ray.serve.config.DeploymentActorConfig#

pydantic model ray.serve.config.DeploymentActorConfig[source]#

Configuration for a deployment-scoped actor.

Deployment-scoped actors are long-lived actors managed by the Serve controller that outlive any single replica but are cleaned up when the deployment is deleted or serve.shutdown() is called. They are shared by all replicas of a deployment.

Example

from ray import serve
from ray.serve.config import DeploymentActorConfig

@ray.remote
class PrefixTreeActor:
    def __init__(self, max_depth: int = 100):
        self.max_depth = max_depth

    def insert(self, text: str):
        self.max_depth += 1

@serve.deployment(
    deployment_actors=[
        DeploymentActorConfig(
            name="prefix_tree",
            actor_class=PrefixTreeActor,
            init_kwargs={"max_depth": 100},
            actor_options={"num_cpus": 0.1},
        ),
    ],
)
class MyDeployment:
    def __init__(self):
        self.tree = serve.get_deployment_actor("prefix_tree")

    def __call__(self, request):
        ray.get(self.tree.insert.remote(request.text))

PublicAPI (alpha): This API is in alpha and may change before becoming stable.

field actor_class: type | str | ActorClass [Required]#: The actor class for this deployment-scoped actor. Can be a @ray.remote-decorated class or an import path string (e.g. ‘my_module:PrefixTreeActor’).

field actor_options: Dict[str, Any] [Optional]#: Ray actor options (e.g. num_cpus, num_gpus, runtime_env). Inherits deployment’s runtime_env; values here override for this actor.

field init_args: Tuple[Any, ...] [Optional]#: Positional arguments to pass to the actor constructor.

field init_kwargs: Dict[str, Any] [Optional]#: Keyword arguments to pass to the actor constructor.

field name: str [Required]#: Unique name for this deployment-scoped actor within the deployment.

get_actor_class() → type[source]#: Deserialize the actor class from cloudpickled bytes.