDBT_Context* ctx, NDBT_Step* step){
+  NdbRestarter restarter;
+  
+  Uint32 val = 0;
+  NdbConfig conf(GETNDB(step)->getNodeId()+1);
+  unsigned int nodeId = conf.getMasterNodeId();
+  if (!conf.getProperty(nodeId,
+			NODE_TYPE_DB, 
+			CFG_DB_TRANSACTION_DEADLOCK_TIMEOUT,
+			&val))
+    return NDBT_FAILED;
+
+  if (val < 120000)
+    val = 120000;
+  ctx->setProperty("TransactionDeadlockTimeout", 4*val);
+  
+  return NDBT_OK;
+}
+
+int
+resetDeadlockTimeout(NDBT_Context* ctx, NDBT_Step* step){
+  NdbRestarter restarter;
+  
+  int val[] = { DumpStateOrd::TcSetTransactionTimeout, g_org_deadlock };
+  if(restarter.dumpStateAllNodes(val, 2) != 0){
+    return NDBT_FAILED;
+  }
+  
+  return NDBT_OK;
+}
+
 
 int runLoadTable(NDBT_Context* ctx, NDBT_Step* step){
 
@@ -374,6 +429,43 @@ int runBuddyTransNoTimeout(NDBT_Context* ctx, NDBT_Step* step){
   return result;
 }
 
+int 
+runError4012(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int stepNo = step->getStepNo();
+  
+  int timeout = ctx->getProperty("TransactionDeadlockTimeout", TIMEOUT);
+
+  HugoOperations hugoOps(*ctx->getTab());
+  Ndb* pNdb = GETNDB(step);
+
+  do{
+    // Commit transaction
+    CHECK(hugoOps.startTransaction(pNdb) == 0);
+    CHECK(hugoOps.pkUpdateRecord(pNdb, 0) == 0);
+    int ret = hugoOps.execute_NoCommit(pNdb);
+    if (ret == 0)
+    {
+      int sleep = timeout;
+      ndbout << "Sleeping for " << sleep << " milliseconds" << endl;
+      NdbSleep_MilliSleep(sleep);
+      
+      // Expect that transaction has NOT timed-out
+      CHECK(hugoOps.execute_Commit(pNdb) == 0);
+    }
+    else
+    {
+      CHECK(ret == 4012);
+    }
+  } while(false);
+  
+  hugoOps.closeTransaction(pNdb);
+  
+  return result;
+}
+
+
 NDBT_TESTSUITE(testTimeout);
 TESTCASE("DontTimeoutTransaction", 
 	 "Test that the transaction does not timeout "\
@@ -465,6 +557,15 @@ TESTCASE("BuddyTransNoTimeout5",
   FINALIZER(resetTransactionTimeout);
   FINALIZER(runClearTable);
 }
+TESTCASE("Error4012", ""){
+  TC_PROPERTY("TransactionDeadlockTimeout", 120000);
+  INITIALIZER(runLoadTable);
+  INITIALIZER(getDeadlockTimeout);
+  INITIALIZER(setDeadlockTimeout);
+  STEPS(runError4012, 2);
+  FINALIZER(runClearTable);
+}
+
 NDBT_TESTSUITE_END(testTimeout);
 
 int main(int argc, const char** argv){

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

@@ -236,6 +236,10 @@ max-time: 500
 cmd: testTimeout
 args: -n TimeoutRandTransaction T1
 
+max-time: 600
+cmd: testTimeout
+args: -n Error4012 T1
+
 # SCAN TESTS
 #
 max-time: 500