Add ability to specify worker and driver ports (#8071)

python/ray/tests/conftest.py

@@ -153,7 +153,10 @@ def ray_start_object_store_memory(request):
 
 @pytest.fixture
 def call_ray_start(request):
-    parameter = getattr(request, "param", "ray start --head --num-cpus=1")
+    parameter = getattr(
+        request, "param",
+        "ray start --head --num-cpus=1 --min-worker-port=0 --max-worker-port=0"
+    )
     command_args = parameter.split(" ")
     out = ray.utils.decode(
         subprocess.check_output(command_args, stderr=subprocess.STDOUT))

python/ray/tests/test_dynres.py

@@ -43,8 +43,10 @@ def test_dynamic_res_deletion(shutdown_only):
     available_res = ray.available_resources()
     cluster_res = ray.cluster_resources()
 
-    assert res_name not in available_res
-    assert res_name not in cluster_res
+    def check_resources():
+        return res_name not in available_res and res_name not in cluster_res
+
+    ray.test_utils.wait_for_condition(check_resources)
 
 
 def test_dynamic_res_infeasible_rescheduling(ray_start_regular):

python/ray/tests/test_multi_node.py

@@ -285,7 +285,10 @@ ray.get([a.log.remote(), f.remote()])
 
 
 @pytest.mark.parametrize(
-    "call_ray_start", ["ray start --head --num-cpus=1 --num-gpus=1"],
+    "call_ray_start", [
+        "ray start --head --num-cpus=1 --num-gpus=1 " +
+        "--min-worker-port=0 --max-worker-port=0"
+    ],
     indirect=True)
 def test_drivers_release_resources(call_ray_start):
     address = call_ray_start
@@ -334,6 +337,7 @@ print("success")
             print(output_line)
             if output_line == "success":
                 return
+            time.sleep(1)
         raise RayTestTimeoutException(
             "Timed out waiting for process to print success.")
 
@@ -376,6 +380,13 @@ def test_calling_start_ray_head(call_ray_stop_only):
     ])
     subprocess.check_output(["ray", "stop"])
 
+    # Test starting Ray with the worker port range specified.
+    subprocess.check_output([
+        "ray", "start", "--head", "--min-worker-port", "50000",
+        "--max-worker-port", "51000"
+    ])
+    subprocess.check_output(["ray", "stop"])
+
     # Test starting Ray with the number of CPUs specified.
     subprocess.check_output(["ray", "start", "--head", "--num-cpus", "2"])
     subprocess.check_output(["ray", "stop"])
@@ -419,7 +430,7 @@ def test_calling_start_ray_head(call_ray_stop_only):
     assert blocked.returncode is None
 
     kill_process_by_name("raylet")
-    wait_for_children_of_pid_to_exit(blocked.pid, timeout=120)
+    wait_for_children_of_pid_to_exit(blocked.pid, timeout=30)
     blocked.wait()
     assert blocked.returncode != 0, "ray start shouldn't return 0 on bad exit"
 
@@ -431,7 +442,7 @@ def test_calling_start_ray_head(call_ray_stop_only):
     wait_for_children_of_pid(blocked.pid, num_children=7, timeout=30)
 
     blocked.terminate()
-    wait_for_children_of_pid_to_exit(blocked.pid, timeout=120)
+    wait_for_children_of_pid_to_exit(blocked.pid, timeout=30)
     blocked.wait()
     assert blocked.returncode != 0, "ray start shouldn't return 0 on bad exit"