tests

python/ray/serve/BUILD

@@ -33,6 +33,14 @@ py_test(
     deps = [":serve_lib"],
 )
 
+py_test(
+    name = "test_get_deployment",
+    size = "medium",
+    srcs = serve_tests_srcs,
+    tags = ["exclusive"],
+    deps = [":serve_lib"],
+)
+
 py_test(
     name = "test_advanced",
     size = "medium",

python/ray/serve/__init__.py

@@ -2,7 +2,7 @@
     accept_batch, connect, start, get_replica_context, get_handle,
     shadow_traffic, set_traffic, delete_backend, list_backends, create_backend,
     get_backend_config, update_backend_config, list_endpoints, delete_endpoint,
-    create_endpoint, shutdown, ingress, deployment)
+    create_endpoint, shutdown, ingress, deployment, get_deployment)
 from ray.serve.batching import batch
 from ray.serve.config import BackendConfig, HTTPOptions
 from ray.serve.utils import ServeRequest
@@ -18,5 +18,5 @@
     "shadow_traffic", "set_traffic", "delete_backend", "list_backends",
     "create_backend", "get_backend_config", "update_backend_config",
     "list_endpoints", "delete_endpoint", "create_endpoint", "shutdown",
-    "ingress", "deployment"
+    "ingress", "deployment", "get_deployment"
 ]

python/ray/serve/api.py

@@ -1318,15 +1318,14 @@ def get_deployment(name: str):
         ServeDeployment
     """
     try:
-        backend_info, version, route = _get_global_client(
-        ).get_deployment_info(name)
+        backend_info, route = _get_global_client().get_deployment_info(name)
     except KeyError:
         raise KeyError(f"Deployment {name} was not found. "
                        "Did you call Deployment.deploy()?")
     return make_deployment_cls(
         backend_info.replica_config.backend_def,
         name,
         backend_info.backend_config,
-        version=version,
+        version=backend_info.version,
         init_args=backend_info.replica_config.init_args,
         ray_actor_options=backend_info.replica_config.ray_actor_options)

python/ray/serve/config.py

@@ -150,7 +150,7 @@ def __init__(self, backend_def, *init_args, ray_actor_options=None):
         self.is_blocking = _callable_is_blocking(backend_def)
         self.is_asgi_app = hasattr(backend_def, "_serve_asgi_app")
         self.path_prefix = getattr(backend_def, "_serve_path_prefix", None)
-        self.init_args = list(init_args)
+        self.init_args = init_args
         if ray_actor_options is None:
             self.ray_actor_options = {}
         else:

python/ray/serve/tests/test_get_deployment.py

@@ -0,0 +1,141 @@
+import os
+
+import pytest
+
+import ray
+from ray import serve
+
+
+def test_basic_get(serve_instance):
+    name = "test"
+
+    @serve.deployment(name, version="1")
+    def d(*args):
+        return "1", os.getpid()
+
+    with pytest.raises(KeyError):
+        serve.get_deployment(name)
+
+    d.deploy()
+    val1, pid1 = ray.get(d.get_handle().remote())
+    assert val1 == "1"
+
+    del d
+
+    d2 = serve.get_deployment(name)
+    val2, pid2 = ray.get(d2.get_handle().remote())
+    assert val2 == "1"
+    assert pid2 == pid1
+
+
+def test_get_after_delete(serve_instance):
+    name = "test"
+
+    @serve.deployment(name, version="1")
+    def d(*args):
+        return "1", os.getpid()
+
+    d.deploy()
+    del d
+
+    d2 = serve.get_deployment(name)
+    d2.delete()
+    del d2
+
+    with pytest.raises(KeyError):
+        serve.get_deployment(name)
+
+
+def test_deploy_new_version(serve_instance):
+    name = "test"
+
+    @serve.deployment(name, version="1")
+    def d(*args):
+        return "1", os.getpid()
+
+    d.deploy()
+    val1, pid1 = ray.get(d.get_handle().remote())
+    assert val1 == "1"
+
+    del d
+
+    d2 = serve.get_deployment(name)
+    d2.options(version="2").deploy()
+    val2, pid2 = ray.get(d2.get_handle().remote())
+    assert val2 == "1"
+    assert pid2 != pid1
+
+
+def test_deploy_empty_version(serve_instance):
+    name = "test"
+
+    @serve.deployment(name)
+    def d(*args):
+        return "1", os.getpid()
+
+    d.deploy()
+    val1, pid1 = ray.get(d.get_handle().remote())
+    assert val1 == "1"
+
+    del d
+
+    d2 = serve.get_deployment(name)
+    d2.deploy()
+    val2, pid2 = ray.get(d2.get_handle().remote())
+    assert val2 == "1"
+    assert pid2 != pid1
+
+
+def test_init_args(serve_instance):
+    name = "test"
+
+    @serve.deployment(name)
+    class D:
+        def __init__(self, val):
+            self._val = val
+
+        def __call__(self, *arg):
+            return self._val, os.getpid()
+
+    D.deploy("1")
+    val1, pid1 = ray.get(D.get_handle().remote())
+    assert val1 == "1"
+
+    del D
+
+    D2 = serve.get_deployment(name)
+    D2.deploy()
+    val2, pid2 = ray.get(D2.get_handle().remote())
+    assert val2 == "1"
+    assert pid2 != pid1
+
+    D2 = serve.get_deployment(name)
+    D2.deploy("2")
+    val3, pid3 = ray.get(D2.get_handle().remote())
+    assert val3 == "2"
+    assert pid3 != pid2
+
+
+def test_scale_replicas(serve_instance):
+    name = "test"
+
+    @serve.deployment(name)
+    def d(*args):
+        return os.getpid()
+
+    def check_num_replicas(num):
+        handle = serve.get_deployment(name).get_handle()
+        assert len(set(ray.get([handle.remote() for _ in range(50)]))) == num
+
+    d.deploy()
+    check_num_replicas(1)
+    del d
+
+    d2 = serve.get_deployment(name)
+    d2.options(num_replicas=2).deploy()
+    check_num_replicas(2)
+
+
+if __name__ == "__main__":
+    import sys
+    sys.exit(pytest.main(["-v", "-s", __file__]))