Fix actor creation hang due to race in SWAP queue (#6280)

python/ray/cluster_utils.py

@@ -92,6 +92,8 @@ class Cluster(object):
             self.webui_url = self.head_node.webui_url
         else:
             ray_params.update_if_absent(redis_address=self.redis_address)
+            # We only need one log monitor per physical node.
+            ray_params.update_if_absent(include_log_monitor=False)
             # Let grpc pick a port.
             ray_params.update(node_manager_port=0)
             node = ray.node.Node(

python/ray/tests/BUILD

@@ -10,7 +10,7 @@ py_test(
     name = "test_actor_direct",
     size = "medium",
     srcs = ["test_actor_direct.py", "test_actor.py"],
-    tags = ["exclusive", "manual"],
+    tags = ["exclusive"],
     deps = ["//:ray_lib"],
 )
 

python/ray/tests/test_multi_node_2.py

@@ -215,8 +215,6 @@ def test_worker_plasma_store_failure(ray_start_cluster_head):
     cluster = ray_start_cluster_head
     worker = cluster.add_node()
     cluster.wait_for_nodes()
-    # Log monitor doesn't die for some reason
-    worker.kill_log_monitor()
     worker.kill_reporter()
     worker.kill_plasma_store()
     worker.all_processes[ray_constants.PROCESS_TYPE_RAYLET][0].process.wait()