snowflake-ml-python 1.12.0__tar.gz → 1.14.0__tar.gz

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/CHANGELOG.md

@@ -1,11 +1,37 @@
 # Release History
 
+## 1.14.0
+
+### Bug Fixes
+
+### Behavior Changes
+
+### New Features
+
+* ML Job: The `additional_payloads` argument is now **deprecated** in favor of `imports`.
+
+## 1.13.0
+
+### Bug Fixes
+
+### Behavior Changes
+
+### New Features
+
+* Registry: Log a HuggingFace model without having to load the model in memory using
+ the `huggingface_pipeline.HuggingFacePipelineModel`. Requires `huggingface_hub` package to installed.
+ To disable downloading HuggingFace repository, provide `download_snapshot=False` while creating the
+ `huggingface_pipeline.HuggingFacePipelineModel` object.
+* Registry: Added support for XGBoost models to use `enable_categorical=True` with pandas DataFrame
+* Registry: Added support to display privatelink inference endpoint in ModelVersion list services.
+
 ## 1.12.0
 
 ### Bug Fixes
 
 * Registry: Fixed an issue where the string representation of dictionary-type output columns was being incorrectly
   created during structured output deserialization. Now, the original data type is properly preserved.
+* Registry: Fixed the inference server performance issue for wide (500+ features) and JSON inputs.
 
 ### Behavior Changes
 

{snowflake_ml_python-1.12.0/snowflake_ml_python.egg-info → snowflake_ml_python-1.14.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: snowflake-ml-python
-Version: 1.12.0
+Version: 1.14.0
 Summary: The machine learning client library that is used for interacting with Snowflake to build machine learning solutions.
 Author-email: "Snowflake, Inc" <support@snowflake.com>
 License: 
@@ -410,12 +410,38 @@ NOTE: Version 1.7.0 is used as example here. Please choose the the latest versio
 
 # Release History
 
+## 1.14.0
+
+### Bug Fixes
+
+### Behavior Changes
+
+### New Features
+
+* ML Job: The `additional_payloads` argument is now **deprecated** in favor of `imports`.
+
+## 1.13.0
+
+### Bug Fixes
+
+### Behavior Changes
+
+### New Features
+
+* Registry: Log a HuggingFace model without having to load the model in memory using
+ the `huggingface_pipeline.HuggingFacePipelineModel`. Requires `huggingface_hub` package to installed.
+ To disable downloading HuggingFace repository, provide `download_snapshot=False` while creating the
+ `huggingface_pipeline.HuggingFacePipelineModel` object.
+* Registry: Added support for XGBoost models to use `enable_categorical=True` with pandas DataFrame
+* Registry: Added support to display privatelink inference endpoint in ModelVersion list services.
+
 ## 1.12.0
 
 ### Bug Fixes
 
 * Registry: Fixed an issue where the string representation of dictionary-type output columns was being incorrectly
   created during structured output deserialization. Now, the original data type is properly preserved.
+* Registry: Fixed the inference server performance issue for wide (500+ features) and JSON inputs.
 
 ### Behavior Changes
 

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/_internal/telemetry.py

@@ -680,8 +680,10 @@ def _extract_arg_value(field: str, func_spec: inspect.FullArgSpec, args: Any, kw
     elif func_spec.kwonlydefaults and field in func_spec.kwonlyargs:
         if field in kwargs:
             return True, kwargs[field]
-        else:
+        elif field in func_spec.kwonlydefaults:
             return True, func_spec.kwonlydefaults[field]
+        else:
+            return False, None
     else:
         return False, None
 

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/experiment/experiment_tracking.py

@@ -7,7 +7,7 @@ from urllib.parse import quote
 import snowflake.snowpark._internal.utils as snowpark_utils
 from snowflake.ml import model as ml_model, registry
 from snowflake.ml._internal.human_readable_id import hrid_generator
-from snowflake.ml._internal.utils import sql_identifier
+from snowflake.ml._internal.utils import mixins, sql_identifier
 from snowflake.ml.experiment import (
     _entities as entities,
     _experiment_info as experiment_info,
@@ -23,7 +23,7 @@ from snowflake.snowpark import session
 DEFAULT_EXPERIMENT_NAME = sql_identifier.SqlIdentifier("DEFAULT")
 
