ndb - bug#29364 - "SQL queries hang while data node in start phase 5"

  In TC init node status for already started nodes during node restart
  (not present in 5.1)


ndb/src/kernel/blocks/dbtc/DbtcMain.cpp:
  on node restart, make sure to init all already started nodes to NF_STARTED
ndb/test/ndbapi/testNodeRestart.cpp:
  test prg
ndb/test/run-test/daily-basic-tests.txt:
  test prg

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

@@ -882,6 +882,12 @@ void Dbtc::execREAD_NODESCONF(Signal* signal)
         hostptr.p->hostStatus = HS_ALIVE;
 	c_alive_nodes.set(i);
       }//if
+
+      if (NodeBitmask::get(readNodes->startedNodes, i))
+      {
+        jam();
+        hostptr.p->m_nf_bits = HostRecord::NF_STARTED;
+      }
     }//if
   }//for
   ndbsttorry010Lab(signal);
@@ -10108,6 +10114,7 @@ void Dbtc::inithost(Signal* signal)
     hostptr.p->noOfWordsTCINDXCONF = 0;
     hostptr.p->noOfPackedWordsLqh = 0;
     hostptr.p->hostLqhBlockRef = calcLqhBlockRef(hostptr.i);
+    hostptr.p->m_nf_bits = 0;
   }//for
   c_alive_nodes.clear();
 }//Dbtc::inithost()

ndb/test/ndbapi/testNodeRestart.cpp

@@ -944,12 +944,62 @@ int runBug24717(NDBT_Context* ctx, NDBT_Step* step){
     
     restarter.startNodes(&nodeId, 1);
     
-    for (Uint32 i = 0; i < 100; i++)
-    {
-      hugoTrans.pkReadRecords(pNdb, 100, 1, NdbOperation::LM_CommittedRead);
-    }
-    
+    do {
+      for (Uint32 i = 0; i < 100; i++)
+      {
+        hugoTrans.pkReadRecords(pNdb, 100, 1, NdbOperation::LM_CommittedRead);
+      }
+    } while (restarter.waitClusterStarted(5) != 0);
+  }
+  
+  return NDBT_OK;
+}
+
+int 
+runBug29364(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  Ndb* pNdb = GETNDB(step);
+  
+  HugoTransactions hugoTrans(*ctx->getTab());
+
+  if (restarter.getNumDbNodes() < 4)
+    return NDBT_OK;
+
+  int dump0[] = { 9000, 0 } ;
+  int dump1[] = { 9001, 0 } ;
+  Uint32 ownNode = refToNode(pNdb->getReference());
+  dump0[1] = ownNode;
+
+  for (; loops; loops --)
+  {
+    int node0 = restarter.getDbNodeId(rand() % restarter.getNumDbNodes());
+    int node1 = restarter.getRandomNodeOtherNodeGroup(node0, rand());
+
+    restarter.restartOneDbNode(node0, false, true, true);
+    restarter.waitNodesNoStart(&node0, 1);
+    restarter.startNodes(&node0, 1);
     restarter.waitClusterStarted();
+
+    restarter.restartOneDbNode(node1, false, true, true);    
+    restarter.waitNodesNoStart(&node1, 1);
+    if (restarter.dumpStateOneNode(node1, dump0, 2))
+      return NDBT_FAILED;
+
+    restarter.startNodes(&node1, 1);    
+    
+    do {
+      
+      for (Uint32 i = 0; i < 100; i++)
+      {
+        hugoTrans.pkReadRecords(pNdb, 100, 1, NdbOperation::LM_CommittedRead);
+      }
+    } while (restarter.waitClusterStarted(5) != 0);
+    
+    if (restarter.dumpStateOneNode(node1, dump1, 1))
+      return NDBT_FAILED;
   }
   
   return NDBT_OK;
@@ -1553,6 +1603,9 @@ TESTCASE("Bug27003", ""){
 TESTCASE("Bug27283", ""){
   INITIALIZER(runBug27283);
 }
+TESTCASE("Bug29364", ""){
+  INITIALIZER(runBug29364);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -477,6 +477,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug24717 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug29364 T1
+
 max-time: 1000
 cmd: testNodeRestart
 args: -n Bug25364 T1