anyscale 0.24.78__tar.gz → 0.24.79__tar.gz

{anyscale-0.24.78 → anyscale-0.24.79}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: anyscale
-Version: 0.24.78
+Version: 0.24.79
 Summary: Command Line Interface for Anyscale
 Author: Anyscale Inc.
 License: AS License

{anyscale-0.24.78 → anyscale-0.24.79}/anyscale/_private/anyscale_client/anyscale_client.py

@@ -77,6 +77,8 @@ from anyscale.sdk.anyscale_client.models import (
     ClusterenvironmentbuildListResponse,
     ClusterEnvironmentBuildStatus,
     ClusterEnvironmentsQuery,
+    CreateBYODClusterEnvironment,
+    CreateBYODClusterEnvironmentConfigurationSchema,
     CreateClusterEnvironment,
     CreateClusterEnvironmentBuild,
     Job as APIJobRun,
@@ -759,31 +761,59 @@ class AnyscaleClient(AnyscaleClientInterface):
         )
 
     def _find_or_create_cluster_env(
-        self, cluster_env_name: str, anonymous: bool
+        self,
+        cluster_env_name: str,
+        anonymous: bool,
+        *,
+        image_uri: Optional[str] = None,
+        registry_login_secret: Optional[str] = None,
+        ray_version: Optional[str] = None,
     ) -> ClusterEnvironment:
+        """
+        Find or create a cluster environment with the given name.
+
+        There're two possible race conditions:
+        1) A tries to create a cluster env with the same name as B, but B has already created it.
+        -> A will get a 409 conflict error and should retry to get the existing cluster env.
+        2) A and B creates two identical builds under the same cluster env. This would cause job queue to
+        reject the job submission.
+        -> Cluster env and BYOD build are created within the same transaction, so the latter one will fail
+        with a 409 conflict error. The former one will succeed and the latter one should retry to get the
+        existing cluster env.
+        """
         existing_cluster_env = self.get_cluster_env_by_name(cluster_env_name)
         if existing_cluster_env is not None:
             return existing_cluster_env
 
-        # this creates a cluster env only and it does not trigger a build to avoid the race condition
-        # when creating a cluster environment and a build at the same time and then list builds for the cluster environment later.
-        # ```
-        #     cluster_env == self._external_api_client.create_cluster_environment(..., image_uri=image_uri)
-        #     build == self._external_api_client.create_cluster_environment_build(..., cluster_env_id=cluster_env.id)
-        # ```
-        # The race condition can happen if another build for the same cluster envrionment is created between the two calls.
         try:
-            cluster_environment = self._external_api_client.create_cluster_environment(
-                CreateClusterEnvironment(name=cluster_env_name, anonymous=anonymous)
-            ).result
+            if image_uri:
+                # For BYOD builds, we should create a build along with the cluster env.
+                cluster_environment = self._external_api_client.create_byod_cluster_environment(
+                    CreateBYODClusterEnvironment(
+                        name=cluster_env_name,
+                        config_json=CreateBYODClusterEnvironmentConfigurationSchema(
+                            docker_image=image_uri,
+                            ray_version=ray_version
+                            if ray_version
+                            else LATEST_RAY_VERSION,
+                            registry_login_secret=registry_login_secret,
+                        ),
+                        anonymous=anonymous,
+                    )
+                ).result
+            else:
+                cluster_environment = self._external_api_client.create_cluster_environment(
+                    CreateClusterEnvironment(name=cluster_env_name, anonymous=anonymous)
+                ).result
             return cluster_environment
         except ExternalApiException as e:
-            if e.status == 409:
-                # Retry to get the existing cluster env because it might be created by another process.
-                existing_cluster_env = self.get_cluster_env_by_name(cluster_env_name)
-                if existing_cluster_env is not None:
-                    return existing_cluster_env
-            raise e from None
+            if e.status != 409:
+                raise e from None
+            # Retry to get the existing cluster env because it might be created by another process.
+            existing_cluster_env = self.get_cluster_env_by_name(cluster_env_name)
+            if existing_cluster_env is None:
+                raise e from None
+            return existing_cluster_env
 
     @handle_api_exceptions
     def get_cluster_env_build_id_from_containerfile(
@@ -853,13 +883,30 @@ class AnyscaleClient(AnyscaleClientInterface):
                     raise RuntimeError(
                         f"Legacy cluster environment '{identifier}' is not found."
                     )
+        elif image_uri.is_default_image():
+            # Default image
+            cluster_envs = self._internal_api_client.list_application_templates_api_v2_application_templates_get(
+                image_name_contains=image_uri.image_uri
+            ).results
+            for cluster_env in cluster_envs:
+                if (
+                    cluster_env.latest_build is not None
+                    and cluster_env.latest_build.docker_image_name
+                    == image_uri.image_uri
+                ):
+                    return cluster_env.latest_build.id
+            raise RuntimeError(f"Default image '{image_uri.image_uri}' is not found.")
 