 
-class ExperimentTracking:
+class ExperimentTracking(mixins.SerializableSessionMixin):
     """
     Class to manage experiments in Snowflake.
     """
@@ -73,12 +73,34 @@ class ExperimentTracking:
             database_name=self._database_name,
             schema_name=self._schema_name,
         )
+        self._session = session
 
         # The experiment in context
         self._experiment: Optional[entities.Experiment] = None
         # The run in context
         self._run: Optional[entities.Run] = None
 
+    def __getstate__(self) -> dict[str, Any]:
+        state = super().__getstate__()
+        # Remove unpicklable attributes
+        state["_sql_client"] = None
+        state["_registry"] = None
+        return state
+
+    def __setstate__(self, state: dict[str, Any]) -> None:
+        super().__setstate__(state)
+        # Restore unpicklable attributes
+        self._sql_client = sql_client.ExperimentTrackingSQLClient(
+            session=self._session,
+            database_name=self._database_name,
+            schema_name=self._schema_name,
+        )
+        self._registry = registry.Registry(
+            session=self._session,
+            database_name=self._database_name,
+            schema_name=self._schema_name,
+        )
+
     def set_experiment(
         self,
         experiment_name: str,

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/jobs/_utils/constants.py

@@ -25,7 +25,7 @@ RESULT_PATH_DEFAULT_VALUE = f"{OUTPUT_STAGE_SUBPATH}/mljob_result.pkl"
 DEFAULT_IMAGE_REPO = "/snowflake/images/snowflake_images"
 DEFAULT_IMAGE_CPU = "st_plat/runtime/x86/runtime_image/snowbooks"
 DEFAULT_IMAGE_GPU = "st_plat/runtime/x86/generic_gpu/runtime_image/snowbooks"
-DEFAULT_IMAGE_TAG = "1.6.2"
+DEFAULT_IMAGE_TAG = "1.8.0"
 DEFAULT_ENTRYPOINT_PATH = "func.py"
 
 # Percent of container memory to allocate for /dev/shm volume

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/jobs/_utils/scripts/mljob_launcher.py

@@ -234,10 +234,6 @@ def run_script(script_path: str, *script_args: Any, main_func: Optional[str] = N
     if payload_dir and payload_dir not in sys.path:
         sys.path.insert(0, payload_dir)
 
-    # Create a Snowpark session before running the script
-    # Session can be retrieved from using snowflake.snowpark.context.get_active_session()
-    session = Session.builder.configs(SnowflakeLoginOptions()).create()  # noqa: F841
-
     try:
 
         if main_func:
@@ -287,6 +283,19 @@ def main(script_path: str, *script_args: Any, script_main_func: Optional[str] =
     output_dir = os.path.dirname(result_abs_path)
     os.makedirs(output_dir, exist_ok=True)
 
+    try:
+        import ray
+
+        ray.init(address="auto")
+    except ModuleNotFoundError:
+        warnings.warn("Ray is not installed, skipping Ray initialization", ImportWarning, stacklevel=1)
+
+    # Create a Snowpark session before starting
+    # Session can be retrieved from using snowflake.snowpark.context.get_active_session()
+    config = SnowflakeLoginOptions()
+    config["client_session_keep_alive"] = "True"
+    session = Session.builder.configs(config).create()  # noqa: F841
+
     try:
         # Wait for minimum required instances if specified
         min_instances_str = os.environ.get(MIN_INSTANCES_ENV_VAR) or "1"
@@ -342,6 +351,9 @@ def main(script_path: str, *script_args: Any, script_main_func: Optional[str] =
                 f"Failed to serialize JSON result to {result_json_path}: {json_exc}", RuntimeWarning, stacklevel=1
             )
 
+        # Close the session after serializing the result
+        session.close()
+
 
 if __name__ == "__main__":
     # Parse command line arguments

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/jobs/job.py

@@ -50,7 +50,7 @@ class MLJob(Generic[T], SerializableSessionMixin):
     def min_instances(self) -> int:
         try:
             return int(self._container_spec["env"].get(constants.MIN_INSTANCES_ENV_VAR, 1))
-        except TypeError:
+        except (TypeError, ValueError):
             return 1
 
