fixed race conditions

arangod/Cluster/ApplicationCluster.cpp

@@ -154,10 +154,15 @@ bool ApplicationCluster::prepare () {
       LOG_FATAL_AND_EXIT("invalid value specified for --cluster.my-id");
     }
   }
+  
+  // initialise ClusterComm library
+  ClusterComm::instance()->initialise();
 
   // initialise cluster info library
   ClusterInfo::instance();
 
+  usleep(1000);
+
   return true;
 }
 
@@ -258,9 +263,6 @@ bool ApplicationCluster::start () {
            _myAddress.c_str(),
            ServerState::roleToString(role).c_str());
   
-  // initialise ClusterComm library
-  ClusterComm::instance()->initialise();
-
   if (! _disableHeartbeat) {
     AgencyCommResult result = comm.getValues("Sync/HeartbeatIntervalMs", false);
 

arangod/Cluster/ClusterComm.cpp

@@ -445,7 +445,7 @@ ClusterCommResult* ClusterComm::wait (
   double timeleft;
 
   if (0.0 == timeout) {
-    endtime = 1.0e50;   // this is the Sankt Nimmerleinstag
+    endtime = TRI_microtime() + 24.0 * 60.0 * 60.0; // this is the Sankt Nimmerleinstag
   }
   else {
     endtime = TRI_microtime() + timeout;
@@ -895,6 +895,7 @@ void ClusterCommThread::run () {
       if (0 != _stop) {
         break;
       }
+
       {
         basics::ConditionLocker locker(&cc->somethingToSend);
         if (cc->toSend.empty()) {

js/server/modules/org/arangodb/cluster/kickstarter.js

@@ -285,6 +285,10 @@ launchActions.startServers = function (dispatchers, cmd, isRelaunch) {
     pids.push(executeExternal(arangodPath, args));
     endpoints.push(exchangePort(dispatchers[cmd.dispatcher].endpoint,port));
   }
+
+  console.info("Waiting for servers to come to live...");
+  wait(5);
+
   return {"error": false, "isStartServers": true, 
           "pids": pids, "endpoints": endpoints, "roles": roles};
 };