Merge branch 'main' into hf_dup_model

Author: dbasunag

CONSTITUTION.md

@@ -79,8 +79,11 @@ All code MUST consider security implications.
 - Avoid running destructive commands without explicit user confirmation
 - Use detect-secrets and gitleaks pre-commit hooks to prevent secret leakage
 - Test code MUST NOT introduce vulnerabilities into the tested systems
+- JIRA ticket links are allowed in PRs and commit messages (our Jira is public)
+- Do NOT reference internal-only resources (Jenkins, Confluence, Slack threads) in code, PRs, or commit messages
+- Do NOT link embargoed or security-restricted (RH-employee-only) tickets
 
-**Rationale**: Tests interact with production-like clusters; security lapses can have real consequences.
+**Rationale**: Tests interact with production-like clusters; security lapses can have real consequences. This is a public repository — only reference publicly accessible resources.
 
 ## Test Development Standards
 

tests/llama_stack/conftest.py

@@ -6,7 +6,7 @@
 import pytest
 from _pytest.fixtures import FixtureRequest
 from kubernetes.dynamic import DynamicClient
-from llama_stack_client import LlamaStackClient
+from llama_stack_client import APIError, LlamaStackClient
 from llama_stack_client.types.vector_store import VectorStore
 from ocp_resources.data_science_cluster import DataScienceCluster
 from ocp_resources.deployment import Deployment
@@ -73,59 +73,6 @@
 distribution_name = generate_random_name(prefix="llama-stack-distribution")
 
 
-def _cleanup_s3_files(
-    bucket_name: str,
-    endpoint_url: str,
-    region: str,
-    access_key_id: str,
-    secret_access_key: str,
-) -> None:
-    """
-    Clean up files from S3 bucket that were uploaded during tests.
-
-    Args:
-        bucket_name: S3 bucket name
-        endpoint_url: S3 endpoint URL
-        region: S3 region
-        access_key_id: AWS access key ID
-        secret_access_key: AWS secret access key
-    """
-
-    try:
-        import boto3
-        from botocore.exceptions import ClientError
-
-        s3_client = boto3.client(
-            service_name="s3",
-            endpoint_url=endpoint_url,
-            aws_access_key_id=access_key_id,
-            aws_secret_access_key=secret_access_key,
-            region_name=region,
-        )
-
-        response = s3_client.list_objects_v2(Bucket=bucket_name)
-
-        if "Contents" not in response:
-            LOGGER.info("No files found to clean up from S3")
-            return
-
-        # We only want to delete files that start with "file-"
-        for obj in response["Contents"]:
-            key = obj["Key"]
-            if key.startswith("file-"):
-                s3_client.delete_object(Bucket=bucket_name, Key=key)
-                LOGGER.debug(f"Deleted file from S3: {key}")
-
-        response = s3_client.list_objects_v2(Bucket=bucket_name)
-
-        if "Contents" not in response:
-            LOGGER.info("No files found to clean up from S3")
-            return
-
-    except ClientError as e:
-        LOGGER.warning(f"Failed to clean up S3 files: {e}")
-
-
 @pytest.fixture(scope="class")
 def enabled_llama_stack_operator(dsc_resource: DataScienceCluster) -> Generator[DataScienceCluster, Any, Any]:
     with update_components_in_dsc(
@@ -385,11 +332,6 @@ def unprivileged_llama_stack_distribution(
     enabled_llama_stack_operator: DataScienceCluster,
     request: FixtureRequest,
     llama_stack_server_config: dict[str, Any],
-    ci_s3_bucket_name: str,
-    ci_s3_bucket_endpoint: str,
-    ci_s3_bucket_region: str,
-    aws_access_key_id: str,
-    aws_secret_access_key: str,
     unprivileged_llama_stack_distribution_secret: Secret,
     unprivileged_postgres_deployment: Deployment,
     unprivileged_postgres_service: Service,
@@ -406,25 +348,6 @@ def unprivileged_llama_stack_distribution(
         lls_dist.wait_for_status(status=LlamaStackDistribution.Status.READY, timeout=600)
         yield lls_dist
 
-        try:
-            env_vars = llama_stack_server_config.get("containerSpec", {}).get("env", [])
-            enable_s3 = any(env.get("name") == "ENABLE_S3" and env.get("value") == "s3" for env in env_vars)
-
-            if enable_s3:
-                try:
-                    _cleanup_s3_files(
-                        bucket_name=ci_s3_bucket_name,
-                        endpoint_url=ci_s3_bucket_endpoint,
-                        region=ci_s3_bucket_region,
-                        access_key_id=aws_access_key_id,
-                        secret_access_key=aws_secret_access_key,
-                    )
-                except Exception as e:  # noqa: BLE001
-                    LOGGER.warning(f"Failed to clean up S3 files: {e}")
-
-        except Exception as e:  # noqa: BLE001
-            LOGGER.warning(f"Failed to clean up S3 files: {e}")
-
 
