Fix actor garbage collection by breaking cyclic references (#1064)

* Fix bug in wait_for_pid_to_exit, add test for actor deletion.

* Fix actor garbage collection by breaking cyclic references

* Add test for calling actor method immediately after actor creation.

* Fix bug, must dispatch tasks when workers are killed.

* Fix python test

* Fix cyclic reference problem by creating ActorMethod objects on the fly.

* Try simply increasing the time allowed for many_drivers_test.py.

test/jenkins_tests/multi_node_tests/many_drivers_test.py

@@ -30,10 +30,10 @@ class Actor1(object):
 
 
 def driver(redis_address, driver_index):
-    """The script for driver 0.
+    """The script for all drivers.
 
-    This driver should create five actors that each use one GPU and some actors
-    that use no GPUs. After a while, it should exit.
+    This driver should create five actors that each use one GPU. After a while,
+    it should exit.
     """
     ray.init(redis_address=redis_address)
 
@@ -44,7 +44,7 @@ def driver(redis_address, driver_index):
     for i in range(driver_index - max_concurrent_drivers + 1):
         _wait_for_event("DRIVER_{}_DONE".format(i), redis_address)
 
-    def try_to_create_actor(actor_class, timeout=100):
+    def try_to_create_actor(actor_class, timeout=500):
         # Try to create an actor, but allow failures while we wait for the
         # monitor to release the resources for the removed drivers.
         start_time = time.time()