     @property
@@ -83,7 +83,12 @@ class MLJob(Generic[T], SerializableSessionMixin):
     def _container_spec(self) -> dict[str, Any]:
         """Get the job's main container spec."""
         containers = self._service_spec["spec"]["containers"]
-        container_spec = next(c for c in containers if c["name"] == constants.DEFAULT_CONTAINER_NAME)
+        if len(containers) == 1:
+            return cast(dict[str, Any], containers[0])
+        try:
+            container_spec = next(c for c in containers if c["name"] == constants.DEFAULT_CONTAINER_NAME)
+        except StopIteration:
+            raise ValueError(f"Container '{constants.DEFAULT_CONTAINER_NAME}' not found in job {self.name}")
         return cast(dict[str, Any], container_spec)
 
     @property
@@ -160,7 +165,7 @@ class MLJob(Generic[T], SerializableSessionMixin):
         Returns:
             The job's execution logs.
         """
-        logs = _get_logs(self._session, self.id, limit, instance_id, verbose)
+        logs = _get_logs(self._session, self.id, limit, instance_id, self._container_spec["name"], verbose)
         assert isinstance(logs, str)  # mypy
         if as_list:
             return logs.splitlines()
@@ -278,7 +283,12 @@ def _get_service_spec(session: snowpark.Session, job_id: str) -> dict[str, Any]:
 
 @telemetry.send_api_usage_telemetry(project=_PROJECT, func_params_to_log=["job_id", "limit", "instance_id"])
 def _get_logs(
-    session: snowpark.Session, job_id: str, limit: int = -1, instance_id: Optional[int] = None, verbose: bool = True
+    session: snowpark.Session,
+    job_id: str,
+    limit: int = -1,
+    instance_id: Optional[int] = None,
+    container_name: str = constants.DEFAULT_CONTAINER_NAME,
+    verbose: bool = True,
 ) -> str:
     """
     Retrieve the job's execution logs.
@@ -288,6 +298,7 @@ def _get_logs(
         limit: The maximum number of lines to return. Negative values are treated as no limit.
         session: The Snowpark session to use. If none specified, uses active session.
         instance_id: Optional instance ID to get logs from a specific instance.
+        container_name: The container name to get logs from a specific container.
         verbose: Whether to return the full log or just the portion between START and END messages.
 
     Returns:
@@ -308,7 +319,7 @@ def _get_logs(
     params: list[Any] = [
         job_id,
         0 if instance_id is None else instance_id,
-        constants.DEFAULT_CONTAINER_NAME,
+        container_name,
     ]
     if limit > 0:
         params.append(limit)
@@ -334,7 +345,7 @@ def _get_logs(
                     job_id,
                     limit=limit,
                     instance_id=instance_id if instance_id else 0,
-                    container_name=constants.DEFAULT_CONTAINER_NAME,
+                    container_name=container_name,
                 )
                 full_log = os.linesep.join(row[0] for row in logs)
 

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/jobs/manager.py

@@ -58,7 +58,8 @@ def list_jobs(
         >>> from snowflake.ml.jobs import list_jobs
         >>> list_jobs(limit=5)
     """
-    session = session or get_active_session()
+
+    session = _ensure_session(session)
     try:
         df = _get_job_history_spcs(
             session,
@@ -162,7 +163,7 @@ def _get_job_history_spcs(
 @telemetry.send_api_usage_telemetry(project=_PROJECT)
 def get_job(job_id: str, session: Optional[snowpark.Session] = None) -> jb.MLJob[Any]:
     """Retrieve a job service from the backend."""
-    session = session or get_active_session()
+    session = _ensure_session(session)
     try:
         database, schema, job_name = identifier.parse_schema_level_object_identifier(job_id)
         database = identifier.resolve_identifier(cast(str, database or session.get_current_database()))
@@ -231,6 +232,7 @@ def submit_file(
             enable_metrics (bool): Whether to enable metrics publishing for the job.
             query_warehouse (str): The query warehouse to use. Defaults to session warehouse.
             spec_overrides (dict): A dictionary of overrides for the service spec.
+            imports (list[Union[tuple[str, str], tuple[str]]]): A list of additional payloads used in the job.
 
     Returns:
         An object representing the submitted job.
@@ -285,6 +287,7 @@ def submit_directory(
             enable_metrics (bool): Whether to enable metrics publishing for the job.
             query_warehouse (str): The query warehouse to use. Defaults to session warehouse.
             spec_overrides (dict): A dictionary of overrides for the service spec.
+            imports (list[Union[tuple[str, str], tuple[str]]]): A list of additional payloads used in the job.
 
     Returns:
         An object representing the submitted job.
@@ -340,6 +343,7 @@ def submit_from_stage(
             enable_metrics (bool): Whether to enable metrics publishing for the job.
             query_warehouse (str): The query warehouse to use. Defaults to session warehouse.
             spec_overrides (dict): A dictionary of overrides for the service spec.
+            imports (list[Union[tuple[str, str], tuple[str]]]): A list of additional payloads used in the job.
 
     Returns:
         An object representing the submitted job.
@@ -403,6 +407,8 @@ def _submit_job(
         "num_instances",  # deprecated
         "target_instances",
         "min_instances",
+        "enable_metrics",
+        "query_warehouse",
     ],
 )
 def _submit_job(
@@ -434,8 +440,10 @@ def _submit_job(
 
     Raises:
         ValueError: If database or schema value(s) are invalid
+        RuntimeError: If schema is not specified in session context or job submission
+        snowpark.exceptions.SnowparkSQLException: if failed to upload payload
     """
-    session = session or get_active_session()
+    session = _ensure_session(session)
 
     # Check for deprecated args
     if "num_instances" in kwargs:
@@ -444,6 +452,13 @@ def _submit_job(
         )
         target_instances = max(target_instances, kwargs.pop("num_instances"))
 
+    imports = None
+    if "additional_payloads" in kwargs:
+        logger.warning(
+            "'additional_payloads' is deprecated and will be removed in a future release. Use 'imports' instead."
+        )
+        imports = kwargs.pop("additional_payloads")
+
     # Use kwargs for less common optional parameters
     database = kwargs.pop("database", None)
     schema = kwargs.pop("schema", None)
@@ -454,10 +469,7 @@ def _submit_job(
     spec_overrides = kwargs.pop("spec_overrides", None)
     enable_metrics = kwargs.pop("enable_metrics", True)
     query_warehouse = kwargs.pop("query_warehouse", session.get_current_warehouse())
-    additional_payloads = kwargs.pop("additional_payloads", None)
-
-    if additional_payloads:
-        logger.warning("'additional_payloads' is in private preview since 1.9.1. Do not use it in production.")
+    imports = kwargs.pop("imports", None) or imports
 
     # Warn if there are unknown kwargs
     if kwargs:
@@ -486,10 +498,33 @@ def _submit_job(
     stage_name = f"@{'.'.join(filter(None, stage_path_parts[:3]))}"
     stage_path = pathlib.PurePosixPath(f"{stage_name}{stage_path_parts[-1].rstrip('/')}/{job_name}")
 
-    # Upload payload
-    uploaded_payload = payload_utils.JobPayload(
-        source, entrypoint=entrypoint, pip_requirements=pip_requirements, additional_payloads=additional_payloads
-    ).upload(session, stage_path)
+    try:
+        # Upload payload
+        uploaded_payload = payload_utils.JobPayload(
+            source, entrypoint=entrypoint, pip_requirements=pip_requirements, additional_payloads=imports
+        ).upload(session, stage_path)
+    except snowpark.exceptions.SnowparkSQLException as e:
+        if e.sql_error_code == 90106:
+            raise RuntimeError(
+                "Please specify a schema, either in the session context or as a parameter in the job submission"
+            )
+        raise
+
+    # FIXME: Temporary patches, remove this after v1 is deprecated
+    if target_instances > 1:
+        default_spec_overrides = {
+            "spec": {
+                "endpoints": [
+                    {"name": "ray-dashboard-endpoint", "port": 12003, "protocol": "TCP"},
+                ]
+            },
+        }
+        if spec_overrides:
+            spec_overrides = spec_utils.merge_patch(
+                default_spec_overrides, spec_overrides, display_name="spec_overrides"
+            )
+        else:
+            spec_overrides = default_spec_overrides
 
     if feature_flags.FeatureFlags.USE_SUBMIT_JOB_V2.is_enabled():
         # Add default env vars (extracted from spec_utils.generate_service_spec)
@@ -651,3 +686,17 @@ def _do_submit_job_v2(
     actual_job_id = query_helper.run_query(session, query_template, params=params)[0][0]
 
