Executors connect to wrong port when collision occurs

This commit fixes this by setting "spark.driver.port" to the actual
port the akka system bound to. Note that we already do this for port
0, where the original port is not the same as the bound port. However,
we still face the same issue if we end up using port n + 1.

core/src/main/scala/org/apache/spark/SparkEnv.scala

@@ -150,10 +150,10 @@ object SparkEnv extends Logging {
     val (actorSystem, boundPort) = AkkaUtils.createActorSystem("spark", hostname, port, conf = conf,
       securityManager = securityManager)
 
-    // Bit of a hack: If this is the driver and our port was 0 (meaning bind to any free port),
-    // figure out which port number Akka actually bound to and set spark.driver.port to it.
-    if (isDriver && port == 0) {
-      conf.set("spark.driver.port",  boundPort.toString)
+    // Figure out which port Akka actually bound to in case the original port is 0 or occupied.
+    // This is so that we tell the executors the correct port to connect to.
+    if (isDriver) {
+      conf.set("spark.driver.port", boundPort.toString)
     }
 
     // Create an instance of the class named by the given Java system property, or by