+        # BYOD image
         cluster_env_name = name if name else image_uri.to_cluster_env_name()
+        image_uri_str = str(image_uri)
         cluster_env = self._find_or_create_cluster_env(
-            cluster_env_name, anonymous=not name
+            cluster_env_name,
+            anonymous=not name,
+            image_uri=image_uri_str,
+            registry_login_secret=registry_login_secret,
+            ray_version=ray_version,
         )
-        image_uri_str = str(image_uri)
-
         for build in self.list_cluster_env_builds(cluster_env.id):
             if (
                 build.docker_image_name == image_uri_str

{anyscale-0.24.78 → anyscale-0.24.79}/anyscale/_private/models/image_uri.py

@@ -24,6 +24,7 @@ IMAGE_URI_PATTERN_RE = re.compile(
 )
 
 ANYSCALE_CLUSTER_ENV_PREFIX = "anyscale/image/"
+DEFAULT_IMAGE_URI_PREFIX = "anyscale/ray"
 
 
 class ImageURI:
@@ -68,7 +69,7 @@ class ImageURI:
             else:
                 return ImageURI.from_str("")
         elif cls._is_default_build(build):
-            return ImageURI.from_str(str(cluster_env.name))
+            return ImageURI.from_str(build.docker_image_name)
         else:
             image_uri_str = (
                 ANYSCALE_CLUSTER_ENV_PREFIX + f"{cluster_env.name}:{build.revision}"
@@ -91,6 +92,10 @@ class ImageURI:
         """Check if the image URI is a cluster environment image."""
         return self.image_uri.startswith(ANYSCALE_CLUSTER_ENV_PREFIX)
 
+    def is_default_image(self) -> bool:
+        """Check if the image URI is a default image."""
+        return self.image_uri.startswith(DEFAULT_IMAGE_URI_PREFIX)
+
     def to_cluster_env_identifier(self) -> str:
         assert self.is_cluster_env_image()
         identifier = self.image_uri.split("/")[2]

anyscale-0.24.79/anyscale/version.py

@@ -0,0 +1 @@
+__version__ = "0.24.79"

{anyscale-0.24.78 → anyscale-0.24.79}/anyscale.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: anyscale
-Version: 0.24.78
+Version: 0.24.79
 Summary: Command Line Interface for Anyscale
 Author: Anyscale Inc.
 License: AS License

{anyscale-0.24.78 → anyscale-0.24.79}/tests/unit/test_anyscale_client.py

@@ -35,12 +35,15 @@ from anyscale.client.openapi_client.models import (
     ComputeTemplateQuery,
     CreateComputeTemplate,
     CreateInternalProductionJob,
+    DecoratedApplicationTemplate,
+    DecoratedapplicationtemplateListResponse,
     DecoratedComputeTemplate,
     DecoratedcomputetemplateListResponse,
     FeatureFlagResponse,
     HaJobGoalStates,
     HaJobStates,
     InternalProductionJob,
+    MiniBuild,
     ProductionJobConfig,
     ProductionJobStateTransition,
     ProjectListResponse,
@@ -76,6 +79,7 @@ from anyscale.sdk.anyscale_client.models import (
     ClusterenvironmentListResponse,
     ComputeNodeType,
     ComputeTemplate,
+    CreateBYODClusterEnvironment,
     Job as APIJobRun,
     ListResponseMetadata,
     ProductionServiceV2VersionModel,
@@ -101,6 +105,7 @@ from anyscale.sdk.anyscale_client.models.create_cluster_environment_build import
 from anyscale.sdk.anyscale_client.models.job_list_response import JobListResponse
 from anyscale.sdk.anyscale_client.models.jobs_query import JobsQuery
 from anyscale.sdk.anyscale_client.rest import ApiException as ExternalApiException
+from anyscale.shared_anyscale_utils.latest_ray_version import LATEST_RAY_VERSION
 from anyscale.utils.workspace_notification import (
     WORKSPACE_NOTIFICATION_ADDRESS,
     WorkspaceNotification,
@@ -352,6 +357,26 @@ class FakeExternalAPIClient:
             )
         )
 