 @pytest.fixture(scope="class")
 def llama_stack_distribution(
@@ -433,11 +356,6 @@ def llama_stack_distribution(
     enabled_llama_stack_operator: DataScienceCluster,
     request: FixtureRequest,
     llama_stack_server_config: dict[str, Any],
-    ci_s3_bucket_name: str,
-    ci_s3_bucket_endpoint: str,
-    ci_s3_bucket_region: str,
-    aws_access_key_id: str,
-    aws_secret_access_key: str,
     llama_stack_distribution_secret: Secret,
     postgres_deployment: Deployment,
     postgres_service: Service,
@@ -453,25 +371,6 @@ def llama_stack_distribution(
         lls_dist.wait_for_status(status=LlamaStackDistribution.Status.READY, timeout=600)
         yield lls_dist
 
-        try:
-            env_vars = llama_stack_server_config.get("containerSpec", {}).get("env", [])
-            enable_s3 = any(env.get("name") == "ENABLE_S3" and env.get("value") == "s3" for env in env_vars)
-
-            if enable_s3:
-                try:
-                    _cleanup_s3_files(
-                        bucket_name=ci_s3_bucket_name,
-                        endpoint_url=ci_s3_bucket_endpoint,
-                        region=ci_s3_bucket_region,
-                        access_key_id=aws_access_key_id,
-                        secret_access_key=aws_secret_access_key,
-                    )
-                except Exception as e:  # noqa: BLE001
-                    LOGGER.warning(f"Failed to clean up S3 files: {e}")
-
-        except Exception as e:  # noqa: BLE001
-            LOGGER.warning(f"Failed to clean up S3 files: {e}")
-
 
 def _get_llama_stack_distribution_deployment(
     client: DynamicClient,
@@ -642,11 +541,37 @@ def _create_llama_stack_client(
             http_client=http_client,
         )
         wait_for_llama_stack_client_ready(client=client)
+        existing_file_ids = {f.id for f in client.files.list().data}
+
         yield client
+
+        _cleanup_files(client=client, existing_file_ids=existing_file_ids)
     finally:
         http_client.close()
 
 
+def _cleanup_files(client: LlamaStackClient, existing_file_ids: set[str]) -> None:
+    """Delete files created during test execution via the LlamaStack files API.
+
+    Only deletes files whose IDs were not present before the test ran,
+    avoiding interference with other test sessions.
+
+    Args:
+        client: The LlamaStackClient used during the test
+        existing_file_ids: File IDs that existed before the test started
+    """
+    try:
+        for file in client.files.list().data:
+            if file.id not in existing_file_ids:
+                try:
+                    client.files.delete(file_id=file.id)
+                    LOGGER.debug(f"Deleted file: {file.id}")
+                except APIError as e:
+                    LOGGER.warning(f"Failed to delete file {file.id}: {e}")
+    except APIError as e:
+        LOGGER.warning(f"Failed to clean up files: {e}")
+
+
 @pytest.fixture(scope="class")
 def unprivileged_llama_stack_client(
     unprivileged_llama_stack_test_route: Route,

tests/model_registry/mcp_servers/search/test_filtering.py

@@ -24,6 +24,7 @@ class TestMCPServerFiltering:
                 id="by_provider",
             ),
             pytest.param("tags='math'", 1, CALCULATOR_SERVER_NAME, None, id="by_tags"),
+            pytest.param("license='BSD 3-Clause'", 1, "file-manager", ("license", "BSD 3-Clause"), id="by_license"),
         ],
     )
     def test_filter_by_field(
@@ -35,7 +36,7 @@ def test_filter_by_field(
         expected_name: str,
         field_check: tuple[str, str] | None,
     ):
-        """TC-API-003, TC-API-005: Test filtering MCP servers by provider and tags."""
+        """TC-API-003, TC-API-005, TC-API-009: Test filtering MCP servers by provider, tags, and license."""
         response = execute_get_command(
             url=f"{mcp_catalog_rest_urls[0]}mcp_servers",
             headers=model_registry_rest_headers,

tests/model_registry/mcp_servers/search/test_ordering.py

@@ -12,7 +12,6 @@
 class TestMCPServerOrdering:
     """RHOAIENG-51584: Tests for MCP server ordering functionality."""
 
-    @pytest.mark.xfail(reason="RHOAIENG-52448: sortOrder/orderBy not working, fix in PR #2367")
     @pytest.mark.parametrize(
         "sort_order",
         [

tests/model_serving/model_server/llmd/conftest.py

@@ -247,6 +247,7 @@ def _create_llmisvc_from_config(
     namespace: str,
     client: DynamicClient,
     service_account: str | None = None,
+    teardown: bool = True,
 ) -> Generator[LLMInferenceService, Any]:
     """Create an LLMInferenceService from a config class."""
     LOGGER.info(f"\n{config_cls.describe(namespace=namespace)}")
@@ -283,7 +284,7 @@ def _create_llmisvc_from_config(
         "namespace": namespace,
         "annotations": config_cls.annotations(),
         "label": config_cls.labels(),
-        "teardown": True,
+        "teardown": teardown,
         "model": model,
         "replicas": config_cls.replicas,
         "router": config_cls.router_config(),

tests/model_serving/model_server/llmd/llmd_configs/config_base.py

@@ -18,7 +18,7 @@ class LLMISvcConfig:
     container_image = None
     template_config_ref = "kserve-config-llm-template"
     enable_auth = False
-    wait_timeout = 180
+    wait_timeout = 240
 
