[ray client] [runtime env] Print error logs in driver upon connection failure (#18451)

python/ray/tests/test_client_proxy.py

@@ -211,6 +211,30 @@ def test_startup_error_yields_clean_result(shutdown_only):
     server.stop(0)
 
 
+@pytest.mark.skipif(
+    sys.platform == "win32",
+    reason="PSUtil does not work the same on windows.")
+@pytest.mark.parametrize(
+    "call_ray_start", [
+        "ray start --head --ray-client-server-port 25031 "
+        "--port 0 --redis-password=password"
+    ],
+    indirect=True)
+def test_runtime_install_error_message(call_ray_start):
+    """
+    Check that an error while preparing the runtime environment for the client
+    server yields an actionable, clear error on the *client side*.
+    """
+    with pytest.raises(ConnectionAbortedError) as excinfo:
+        ray.client("localhost:25031").env({
+            "pip": ["ray-this-doesnt-exist"]
+        }).connect()
+    assert ("No matching distribution found for ray-this-doesnt-exist" in str(
+        excinfo.value))
+
+    ray.util.disconnect()
+
+
 def test_prepare_runtime_init_req_fails():
     """
     Check that a connection that is initiated with a non-Init request

python/ray/util/client/server/proxier.py

@@ -534,17 +534,31 @@ class DataServicerProxy(ray_client_pb2_grpc.RayletDataStreamerServicer):
                     logger.error(
                         f"Server startup failed for client: {client_id}, "
                         f"using JobConfig: {job_config}!")
+                    # TODO(architkulkarni): Once the client server runtime env
+                    # setup is moved into the runtime env agent, revisit this
+                    # and double check where the error logs end up being saved.
+                    try:
+                        with open("/tmp/ray/session_latest/logs/"
+                                  f"ray_client_server_{server.port}.err") as f:
+                            runtime_env_error_str = f.read()
+                    except FileNotFoundError:
+                        runtime_env_error_str = "(File not found)"
                     raise RuntimeError(
                         "Starting Ray client server failed. This is most "
                         "likely because the runtime_env failed to be "
-                        "installed. See ray_client_server_[port].err on the "
-                        "head node of the cluster for the relevant logs.")
+                        f"installed. Printing the contents of "
+                        f"ray_client_server_{server.port}.err below: \n"
+                        f"{runtime_env_error_str}")
                 channel = self.proxy_manager.get_channel(client_id)
                 if channel is None:
                     logger.error(f"Channel not found for {client_id}")
                     raise RuntimeError(
                         "Proxy failed to Connect to backend! Check "
-                        "`ray_client_server.err` on the cluster.")
+                        "`ray_client_server.err` and "
+                        f"`ray_client_server_{server.port}.err` on the head "
+                        "node of the cluster for the relevant logs. "
+                        "By default these are located at "
+                        "/tmp/ray/session_latest/logs.")
                 stub = ray_client_pb2_grpc.RayletDataStreamerStub(channel)
             except Exception:
                 init_resp = ray_client_pb2.DataResponse(