BUG#13985: Cluster: ndb_mgm "status" command can return incorrect data node status

- only force HB to data nodes
- flag for if we broadcast condition on receipt of HB


ndb/src/mgmsrv/MgmtSrvr.cpp:
  Add get_connected_ndb_nodes to check status for connected data nodes only
ndb/src/mgmsrv/MgmtSrvr.hpp:
  add prototype for get_connected_ndb_nodes
ndb/src/mgmsrv/Services.cpp:
  only force HB to NDBD nodes
ndb/src/ndbapi/ClusterMgr.cpp:
  flag to control if we send the condition
ndb/src/ndbapi/ClusterMgr.hpp:
  flag for if we broadcast condition on receipt of hb

5 files changed, 36 insertions, 6 deletions

diff --git a/ndb/src/mgmsrv/MgmtSrvr.cpp b/ndb/src/mgmsrv/MgmtSrvr.cpp
index f3e75f1afe4..793192dd487 100644
--- a/ndb/src/mgmsrv/MgmtSrvr.cpp
+++ b/ndb/src/mgmsrv/MgmtSrvr.cpp
@@ -1941,6 +1941,25 @@ MgmtSrvr::get_connected_nodes(NodeBitmask &connected_nodes) const
   }
 }
 
+void
+MgmtSrvr::get_connected_ndb_nodes(NodeBitmask &connected_nodes) const
+{
+  NodeBitmask ndb_nodes;
+  if (theFacade && theFacade->theClusterMgr)
+  {
+    for(Uint32 i = 0; i < MAX_NODES; i++)
+    {
+      if (getNodeType(i) == NDB_MGM_NODE_TYPE_NDB)
+      {
+        ndb_nodes.set(i);
+	const ClusterMgr::Node &node= theFacade->theClusterMgr->getNodeInfo(i);
+	connected_nodes.bitOR(node.m_state.m_connected_nodes);
+      }
+    }
+  }
+  connected_nodes.bitAND(ndb_nodes);
+}
+
 bool
 MgmtSrvr::alloc_node_id(NodeId * nodeId, 
 			enum ndb_mgm_node_type type,
diff --git a/ndb/src/mgmsrv/MgmtSrvr.hpp b/ndb/src/mgmsrv/MgmtSrvr.hpp
index 1e59f32b76f..5bacf640a18 100644
--- a/ndb/src/mgmsrv/MgmtSrvr.hpp
+++ b/ndb/src/mgmsrv/MgmtSrvr.hpp
@@ -485,6 +485,7 @@ public:
 
   const char *get_connect_address(Uint32 node_id);
   void get_connected_nodes(NodeBitmask &connected_nodes) const;
+  void get_connected_ndb_nodes(NodeBitmask &connected_nodes) const;
   SocketServer *get_socket_server() { return m_socket_server; }
 
   void updateStatus(NodeBitmask nodes);
diff --git a/ndb/src/mgmsrv/Services.cpp b/ndb/src/mgmsrv/Services.cpp
index 1df7d9be7b4..ae3433fe95b 100644
--- a/ndb/src/mgmsrv/Services.cpp
+++ b/ndb/src/mgmsrv/Services.cpp
@@ -952,7 +952,7 @@ printNodeStatus(OutputStream *output,
 		enum ndb_mgm_node_type type) {
   NodeId nodeId = 0;
   NodeBitmask hbnodes;
-  mgmsrv.get_connected_nodes(hbnodes);
+  mgmsrv.get_connected_ndb_nodes(hbnodes);
   mgmsrv.updateStatus(hbnodes);
   while(mgmsrv.getNextNodeId(&nodeId, type)) {
     enum ndb_mgm_node_status status;
diff --git a/ndb/src/ndbapi/ClusterMgr.cpp b/ndb/src/ndbapi/ClusterMgr.cpp
index 772cbaf99b5..28f65eebde8 100644
--- a/ndb/src/ndbapi/ClusterMgr.cpp
+++ b/ndb/src/ndbapi/ClusterMgr.cpp
@@ -70,6 +70,7 @@ ClusterMgr::ClusterMgr(TransporterFacade & _facade):
   ndbSetOwnVersion();
   clusterMgrThreadMutex = NdbMutex_Create();
   waitForHBCond= NdbCondition_Create();
+  waitingForHB= false;
   noOfAliveNodes= 0;
   noOfConnectedNodes= 0;
   theClusterMgrThread= 0;
@@ -172,7 +173,7 @@ ClusterMgr::forceHB(NodeBitmask waitFor)
 {
     theFacade.lock_mutex();
 
-    if(!waitForHBFromNodes.isclear())
+    if(waitingForHB)
     {
       NdbCondition_WaitTimeout(waitForHBCond, theFacade.theMutexPtr, 1000);
       theFacade.unlock_mutex();
@@ -180,6 +181,7 @@ ClusterMgr::forceHB(NodeBitmask waitFor)
     }
 
     global_flag_send_heartbeat_now= 1;
+    waitingForHB= true;
 
     waitForHBFromNodes= waitFor;
 #ifdef DEBUG_REG
@@ -209,10 +211,11 @@ ClusterMgr::forceHB(NodeBitmask waitFor)
     }
 
     NdbCondition_WaitTimeout(waitForHBCond, theFacade.theMutexPtr, 1000);
-    theFacade.unlock_mutex();
+    waitingForHB= false;
 #ifdef DEBUG_REG
     ndbout << "Still waiting for HB from " << waitForHBFromNodes.getText(buf) << endl;
 #endif
+    theFacade.unlock_mutex();
 }
 
 void
@@ -404,10 +407,16 @@ ClusterMgr::execAPI_REGCONF(const Uint32 * theData){
     node.hbFrequency = (apiRegConf->apiHeartbeatFrequency * 10) - 50;
   }
 
-  waitForHBFromNodes.clear(nodeId);
+  if(waitingForHB)
+  {
+    waitForHBFromNodes.clear(nodeId);
 
-  if(waitForHBFromNodes.isclear())
-    NdbCondition_Broadcast(waitForHBCond);
+    if(waitForHBFromNodes.isclear())
+    {
+      waitingForHB= false;
+      NdbCondition_Broadcast(waitForHBCond);
+    }
+  }
 }
 
 void
diff --git a/ndb/src/ndbapi/ClusterMgr.hpp b/ndb/src/ndbapi/ClusterMgr.hpp
index 3284fd8aa8a..b9863821b4f 100644
--- a/ndb/src/ndbapi/ClusterMgr.hpp
+++ b/ndb/src/ndbapi/ClusterMgr.hpp
@@ -90,6 +90,7 @@ private:
 
   NodeBitmask   waitForHBFromNodes; // used in forcing HBs
   NdbCondition* waitForHBCond;
+  bool          waitingForHB;
 
   /**
    * Used for controlling start/stop of the thread