fixed startup race condition

arangod/Cluster/ApplicationCluster.cpp

@@ -155,14 +155,12 @@ bool ApplicationCluster::prepare () {
     }
   }
   
+  // initialise ClusterInfo library
+  ClusterInfo::initialise(); 
+
   // initialise ClusterComm library
   ClusterComm::initialise();
 
-  // initialise cluster info library
-  ClusterInfo::instance();
-
-  usleep(1000);
-
   return true;
 }
 

arangod/Cluster/ClusterComm.cpp

@@ -59,14 +59,8 @@ void triagens::arango::ClusterCommRestCallback(string& coordinator,
 /// @brief ClusterComm constructor
 ////////////////////////////////////////////////////////////////////////////////
 
-ClusterComm::ClusterComm () {
+ClusterComm::ClusterComm () :
-  _backgroundThread = new ClusterCommThread();
+  _backgroundThread(0) {
-  if (0 == _backgroundThread) {
-    LOG_FATAL_AND_EXIT("unable to start ClusterComm background thread");
-  }
-  if (! _backgroundThread->init() || ! _backgroundThread->start()) {
-    LOG_FATAL_AND_EXIT("ClusterComm background thread does not work");
-  }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -74,10 +68,13 @@ ClusterComm::ClusterComm () {
 ////////////////////////////////////////////////////////////////////////////////
 
 ClusterComm::~ClusterComm () {
-  _backgroundThread->stop();
+  if (_backgroundThread != 0) {
-  _backgroundThread->shutdown();
+    _backgroundThread->stop();
-  delete _backgroundThread;
+    _backgroundThread->shutdown();
-  _backgroundThread = 0;
+    delete _backgroundThread;
+    _backgroundThread = 0;
+  }
+
   cleanupAllQueues();
 }
 
@@ -100,12 +97,30 @@ ClusterComm* ClusterComm::instance () {
 }
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief only used to trigger creation
+/// @brief initialise the cluster comm singleton object
 ////////////////////////////////////////////////////////////////////////////////
 
 void ClusterComm::initialise () {
   assert(_theinstance == 0);
-  _theinstance = new ClusterComm( );  // this now happens exactly once
+  _theinstance = new ClusterComm();  // this now happens exactly once
+ 
+  _theinstance->startBackgroundThread(); 
+}
+
+////////////////////////////////////////////////////////////////////////////////
+/// @brief start the communication background thread
+////////////////////////////////////////////////////////////////////////////////
+
+void ClusterComm::startBackgroundThread () {
+  _backgroundThread = new ClusterCommThread();
+
+  if (0 == _backgroundThread) {
+    LOG_FATAL_AND_EXIT("unable to start ClusterComm background thread");
+  }
+
+  if (! _backgroundThread->init() || ! _backgroundThread->start()) {
+    LOG_FATAL_AND_EXIT("ClusterComm background thread does not work");
+  }
 }
 
 ////////////////////////////////////////////////////////////////////////////////

arangod/Cluster/ClusterComm.h

@@ -261,6 +261,12 @@ void ClusterCommRestCallback(string& coordinator, rest::HttpResponse* response);
           _theinstance = 0;
         }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief start the communication background thread
+////////////////////////////////////////////////////////////////////////////////
+
+        void startBackgroundThread ();
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief submit an HTTP request to a shard asynchronously.
 ////////////////////////////////////////////////////////////////////////////////

arangod/Cluster/ClusterInfo.cpp

@@ -244,12 +244,19 @@ ClusterInfo* ClusterInfo::instance () {
   // This does not have to be thread-safe, because we guarantee that
   // this is called very early in the startup phase when there is still
   // a single thread.
-  if (0 == _theinstance) {
+  assert(_theinstance != 0);
-    _theinstance = new ClusterInfo();  // this now happens exactly once
-  }
   return _theinstance;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief initialise the cluster info singleton object
+////////////////////////////////////////////////////////////////////////////////
+
+void ClusterInfo::initialise () {
+  assert(_theinstance == 0);
+  _theinstance = new ClusterInfo();  // this now happens exactly once
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief creates a cluster info object
 ////////////////////////////////////////////////////////////////////////////////
@@ -2085,7 +2092,6 @@ int ClusterInfo::getResponsibleShard (CollectionID const& collectionID,
   return error;
 }
 
-
 // Local Variables:
 // mode: outline-minor
 // outline-regexp: "^\\(/// @brief\\|/// {@inheritDoc}\\|/// @addtogroup\\|// --SECTION--\\|/// @\\}\\)"

arangod/Cluster/ClusterInfo.h

@@ -720,6 +720,12 @@ namespace triagens {
 
         static ClusterInfo* instance ();
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief initialise function to call once when still single-threaded
+////////////////////////////////////////////////////////////////////////////////
+        
+        static void initialise ();
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief cleanup function to call once when shutting down
 ////////////////////////////////////////////////////////////////////////////////