+    def create_byod_cluster_environment(
+        self, create_byod_cluster_environment: CreateBYODClusterEnvironment
+    ) -> ClusterEnvironment:
+        cluster_env = self.create_cluster_environment(
+            CreateClusterEnvironment(
+                name=create_byod_cluster_environment.name,
+                anonymous=create_byod_cluster_environment.anonymous,
+            )
+        ).result
+        self.create_cluster_environment_build(
+            CreateClusterEnvironmentBuild(
+                cluster_environment_id=cluster_env.id,
+                docker_image_name=create_byod_cluster_environment.config_json.docker_image,
+                ray_version=create_byod_cluster_environment.config_json.ray_version,
+                registry_login_secret=create_byod_cluster_environment.config_json.registry_login_secret,
+                local_vars_configuration=OPENAPI_NO_VALIDATION,
+            )
+        )
+        return FakeClientResult(result=cluster_env)
+
     def search_cluster_environments(
         self, query: ClusterEnvironmentsQuery
     ) -> ClusterenvironmentListResponse:
@@ -593,6 +618,8 @@ class FakeInternalAPIClient:
         self._job_run_id_to_log_chunk_urls: Dict[Optional[str], List[str]] = {}
         # Schedule ID to schedule.
         self._schedules: Dict[str, DecoratedSchedule] = {}
+        # Cluster environment ID to cluster envrionment.
+        self._cluster_envs: Dict[str, DecoratedApplicationTemplate] = {}
 
     def check_is_feature_flag_on_api_v2_userinfo_check_is_feature_flag_on_get(
         self, key: str
@@ -843,6 +870,26 @@ class FakeInternalAPIClient:
     def get_schedule(self, id: str) -> Optional[DecoratedSchedule]:  # noqa: A002
         return self._schedules[id]
 
+    def list_application_templates_api_v2_application_templates_get(
+        self, image_name_contains: str
+    ) -> DecoratedapplicationtemplateListResponse:
+        results: List[DecoratedApplicationTemplate] = []
+        for cluster_env in self._cluster_envs.values():
+            if (
+                cluster_env.latest_build
+                and image_name_contains in cluster_env.latest_build.docker_image_name
+            ):
+                results.append(cluster_env)
+        return DecoratedapplicationtemplateListResponse(
+            results=results,
+            metadata=ListResponseMetadata(total=len(results), next_paging_token=None),
+        )
+
+    def set_cluster_env(
+        self, cluster_env_id: str, cluster_env: DecoratedApplicationTemplate
+    ):
+        self._cluster_envs[cluster_env_id] = cluster_env
+
 
 class FakeAWSS3Client(AWSS3ClientInterface):
     def __init__(self):
@@ -2281,6 +2328,41 @@ class TestClusterEnv:
             image_uri, ray_version=None
         )
 
+    def test_get_cluster_env_build_id_from_image_uri_default_image(
+        self,
+        setup_anyscale_client: Tuple[
+            AnyscaleClient,
+            FakeExternalAPIClient,
+            FakeInternalAPIClient,
+            TestLogger,
+            FakeAWSS3Client,
+            FakeGCPGCSClient,
+        ],
+    ):
+        anyscale_client, _, fake_internal_client, _, _, _ = setup_anyscale_client
+        cluster_env_id = "fake-cluster-env-id"
+        default_image_uri = "anyscale/ray:2.22.0-slim-py39-cu121"
+        image_uri = ImageURI.from_str(default_image_uri)
+        build_id = "anyscaleray2220-slim-py39-cu121"
+
+        # Create successful cluster env build
+        fake_internal_client.set_cluster_env(
+            cluster_env_id,
+            DecoratedApplicationTemplate(
+                id=cluster_env_id,
+                latest_build=MiniBuild(
+                    id=build_id,
+                    docker_image_name=default_image_uri,
+                    local_vars_configuration=OPENAPI_NO_VALIDATION,
+                ),
+                local_vars_configuration=OPENAPI_NO_VALIDATION,
+            ),
+        )
+
+        assert build_id == anyscale_client.get_cluster_env_build_id_from_image_uri(
+            image_uri, ray_version=None
+        )
+
     def test_get_default_build_id(
         self,
         setup_anyscale_client: Tuple[
@@ -2392,6 +2474,16 @@ class TestClusterEnv:
             "fake-cluster-env-build-id", use_image_alias=True
         ) == ImageURI.from_str("anyscale/image/fake-cluster-env-name:5")
 
