ndb - bug#27169

Fix bug in SUMA::resend_bucket which could cause mysqld to crash


storage/ndb/src/kernel/blocks/suma/Suma.cpp:
  Remove *len* part from sz,
    or an extra word will be sent (sometimes) which will cause event-api barf
storage/ndb/test/ndbapi/test_event.cpp:
  test prg for bug#27169
storage/ndb/test/run-test/daily-basic-tests.txt:
  test prg for bug#27169

storage/ndb/src/kernel/blocks/suma/Suma.cpp

@@ -4997,8 +4997,11 @@ Suma::resend_bucket(Signal* signal, Uint32 buck, Uint32 min_gci,
     {
       continue;
     }
+
+    ndbrequire(sz);
+    sz --; // remove *len* part of sz
     
-    if(sz == 1)
+    if(sz == 0)
     {
       SubGcpCompleteRep * rep = (SubGcpCompleteRep*)signal->getDataPtrSend();
       rep->gci = last_gci;

storage/ndb/test/ndbapi/test_event.cpp

@@ -836,6 +836,67 @@ end:
   DBUG_RETURN(result);
 }
 
+int runEventListenerUntilStopped(NDBT_Context* ctx, NDBT_Step* step)
+{
+  
+  int result = NDBT_OK;
+  const NdbDictionary::Table * table= ctx->getTab();
+  HugoTransactions hugoTrans(* table);
+
+  char buf[1024];
+  sprintf(buf, "%s_EVENT", table->getName());
+  NdbEventOperation *pOp, *pCreate = 0;
+  pCreate = pOp = GETNDB(step)->createEventOperation(buf);
+  if ( pOp == NULL ) {
+    g_err << "Event operation creation failed on %s" << buf << endl;
+    return NDBT_FAILED;
+  }
+  
+  int i;
+  int n_columns= table->getNoOfColumns();
+  NdbRecAttr* recAttr[1024];
+  NdbRecAttr* recAttrPre[1024];
+  for (i = 0; i < n_columns; i++) {
+    recAttr[i]    = pOp->getValue(table->getColumn(i)->getName());
+    recAttrPre[i] = pOp->getPreValue(table->getColumn(i)->getName());
+  }
+
+  if (pOp->execute()) 
+  { // This starts changes to "start flowing"
+    g_err << "execute operation execution failed: \n";
+    g_err << pOp->getNdbError().code << " "
+	  << pOp->getNdbError().message << endl;
+    result = NDBT_FAILED;
+    goto end;
+  }
+  
+  Ndb* ndb= GETNDB(step);
+  while(!ctx->isTestStopped())
+  {
+    Uint64 curr_gci = 0;
+    while(!ctx->isTestStopped())
+    {
+      ndb->pollEvents(100, &curr_gci);
+      while ((pOp= ndb->nextEvent()) != 0)
+      {
+	assert(pOp == pCreate);
+      } 
+    }
+  }
+  
+end:
+  if(pCreate)
+  {
+    if (GETNDB(step)->dropEventOperation(pCreate)) {
+      g_err << "dropEventOperation execution failed "
+	    << GETNDB(step)->getNdbError().code << " "
+	    << GETNDB(step)->getNdbError().message << endl;
+      result = NDBT_FAILED;
+    }
+  }
+  return result;
+}
+
 int runRestarter(NDBT_Context* ctx, NDBT_Step* step){
   int result = NDBT_OK;
   int loops = ctx->getNumLoops();
@@ -877,6 +938,51 @@ int runRestarter(NDBT_Context* ctx, NDBT_Step* step){
   return result;
 }
 
+int runRestarterLoop(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  NdbRestarter restarter;
+  int i = 0;
+  int lastId = 0;
+
+  if (restarter.getNumDbNodes() < 2){
+    ctx->stopTest();
+    return NDBT_OK;
+  }
+
+  if(restarter.waitClusterStarted(60) != 0){
+    g_err << "Cluster failed to start" << endl;
+    return NDBT_FAILED;
+  }
+  
+  while(result != NDBT_FAILED 
+	&& !ctx->isTestStopped() 
+	&& i < loops)
+  {
+    int id = lastId % restarter.getNumDbNodes();
+    int nodeId = restarter.getDbNodeId(id);
+    ndbout << "Restart node " << nodeId << endl; 
+    if(restarter.restartOneDbNode(nodeId, false, false, true) != 0){
+      g_err << "Failed to restartNextDbNode" << endl;
+      result = NDBT_FAILED;
+      break;
+    }    
+    
+    if(restarter.waitClusterStarted(60) != 0){
+      g_err << "Cluster failed to start" << endl;
+      result = NDBT_FAILED;
+      break;
+    }
+    
+    lastId++;
+    i++;
+  }
+
+  ctx->stopTest();
+  return result;
+}
+
 Vector<const NdbDictionary::Table*> pTabs;
 Vector<const NdbDictionary::Table*> pShadowTabs;
 
@@ -1616,6 +1722,42 @@ runSubscribeUnsubscribe(NDBT_Context* ctx, NDBT_Step* step)
   return NDBT_OK;
 }
 
+int 
+runScanUpdateUntilStopped(NDBT_Context* ctx, NDBT_Step* step){
+  int records = ctx->getNumRecords();
+  int parallelism = ctx->getProperty("Parallelism", (Uint32)0);
+  int abort = ctx->getProperty("AbortProb", (Uint32)0);
+  HugoTransactions hugoTrans(*ctx->getTab());
+  while (ctx->isTestStopped() == false) 
+  {
+    if (hugoTrans.scanUpdateRecords(GETNDB(step), records, abort, 
+				    parallelism) == NDBT_FAILED){
+      return NDBT_FAILED;
+    }
+  }
+  return NDBT_OK;
+}
+
+int 
+runInsertDeleteUntilStopped(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int records = ctx->getNumRecords();
+  HugoTransactions hugoTrans(*ctx->getTab());
+  UtilTransactions utilTrans(*ctx->getTab());
+  while (ctx->isTestStopped() == false) 
+  {
+    if (hugoTrans.loadTable(GETNDB(step), records, 1) != 0){
+      return NDBT_FAILED;
+    }
+    if (utilTrans.clearTable(GETNDB(step),  records) != 0){
+      return NDBT_FAILED;
+    }
+  }
+  
+  return NDBT_OK;
+}
+
 NDBT_TESTSUITE(test_event);
 TESTCASE("BasicEventOperation", 
 	 "Verify that we can listen to Events"
@@ -1737,6 +1879,14 @@ TESTCASE("SubscribeUnsubscribe",
   STEPS(runSubscribeUnsubscribe, 16);
   FINALIZER(runDropEvent);
 }
+TESTCASE("Bug27169", ""){
+  INITIALIZER(runCreateEvent);
+  STEP(runEventListenerUntilStopped);
+  STEP(runInsertDeleteUntilStopped);
+  STEP(runScanUpdateUntilStopped);
+  STEP(runRestarterLoop);
+  FINALIZER(runDropEvent);
+}
 NDBT_TESTSUITE_END(test_event);
 
 int main(int argc, const char** argv){

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

@@ -788,6 +788,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug25468 T1
 
+max-time: 1000
+cmd: test_event
+args: -l 10 -n Bug27169 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench