Merge branch 'main' into pre-commit-ci-update-config

opendatahub-io · Jan 7, 2025 · 445d05e · 445d05e
2 parents be5c702 + 453f931
commit 445d05e
Show file tree

Hide file tree

Showing 4 changed files with 51 additions and 102 deletions.
diff --git a/tests/model_serving/model_server/private_endpoint/conftest.py b/tests/model_serving/model_server/private_endpoint/conftest.py
@@ -1,104 +1,53 @@
+from typing import Any, Generator
+
 import pytest
-from typing import Generator, Any
+from kubernetes.dynamic import DynamicClient
 from ocp_resources.inference_service import InferenceService
-from ocp_resources.secret import Secret
 from ocp_resources.namespace import Namespace
 from ocp_resources.pod import Pod
-from simple_logger.logger import get_logger
+from ocp_resources.secret import Secret
 from ocp_resources.serving_runtime import ServingRuntime
-from kubernetes.dynamic import DynamicClient
+from simple_logger.logger import get_logger
 
-from utilities.serving_runtime import ServingRuntimeFromTemplate
+from tests.model_serving.model_server.private_endpoint.utils import create_sidecar_pod
 from tests.model_serving.model_server.utils import create_isvc
-from tests.model_serving.model_server.private_endpoint.utils import (
-    create_sidecar_pod,
-)
-from utilities.infra import (
-    create_ns,
-    create_storage_config_secret,
-    s3_endpoint_secret,
-    wait_for_kserve_predictor_deployment_replicas,
-)
-from utilities.constants import KServeDeploymentType, ModelStoragePath, ModelFormat
-
+from utilities.constants import KServeDeploymentType, ModelFormat, ModelStoragePath
+from utilities.infra import create_ns, create_storage_config_secret
 
 LOGGER = get_logger(name=__name__)
 
 
-@pytest.fixture(scope="class")
-def endpoint_namespace(admin_client: DynamicClient) -> Generator[Namespace, None, None]:
-    with create_ns(admin_client=admin_client, name="endpoint-namespace") as ns:
-        yield ns
-
-
 @pytest.fixture(scope="class")
 def diff_namespace(admin_client: DynamicClient) -> Generator[Namespace, None, None]:
     with create_ns(admin_client=admin_client, name="diff-namespace") as ns:
         yield ns
 
 
-@pytest.fixture(scope="class")
-def endpoint_sr(
-    admin_client: DynamicClient,
-    endpoint_namespace: Namespace,
-) -> Generator[ServingRuntime, None, None]:
-    with ServingRuntimeFromTemplate(
-        client=admin_client,
-        name="flan-example-sr",
-        namespace=endpoint_namespace.name,
-        template_name="caikit-tgis-serving-template",
-    ) as model_runtime:
-        yield model_runtime
-
-
-@pytest.fixture(scope="class")
-def endpoint_s3_secret(
-    admin_client: DynamicClient,
-    endpoint_namespace: Namespace,
-    aws_access_key_id: str,
-    aws_secret_access_key: str,
-    models_s3_bucket_name: str,
-    models_s3_bucket_region: str,
-    models_s3_bucket_endpoint: str,
-) -> Generator[Secret, None, None]:
-    with s3_endpoint_secret(
-        admin_client=admin_client,
-        name="endpoint-s3-secret",
-        namespace=endpoint_namespace.name,
-        aws_access_key=aws_access_key_id,
-        aws_secret_access_key=aws_secret_access_key,
-        aws_s3_region=models_s3_bucket_region,
-        aws_s3_bucket=models_s3_bucket_name,
-        aws_s3_endpoint=models_s3_bucket_endpoint,
-    ) as secret:
-        yield secret
-
-
 @pytest.fixture(scope="class")
 def endpoint_isvc(
     admin_client: DynamicClient,
-    endpoint_sr: ServingRuntime,
-    endpoint_s3_secret: Secret,
+    serving_runtime_from_template: ServingRuntime,
+    models_endpoint_s3_secret: Secret,
     storage_config_secret: Secret,
-    endpoint_namespace: Namespace,
 ) -> Generator[InferenceService, None, None]:
     with create_isvc(
         client=admin_client,
-        name="test",
-        namespace=endpoint_namespace.name,
+        name="endpoint-isvc",
+        namespace=serving_runtime_from_template.namespace,
         deployment_mode=KServeDeploymentType.SERVERLESS,
-        storage_key="endpoint-s3-secret",
+        storage_key=models_endpoint_s3_secret.name,
         storage_path=ModelStoragePath.FLAN_T5_SMALL,
         model_format=ModelFormat.CAIKIT,
-        runtime=endpoint_sr.name,
+        runtime=serving_runtime_from_template.name,
+        wait_for_predictor_pods=True,
     ) as isvc:
         yield isvc
 