+    @pytest.mark.parametrize(
+        ("image_uri", "registry_login_secret", "ray_version"),
+        [
+            (None, None, None),
+            ("docker.io/byod:123", "secret", "2.30.0"),
+            ("docker.io/byod:123", None, "2.30.0"),
+            ("docker.io/byod:123", None, None),
+            ("docker.io/byod:123", "secret", None),
+        ],
+    )
     def test_find_or_create_cluster_env(
         self,
         setup_anyscale_client: Tuple[
@@ -2402,38 +2494,57 @@ class TestClusterEnv:
             FakeAWSS3Client,
             FakeGCPGCSClient,
         ],
+        image_uri: Optional[str],
+        registry_login_secret: Optional[str],
+        ray_version: Optional[str],
     ):
         anyscale_client, fake_external_client, _, _, _, _ = setup_anyscale_client
-        cluster_env_id = "fake-cluster-env-id"
         cluster_env_name = "fake-cluster-env-name"
-        fake_external_client.set_cluster_env(
-            cluster_env_id,
-            ClusterEnvironment(
-                id=cluster_env_id,
-                name=cluster_env_name,
-                local_vars_configuration=OPENAPI_NO_VALIDATION,
-            ),
-        )
+
+        # Test create a cluster env
         cluster_env = anyscale_client._find_or_create_cluster_env(
-            cluster_env_name, False
+            cluster_env_name,
+            False,
+            image_uri=image_uri,
+            registry_login_secret=registry_login_secret,
+            ray_version=ray_version,
         )
-        assert cluster_env.id == cluster_env_id
         assert cluster_env.name == cluster_env_name
+        cluster_env_id: str = cluster_env.id
+        if image_uri:
+            builds = fake_external_client.list_cluster_environment_builds(
+                cluster_environment_id=cluster_env_id, count=100
+            ).results
+            assert len(builds) == 1
+            build = builds[0]
+            assert build.docker_image_name == image_uri
+            assert build.registry_login_secret == registry_login_secret
+            assert build.ray_version == (
+                ray_version if ray_version else LATEST_RAY_VERSION
+            )
 
+        # Test get an existing cluster env
         cluster_env = anyscale_client._find_or_create_cluster_env(
-            cluster_env_name, False
+            cluster_env_name,
+            False,
+            image_uri=image_uri,
+            registry_login_secret=registry_login_secret,
+            ray_version=ray_version,
         )
         assert cluster_env.id == cluster_env_id
         assert cluster_env.name == cluster_env_name
 
+        # Test race condition where the cluster env is not found on the first call
         with patch.object(
             anyscale_client, "get_cluster_env_by_name", side_effect=[None, cluster_env]
         ) as mock_get_cluster_env_by_name:
             cluster_env = anyscale_client._find_or_create_cluster_env(
-                cluster_env_name, True
+                cluster_env_name,
+                False,
+                image_uri="docker.io/byod:123",
+                registry_login_secret="secret",
+                ray_version="2.30.0",
             )
-            assert cluster_env.id == cluster_env_id
-            assert cluster_env.name == cluster_env_name
             mock_get_cluster_env_by_name.assert_has_calls(
                 [call(cluster_env_name), call(cluster_env_name)]
             )

{anyscale-0.24.78 → anyscale-0.24.79}/tests/unit/test_image_uri.py

@@ -121,11 +121,11 @@ class TestImageURI:
                     config_json=Mock(),
                     containerfile=None,
                     cluster_environment_id="DEFAULT_APP_CONFIG_ID_2.8.1_py39",
+                    docker_image_name="anyscale/ray:2.8.1-py39",
                     revision=1,
                 ),
                 False,
-                # expect cluster_env.name is used. When it's a default build, cluster_env.name will be the default build id, e.g. "anyscale/ray:2.36.1-slim-py312-cu123".
-                "cluser_env_name",
+                "anyscale/ray:2.8.1-py39",
                 id="default_build",
             ),
         ],
@@ -145,3 +145,15 @@ class TestImageURI:
             ).image_uri
             == expected_image_uri
         )
+
+    @pytest.mark.parametrize(
+        ("image_uri", "expected"),
+        [
+            ("docker.io/library/ubuntu:latest", False),
+            ("anyscale/image/cluser_env_name:3", False),
+            ("anyscale/ray:2.22.0-slim-py39-cu121", True),
+            ("anyscale/ray-ml:2.22.0-cu121", True),
+        ],
+    )
+    def test_is_default_image(self, image_uri: str, expected: bool):
+        assert ImageURI.from_str(image_uri).is_default_image() == expected

anyscale-0.24.78/anyscale/version.py

@@ -1 +0,0 @@
-__version__ = "0.24.78"