     return get_job(actual_job_id, session=session)
+
+
+def _ensure_session(session: Optional[snowpark.Session]) -> snowpark.Session:
+    try:
+        session = session or get_active_session()
+    except snowpark.exceptions.SnowparkSessionException as e:
+        if "More than one active session" in e.message:
+            raise RuntimeError(
+                "More than one active session is found. Please specify the session explicitly as a parameter"
+            ) from None
+        if "No default Session is found" in e.message:
+            raise RuntimeError("No active session is found. Please create a session") from None
+        raise
+    return session

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/lineage/lineage_node.py

@@ -83,7 +83,6 @@ class LineageNode(mixins.SerializableSessionMixin):
         raise NotImplementedError()
 
     @telemetry.send_api_usage_telemetry(project=_PROJECT)
-    @snowpark._internal.utils.private_preview(version="1.5.3")
     def lineage(
         self,
         direction: Literal["upstream", "downstream"] = "downstream",

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/model/_client/model/batch_inference_specs.py

@@ -4,14 +4,11 @@ from pydantic import BaseModel
 
 
 class InputSpec(BaseModel):
-    input_stage_location: str
-    input_file_pattern: str = "*"
+    stage_location: str
 
 
 class OutputSpec(BaseModel):
-    output_stage_location: str
-    output_file_prefix: Optional[str] = None
-    completion_filename: str = "_SUCCESS"
+    stage_location: str
 
 
 class JobSpec(BaseModel):
@@ -25,3 +22,4 @@ class JobSpec(BaseModel):
     warehouse: Optional[str] = None
     cpu_requests: Optional[str] = None
     memory_requests: Optional[str] = None
+    replicas: Optional[int] = None

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/model/_client/model/model_version_impl.py

@@ -590,11 +590,12 @@ class ModelVersion(lineage_node.LineageNode):
             cpu_requests=job_spec.cpu_requests,
             memory_requests=job_spec.memory_requests,
             job_name=job_name,
+            replicas=job_spec.replicas,
             # input and output
-            input_stage_location=input_spec.input_stage_location,
-            input_file_pattern=input_spec.input_file_pattern,
-            output_stage_location=output_spec.output_stage_location,
-            completion_filename=output_spec.completion_filename,
+            input_stage_location=input_spec.stage_location,
+            input_file_pattern="*",
+            output_stage_location=output_spec.stage_location,
+            completion_filename="_SUCCESS",
             # misc
             statement_params=statement_params,
         )
@@ -787,7 +788,7 @@ class ModelVersion(lineage_node.LineageNode):
         inference_engine_args: service_ops.InferenceEngineArgs,
         gpu_requests: Optional[Union[str, int]] = None,
     ) -> Optional[service_ops.InferenceEngineArgs]:
-        """Enrich inference engine args with model path and tensor parallelism settings.
+        """Enrich inference engine args with tensor parallelism settings.
 
         Args:
             inference_engine_args: The original inference engine args
@@ -802,21 +803,6 @@ class ModelVersion(lineage_node.LineageNode):
         if inference_engine_args.inference_engine_args_override is None:
             inference_engine_args.inference_engine_args_override = []
 
-        # Get model stage path and strip off "snow://" prefix
-        model_stage_path = self._model_ops.get_model_version_stage_path(
-            database_name=None,
-            schema_name=None,
-            model_name=self._model_name,
-            version_name=self._version_name,
-        )
-
-        # Strip "snow://" prefix
-        if model_stage_path.startswith("snow://"):
-            model_stage_path = model_stage_path.replace("snow://", "", 1)
-
-        # Always overwrite the model key by appending
-        inference_engine_args.inference_engine_args_override.append(f"--model={model_stage_path}")
-
         gpu_count = None
 
         # Set tensor-parallelism if gpu_requests is specified

{snowflake_ml_python-1.12.0 → snowflake_ml_python-1.14.0}/snowflake/ml/model/_client/ops/model_ops.py

