Revert "[Serve] Set controller and HTTP proxy num_cpus=0 by default (#15000)" (#15091)

This reverts commit 39aa01fc2c.

python/ray/serve/api.py

@@ -651,7 +651,6 @@ def start(
         http_port: int = DEFAULT_HTTP_PORT,
         http_middlewares: List[Any] = [],
         http_options: Optional[Union[dict, HTTPOptions]] = None,
-        dedicated_cpu: bool = False,
 ) -> Client:
     """Initialize a serve instance.
 
@@ -676,7 +675,7 @@ def start(
               this to "0.0.0.0".
             - port(int): Port for HTTP server. Defaults to 8000.
             - middlewares(list): A list of Starlette middlewares that will be
-              applied to the HTTP servers in the cluster. Defaults to [].
+              applied to the HTTP servers in the cluster.
             - location(str, serve.config.DeploymentMode): The deployment
               location of HTTP servers:
 
@@ -685,10 +684,6 @@ def start(
                   on. This is the default.
                 - "EveryNode": start one HTTP server per node.
                 - "NoServer" or None: disable HTTP server.
-            - num_cpus (int): The number of CPU cores to reserve for each
-              internal Serve HTTP proxy actor.  Defaults to 0.
-        dedicated_cpu (bool): Whether to reserve a CPU core for the internal
-          Serve controller actor.  Defaults to False.
     """
     if ((http_host != DEFAULT_HTTP_HOST) or (http_port != DEFAULT_HTTP_PORT)
             or (len(http_middlewares) != 0)):
@@ -733,7 +728,6 @@ def start(
             host=http_host, port=http_port, middlewares=http_middlewares)
 
     controller = ServeController.options(
-        num_cpus=(1 if dedicated_cpu else 0),
         name=controller_name,
         lifetime="detached" if detached else None,
         max_restarts=-1,

python/ray/serve/config.py

@@ -244,7 +244,6 @@ class HTTPOptions(pydantic.BaseModel):
     port: int = DEFAULT_HTTP_PORT
     middlewares: List[Any] = []
     location: Optional[DeploymentMode] = DeploymentMode.HeadOnly
-    num_cpus: int = 0
 
     @validator("location", always=True)
     def location_backfill_no_server(cls, v, values):

python/ray/serve/controller.py

@@ -38,7 +38,7 @@ CHECKPOINT_KEY = "serve-controller-checkpoint"
 CONTROL_LOOP_PERIOD_S = 0.1
 
 
-@ray.remote(num_cpus=0)
+@ray.remote
 class ServeController:
     """Responsible for managing the state of the serving system.
 

python/ray/serve/examples/doc/quickstart_class.py

@@ -2,7 +2,7 @@ import ray
 from ray import serve
 import requests
 
-ray.init()
+ray.init(num_cpus=4)
 serve.start()
 
 

python/ray/serve/examples/doc/quickstart_function.py

@@ -2,7 +2,7 @@ import ray
 from ray import serve
 import requests
 
-ray.init()
+ray.init(num_cpus=4)
 serve.start()
 
 

python/ray/serve/http_proxy.py

@@ -164,7 +164,7 @@ class HTTPProxy:
         await self.router(scope, receive, send)
 
 
-@ray.remote(num_cpus=0)
+@ray.remote
 class HTTPProxyActor:
     def __init__(self,
                  host: str,

python/ray/serve/http_state.py

@@ -69,7 +69,6 @@ class HTTPState:
                                 name, node_id, self._config.host,
                                 self._config.port))
                 proxy = HTTPProxyActor.options(
-                    num_cpus=self._config.num_cpus,
                     name=name,
                     lifetime="detached" if self._detached else None,
                     max_concurrency=ASYNC_CONCURRENCY,

python/ray/serve/tests/test_standalone.py

@@ -15,7 +15,6 @@ from ray.serve.constants import SERVE_PROXY_NAME
 from ray.serve.exceptions import RayServeException
 from ray.serve.utils import (block_until_http_ready, get_all_node_ids,
                              format_actor_name)
-from ray.serve.config import HTTPOptions
 from ray.test_utils import wait_for_condition
 from ray._private.services import new_port
 
@@ -122,30 +121,6 @@ def test_connect(detached, ray_shutdown):
     assert "backend-ception" in serve.list_backends().keys()
 
 
-@pytest.mark.skipif(sys.platform == "win32", reason="Failing on Windows")
-@pytest.mark.parametrize("controller_cpu", [True, False])
-@pytest.mark.parametrize("num_proxy_cpus", [0, 1, 2])
-def test_dedicated_cpu(controller_cpu, num_proxy_cpus, ray_cluster):
-    cluster = ray_cluster
-    num_cluster_cpus = 8
-    head_node = cluster.add_node(num_cpus=num_cluster_cpus)
-
-    ray.init(head_node.address)
-    wait_for_condition(
-        lambda: ray.cluster_resources().get("CPU") == num_cluster_cpus)
-
-    num_cpus_used = int(controller_cpu) + num_proxy_cpus
-
-    serve.start(
-        dedicated_cpu=controller_cpu,
-        http_options=HTTPOptions(num_cpus=num_proxy_cpus))
-    available_cpus = num_cluster_cpus - num_cpus_used
-    wait_for_condition(
-        lambda: (ray.available_resources().get("CPU") == available_cpus))
-    serve.shutdown()
-    ray.shutdown()
-
-
 @pytest.mark.skipif(
     not hasattr(socket, "SO_REUSEPORT"),
     reason=("Port sharing only works on newer verion of Linux. "
@@ -329,7 +304,7 @@ def test_http_head_only(ray_cluster):
         r["CPU"]
         for r in ray.state.state._available_resources_per_node().values()
     }
-    assert cpu_per_nodes == {4, 4}
+    assert cpu_per_nodes == {2, 4}
 
 
 if __name__ == "__main__":