     @classmethod
     def container_resources(cls):

tests/model_serving/model_server/upgrade/conftest.py

@@ -5,7 +5,9 @@
 import yaml
 from kubernetes.dynamic import DynamicClient
 from ocp_resources.config_map import ConfigMap
+from ocp_resources.gateway import Gateway
 from ocp_resources.inference_service import InferenceService
+from ocp_resources.llm_inference_service import LLMInferenceService
 from ocp_resources.namespace import Namespace
 from ocp_resources.role import Role
 from ocp_resources.role_binding import RoleBinding
@@ -23,6 +25,7 @@
     ModelVersion,
     Protocols,
     RuntimeTemplates,
+    Timeout,
 )
 from utilities.inference_utils import create_isvc
 from utilities.infra import (
@@ -32,6 +35,8 @@
     s3_endpoint_secret,
     update_configmap_data,
 )
+from utilities.llmd_constants import KServeGateway, LLMDGateway
+from utilities.llmd_utils import create_llmd_gateway
 from utilities.logger import RedactedString
 from utilities.serving_runtime import ServingRuntimeFromTemplate
 
@@ -42,6 +47,7 @@
 MODEL_CAR_UPGRADE_NAMESPACE = "upgrade-model-car"
 METRICS_UPGRADE_NAMESPACE = "upgrade-metrics"
 PRIVATE_ENDPOINT_UPGRADE_NAMESPACE = "upgrade-private-endpoint"
+LLMD_UPGRADE_NAMESPACE = "upgrade-llmd"
 S3_CONNECTION = "upgrade-connection"
 
 
@@ -765,3 +771,88 @@ def private_endpoint_inference_service_fixture(
             **isvc_kwargs,
         ) as isvc:
             yield isvc
+
+
+# LLMD Upgrade Fixtures
+@pytest.fixture(scope="session")
+def llmd_namespace_fixture(
+    pytestconfig: pytest.Config,
+    admin_client: DynamicClient,
+    teardown_resources: bool,
+) -> Generator[Namespace, Any, Any]:
+    """Namespace for LLMD upgrade tests."""
+    ns = Namespace(client=admin_client, name=LLMD_UPGRADE_NAMESPACE)
+
+    if pytestconfig.option.post_upgrade:
+        yield ns
+        ns.clean_up()
+    else:
+        with create_ns(
+            admin_client=admin_client,
+            name=LLMD_UPGRADE_NAMESPACE,
+            model_mesh_enabled=False,
+            add_dashboard_label=True,
+            teardown=teardown_resources,
+        ) as ns:
+            yield ns
+
+
+@pytest.fixture(scope="session")
+def llmd_gateway_fixture(
+    pytestconfig: pytest.Config,
+    admin_client: DynamicClient,
+    teardown_resources: bool,
+) -> Generator[Gateway, Any, Any]:
+    """Shared LLMD Gateway for upgrade tests."""
+    gateway = Gateway(
+        client=admin_client,
+        name=LLMDGateway.DEFAULT_NAME,
+        namespace=LLMDGateway.DEFAULT_NAMESPACE,
+        api_group=KServeGateway.API_GROUP,
+    )
+
+    if pytestconfig.option.post_upgrade:
+        yield gateway
+        gateway.clean_up()
+    else:
+        with create_llmd_gateway(
+            client=admin_client,
+            namespace=LLMDGateway.DEFAULT_NAMESPACE,
+            gateway_class_name=LLMDGateway.DEFAULT_CLASS,
+            wait_for_condition=True,
+            timeout=Timeout.TIMEOUT_1MIN,
+            teardown=teardown_resources,
+        ) as gateway:
+            yield gateway
+
+
+@pytest.fixture(scope="session")
+def llmd_inference_service_fixture(
+    pytestconfig: pytest.Config,
+    admin_client: DynamicClient,
+    llmd_namespace_fixture: Namespace,
+    llmd_gateway_fixture: Gateway,
+    teardown_resources: bool,
+) -> Generator[LLMInferenceService, Any, Any]:
+    """LLMInferenceService using TinyLlama OCI for upgrade tests."""
+    from tests.model_serving.model_server.llmd.conftest import _create_llmisvc_from_config
+    from tests.model_serving.model_server.llmd.llmd_configs import TinyLlamaOciConfig
+
+    config_cls = TinyLlamaOciConfig
+    llmisvc = LLMInferenceService(
+        client=admin_client,
+        name=config_cls.name,
+        namespace=llmd_namespace_fixture.name,
+    )
+
+    if pytestconfig.option.post_upgrade:
+        yield llmisvc
+        llmisvc.clean_up()
+    else:
+        with _create_llmisvc_from_config(
+            config_cls=config_cls,
+            namespace=llmd_namespace_fixture.name,
+            client=admin_client,
+            teardown=teardown_resources,
+        ) as llmisvc:
+            yield llmisvc