@@ -47,6 +47,8 @@ class ServiceInfo(TypedDict):
 class ModelOperator:
     INFERENCE_SERVICE_ENDPOINT_NAME = "inference"
     INGRESS_ENDPOINT_URL_SUFFIX = "snowflakecomputing.app"
+    # app-service-privatelink might not contain "snowflakecomputing" in the url - using the minimum required substring
+    PRIVATELINK_INGRESS_ENDPOINT_URL_SUBSTRING = "privatelink.snowflake"
 
     def __init__(
         self,
@@ -612,6 +614,36 @@ class ModelOperator:
             statement_params=statement_params,
         )
 
+    def _is_privatelink_connection(self) -> bool:
+        """Detect if the current session is using a privatelink connection."""
+        try:
+            host = self._session.connection.host
+            return ModelOperator.PRIVATELINK_INGRESS_ENDPOINT_URL_SUBSTRING in host
+        except AttributeError:
+            return False
+
+    def _extract_and_validate_ingress_url(self, res_row: "row.Row") -> Optional[str]:
+        """Extract and validate ingress URL from endpoint row."""
+        url_value = res_row[self._service_client.MODEL_INFERENCE_SERVICE_ENDPOINT_INGRESS_URL_COL_NAME]
+        if url_value is None:
+            return None
+        url_str = str(url_value)
+        return url_str if url_str.endswith(ModelOperator.INGRESS_ENDPOINT_URL_SUFFIX) else None
+
+    def _extract_and_validate_privatelink_url(self, res_row: "row.Row") -> Optional[str]:
+        """Extract and validate privatelink ingress URL from endpoint row."""
+        # Check if the privatelink_ingress_url column exists
+        col_name = self._service_client.MODEL_INFERENCE_SERVICE_ENDPOINT_PRIVATELINK_INGRESS_URL_COL_NAME
+        if col_name not in res_row:
+            # Column doesn't exist in query result for non-Business Critical accounts
+            return None
+
+        url_value = res_row[col_name]
+        if url_value is None:
+            return None
+        url_str = str(url_value)
+        return url_str if ModelOperator.PRIVATELINK_INGRESS_ENDPOINT_URL_SUBSTRING in url_str else None
+
     def show_services(
         self,
         *,
@@ -644,8 +676,10 @@ class ModelOperator:
         fully_qualified_service_names = [str(service) for service in json_array if "MODEL_BUILD_" not in service]
 
         result: list[ServiceInfo] = []
+        is_privatelink_connection = self._is_privatelink_connection()
+
         for fully_qualified_service_name in fully_qualified_service_names:
-            ingress_url: Optional[str] = None
+            inference_endpoint: Optional[str] = None
             db, schema, service_name = sql_identifier.parse_fully_qualified_name(fully_qualified_service_name)
             statuses = self._service_client.get_service_container_statuses(
                 database_name=db, schema_name=schema, service_name=service_name, statement_params=statement_params
@@ -659,17 +693,23 @@ class ModelOperator:
             ):
                 if (
                     res_row[self._service_client.MODEL_INFERENCE_SERVICE_ENDPOINT_NAME_COL_NAME]
-                    == self.INFERENCE_SERVICE_ENDPOINT_NAME
-                    and res_row[self._service_client.MODEL_INFERENCE_SERVICE_ENDPOINT_INGRESS_URL_COL_NAME] is not None
+                    != self.INFERENCE_SERVICE_ENDPOINT_NAME
                 ):
-                    ingress_url = str(
-                        res_row[self._service_client.MODEL_INFERENCE_SERVICE_ENDPOINT_INGRESS_URL_COL_NAME]
-                    )
-                    if not ingress_url.endswith(ModelOperator.INGRESS_ENDPOINT_URL_SUFFIX):
-                        ingress_url = None
+                    continue
+
+                ingress_url = self._extract_and_validate_ingress_url(res_row)
+                privatelink_ingress_url = self._extract_and_validate_privatelink_url(res_row)
+
+                if is_privatelink_connection and privatelink_ingress_url is not None:
+                    inference_endpoint = privatelink_ingress_url
+                else:
+                    inference_endpoint = ingress_url
+
             result.append(
                 ServiceInfo(
-                    name=fully_qualified_service_name, status=service_status.value, inference_endpoint=ingress_url
+                    name=fully_qualified_service_name,
+                    status=service_status.value,
+                    inference_endpoint=inference_endpoint,
                 )
             )