[rllib] Add tests for examples using ray client (#16271)

* [rllib] add tests for examples using ray client

* rename test_client to test_ray_client

rllib/BUILD

@@ -1550,6 +1550,13 @@ py_test(
     srcs = ["tests/test_placement_groups.py"]
 )
 
+py_test(
+    name = "tests/test_ray_client",
+    tags = ["tests_dir", "tests_dir_R"],
+    size = "large",
+    srcs = ["tests/test_ray_client.py"]
+)
+
 py_test(
     name = "tests/test_reproducibility",
     tags = ["tests_dir", "tests_dir_R"],

rllib/tests/test_ray_client.py

@@ -0,0 +1,92 @@
+import os
+import sys
+import unittest
+
+import pytest
+import ray
+from ray import tune
+from ray.job_config import JobConfig
+from ray.rllib.agents import ppo
+from ray.rllib.examples.env.stateless_cartpole import StatelessCartPole
+from ray.rllib.utils.test_utils import check_learning_achieved
+from ray.util.client.ray_client_helpers import ray_start_client_server
+
+
+class TestRayClient(unittest.TestCase):
+    def test_connection(self):
+        with ray_start_client_server():
+            assert ray.util.client.ray.is_connected()
+        assert ray.util.client.ray.is_connected() is False
+
+    def test_custom_train_fn(self):
+        with ray_start_client_server():
+            assert ray.util.client.ray.is_connected()
+
+            config = {
+                "lr": 0.01,
+                # Use GPUs iff `RLLIB_NUM_GPUS` env var set to > 0.
+                "num_gpus": int(os.environ.get("RLLIB_NUM_GPUS", "0")),
+                "num_workers": 0,
+                "framework": "tf",
+            }
+            resources = ppo.PPOTrainer.default_resource_request(config)
+            from ray.rllib.examples.custom_train_fn import my_train_fn
+            tune.run(my_train_fn, resources_per_trial=resources, config=config)
+
+    def test_cartpole_lstm(self):
+        with ray_start_client_server():
+            assert ray.util.client.ray.is_connected()
+
+            config = dict(
+                {
+                    "num_sgd_iter": 5,
+                    "model": {
+                        "vf_share_layers": True,
+                    },
+                    "vf_loss_coeff": 0.0001,
+                },
+                **{
+                    "env": StatelessCartPole,
+                    # Use GPUs iff `RLLIB_NUM_GPUS` env var set to > 0.
+                    "num_gpus": int(os.environ.get("RLLIB_NUM_GPUS", "0")),
+                    "model": {
+                        "use_lstm": True,
+                        "lstm_cell_size": 256,
+                        "lstm_use_prev_action": None,
+                        "lstm_use_prev_reward": None,
+                    },
+                    "framework": "tf",
+                    # Run with tracing enabled for tfe/tf2?
+                    "eager_tracing": None,
+                })
+
+            stop = {
+                "training_iteration": 200,
+                "timesteps_total": 100000,
+                "episode_reward_mean": 150.0,
+            }
+
+            results = tune.run("PPO", config=config, stop=stop, verbose=2)
+            check_learning_achieved(results, 150.0)
+
+    def test_custom_experiment(self):
+        def ray_connect_handler(job_config: JobConfig = None):
+            ray.init(num_cpus=3)
+
+        with ray_start_client_server(ray_connect_handler=ray_connect_handler):
+            assert ray.util.client.ray.is_connected()
+
+            config = ppo.DEFAULT_CONFIG.copy()
+            config["train-iterations"] = 10
+            config["env"] = "CartPole-v0"
+
+            from ray.rllib.examples.custom_experiment import experiment
+            tune.run(
+                experiment,
+                config=config,
+                resources_per_trial=ppo.PPOTrainer.default_resource_request(
+                    config))
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main(["-v", __file__]))