[serve] Add list_backends and list_endpoints (#8737)

python/ray/serve/__init__.py

@@ -1,10 +1,20 @@
-from ray.serve.api import (init, create_backend, delete_backend,
-                           create_endpoint, delete_endpoint, set_traffic,
-                           get_handle, stat, update_backend_config,
-                           get_backend_config, accept_batch)  # noqa: E402
+from ray.serve.api import (
+    init, create_backend, delete_backend, create_endpoint, delete_endpoint,
+    set_traffic, get_handle, stat, update_backend_config, get_backend_config,
+    accept_batch, list_backends, list_endpoints)  # noqa: E402
 
 __all__ = [
-    "init", "create_backend", "delete_backend", "create_endpoint",
-    "delete_endpoint", "set_traffic", "get_handle", "stat",
-    "update_backend_config", "get_backend_config", "accept_batch"
+    "init",
+    "create_backend",
+    "delete_backend",
+    "create_endpoint",
+    "delete_endpoint",
+    "set_traffic",
+    "get_handle",
+    "stat",
+    "update_backend_config",
+    "get_backend_config",
+    "accept_batch",
+    "list_backends",
+    "list_endpoints",
 ]

python/ray/serve/api.py

@@ -141,6 +141,16 @@ def delete_endpoint(endpoint):
     retry_actor_failures(master_actor.delete_endpoint, endpoint)
 
 
+@_ensure_connected
+def list_endpoints():
+    """Returns a dictionary of all registered endpoints.
+
+    The dictionary keys are endpoint names and values are dictionaries
+    of the form: {"methods": List[str], "traffic": Dict[str, float]}.
+    """
+    return retry_actor_failures(master_actor.get_all_endpoints)
+
+
 @_ensure_connected
 def update_backend_config(backend_tag, config_options):
     """Update a backend configuration for a backend tag.
@@ -200,6 +210,15 @@ def create_backend(backend_tag,
                          backend_config, replica_config)
 
 
+@_ensure_connected
+def list_backends():
+    """Returns a dictionary of all registered backends.
+
+    Dictionary maps backend tags to backend configs.
+    """
+    return retry_actor_failures(master_actor.get_all_backends)
+
+
 @_ensure_connected
 def delete_backend(backend_tag):
     """Delete the given backend.

python/ray/serve/master.py

@@ -446,12 +446,22 @@ class ServeMaster:
         return self.workers
 
     def get_all_backends(self):
-        """Used for validation by the API client."""
-        return list(self.backends.keys())
+        """Returns a dictionary of backend tag to backend config dict."""
+        backends = {}
+        for backend_tag, (_, config, _) in self.backends.items():
+            backends[backend_tag] = config.__dict__
+        return backends
 
     def get_all_endpoints(self):
-        """Used for validation by the API client."""
-        return [endpoint for endpoint, methods in self.routes.values()]
+        """Returns a dictionary of endpoint to endpoint config."""
+        endpoints = {}
+        for route, (endpoint, methods) in self.routes.items():
+            endpoints[endpoint] = {
+                "route": route if route.startswith("/") else None,
+                "methods": methods,
+                "traffic": self.traffic_policies.get(endpoint, {})
+            }
+        return endpoints
 
     async def set_traffic(self, endpoint_name, traffic_policy_dictionary):
         """Sets the traffic policy for the specified endpoint."""

python/ray/serve/tests/test_api.py

@@ -426,3 +426,65 @@ def test_parallel_start(serve_instance):
     handle = serve.get_handle("test-parallel")
 
     ray.get(handle.remote(), timeout=10)
+
+
+def test_list_endpoints(serve_instance):
+    serve.init()
+
+    def f():
+        pass
+
+    serve.create_endpoint("endpoint", "/api", methods=["GET", "POST"])
+    serve.create_endpoint("endpoint2", methods=["POST"])
+    serve.create_backend("backend", f)
+    serve.set_traffic("endpoint2", {"backend": 1.0})
+
+    endpoints = serve.list_endpoints()
+    assert "endpoint" in endpoints
+    assert endpoints["endpoint"] == {
+        "route": "/api",
+        "methods": ["GET", "POST"],
+        "traffic": {}
+    }
+
+    assert "endpoint2" in endpoints
+    assert endpoints["endpoint2"] == {
+        "route": None,
+        "methods": ["POST"],
+        "traffic": {
+            "backend": 1.0
+        }
+    }
+
+    serve.delete_endpoint("endpoint")
+    assert "endpoint2" in serve.list_endpoints()
+
+    serve.delete_endpoint("endpoint2")
+    assert len(serve.list_endpoints()) == 0
+
+
+def test_list_backends(serve_instance):
+    serve.init()
+
+    @serve.accept_batch
+    def f():
+        pass
+
+    serve.create_backend("backend", f, config={"max_batch_size": 10})
+    backends = serve.list_backends()
+    assert len(backends) == 1
+    assert "backend" in backends
+    assert backends["backend"]["max_batch_size"] == 10
+
+    serve.create_backend("backend2", f, config={"num_replicas": 10})
+    backends = serve.list_backends()
+    assert len(backends) == 2
+    assert backends["backend2"]["num_replicas"] == 10
+
+    serve.delete_backend("backend")
+    backends = serve.list_backends()
+    assert len(backends) == 1
+    assert "backend2" in backends
+
+    serve.delete_backend("backend2")
+    assert len(serve.list_backends()) == 0