 
 @pytest.fixture(scope="class")
 def storage_config_secret(
     admin_client: DynamicClient,
-    endpoint_s3_secret: Secret,
+    models_endpoint_s3_secret: Secret,
     aws_access_key_id: str,
     aws_secret_access_key: str,
     models_s3_bucket_name: str,
@@ -107,8 +56,8 @@ def storage_config_secret(
 ) -> Generator[Secret, None, None]:
     with create_storage_config_secret(
         admin_client=admin_client,
-        endpoint_secret_name=endpoint_s3_secret.name,
-        namespace=endpoint_s3_secret.namespace,
+        endpoint_secret_name=models_endpoint_s3_secret.name,
+        namespace=models_endpoint_s3_secret.namespace,
         aws_access_key=aws_access_key_id,
         aws_secret_access_key=aws_secret_access_key,
         aws_s3_bucket=models_s3_bucket_name,
@@ -120,11 +69,11 @@ def storage_config_secret(
 
 @pytest.fixture()
 def endpoint_pod_with_istio_sidecar(
-    admin_client: DynamicClient, endpoint_namespace: Namespace
+    admin_client: DynamicClient, model_namespace: Namespace
 ) -> Generator[Pod, Any, Any]:
     with create_sidecar_pod(
         admin_client=admin_client,
-        namespace=endpoint_namespace.name,
+        namespace=model_namespace.name,
         use_istio=True,
         pod_name="test-with-istio",
     ) as pod:
@@ -133,11 +82,11 @@ def endpoint_pod_with_istio_sidecar(
 
 @pytest.fixture()
 def endpoint_pod_without_istio_sidecar(
-    admin_client: DynamicClient, endpoint_namespace: Namespace
+    admin_client: DynamicClient, model_namespace: Namespace
 ) -> Generator[Pod, Any, Any]:
     with create_sidecar_pod(
         admin_client=admin_client,
-        namespace=endpoint_namespace.name,
+        namespace=model_namespace.name,
         use_istio=False,
         pod_name="test",
     ) as pod:
@@ -170,11 +119,3 @@ def diff_pod_without_istio_sidecar(
         pod_name="test",
     ) as pod:
         yield pod
-
-
-@pytest.fixture()
-def ready_predictor(admin_client: DynamicClient, endpoint_isvc: InferenceService) -> None:
-    wait_for_kserve_predictor_deployment_replicas(
-        client=admin_client,
-        isvc=endpoint_isvc,
-    )
diff --git a/tests/model_serving/model_server/private_endpoint/test_kserve_private_endpoint.py b/tests/model_serving/model_server/private_endpoint/test_kserve_private_endpoint.py
@@ -2,36 +2,44 @@
 
 import pytest
 from simple_logger.logger import get_logger
-from ocp_resources.namespace import Namespace
 from ocp_resources.inference_service import InferenceService
 from ocp_resources.pod import Pod
-from ocp_resources.deployment import Deployment
 from tests.model_serving.model_server.private_endpoint.utils import curl_from_pod
-from utilities.constants import CurlOutput, ModelEndpoint, Protocols
+from utilities.constants import CurlOutput, ModelEndpoint, Protocols, RuntimeTemplates
 
 LOGGER = get_logger(name=__name__)
 
 
 pytestmark = pytest.mark.usefixtures("skip_if_no_deployed_openshift_serverless", "valid_aws_config")
 
 
+@pytest.mark.parametrize(
+    "model_namespace, serving_runtime_from_template",
+    [
+        pytest.param(
+            {"name": "endpoint"},
+            {
+                "name": "flan-example-runtime",
+                "template-name": RuntimeTemplates.CAIKIT_TGIS_SERVING,
+                "multi-model": False,
+            },
+        )
+    ],
+    indirect=True,
+)
 @pytest.mark.serverless
 class TestKserveInternalEndpoint:
     """Tests the internal endpoint of a kserve predictor"""
 
-    def test_deploy_model_state_loaded(
-        self: Self, endpoint_namespace: Namespace, endpoint_isvc: InferenceService, ready_predictor: Deployment
-    ) -> None:
+    def test_deploy_model_state_loaded(self: Self, endpoint_isvc: InferenceService) -> None:
         """Verifies that the predictor gets to state Loaded"""
         assert endpoint_isvc.instance.status.modelStatus.states.activeModelState == "Loaded"
 
-    def test_deploy_model_url(
-        self: Self, endpoint_namespace: Namespace, endpoint_isvc: InferenceService, ready_predictor: Deployment
-    ) -> None:
+    def test_deploy_model_url(self: Self, endpoint_isvc: InferenceService) -> None:
         """Verifies that the internal endpoint has the expected formatting"""
         assert (
             endpoint_isvc.instance.status.address.url
-            == f"https://{endpoint_isvc.name}.{endpoint_namespace.name}.svc.cluster.local"
+            == f"https://{endpoint_isvc.name}.{endpoint_isvc.namespace}.svc.cluster.local"
         )
 
     def test_curl_with_istio_same_ns(

diff --git a/tests/model_serving/model_server/storage/pvc/conftest.py b/tests/model_serving/model_server/storage/pvc/conftest.py
@@ -15,7 +15,7 @@
 from tests.model_serving.model_server.storage.constants import NFS_STR
 from tests.model_serving.model_server.utils import create_isvc
 from utilities.constants import KServeDeploymentType
-from utilities.infra import get_pods_by_isvc_label, wait_for_kserve_predictor_deployment_replicas
+from utilities.infra import get_pods_by_isvc_label
 
 
 @pytest.fixture(scope="class")
@@ -98,7 +98,6 @@ def predictor_pods_scope_function(admin_client: DynamicClient, pvc_inference_ser
 def predictor_pods_scope_class(
     admin_client: DynamicClient,
     pvc_inference_service: InferenceService,
-    isvc_deployment_ready: None,
 ) -> List[Pod]:
     return get_pods_by_isvc_label(
         client=admin_client,
@@ -140,6 +139,7 @@ def pvc_inference_service(
         "storage_uri": f"pvc://{model_pvc.name}/{downloaded_model_data}",
         "model_format": serving_runtime_from_template.instance.spec.supportedModelFormats[0].name,
         "deployment_mode": request.param.get("deployment-mode", KServeDeploymentType.SERVERLESS),
+        "wait_for_predictor_pods": True,
     }
 
     if min_replicas := request.param.get("min-replicas"):
@@ -149,14 +149,6 @@ def pvc_inference_service(
         yield isvc
 
 
-@pytest.fixture(scope="class")
-def isvc_deployment_ready(admin_client: DynamicClient, pvc_inference_service: InferenceService) -> None:
-    wait_for_kserve_predictor_deployment_replicas(
-        client=admin_client,
-        isvc=pvc_inference_service,
-    )
-
-
 @pytest.fixture()
 def first_predictor_pod(predictor_pods_scope_function: List[Pod]) -> Pod:
     return predictor_pods_scope_function[0]

diff --git a/tests/model_serving/model_server/utils.py b/tests/model_serving/model_server/utils.py
@@ -13,7 +13,7 @@
 )
 from utilities.constants import KServeDeploymentType
 from utilities.inference_utils import UserInference
-
+from utilities.infra import wait_for_kserve_predictor_deployment_replicas
 
 LOGGER = get_logger(name=__name__)
 
@@ -43,6 +43,7 @@ def create_isvc(
     volumes: Optional[dict[str, Any]] = None,
     volumes_mounts: Optional[dict[str, Any]] = None,
     model_version: Optional[str] = None,
+    wait_for_predictor_pods: bool = True,
 ) -> Generator[InferenceService, Any, Any]:
     labels: Dict[str, str] = {}
     predictor_dict: Dict[str, Any] = {
@@ -115,6 +116,13 @@ def create_isvc(
                 status=inference_service.Condition.Status.TRUE,
                 timeout=10 * 60,
             )
+
+        if wait_for_predictor_pods:
+            wait_for_kserve_predictor_deployment_replicas(
+                client=client,
+                isvc=inference_service,
+            )
+
         yield inference_service