SIENTIAPDE-1773

Enhance environment configuration and update dependencies

- Added new environment variables for PluginStore and MLflow configuration in `.env.example`, including `RUNTIME`, `STORE_BASE_URL`, `STORE_OWNER`, `STORE_REPO`, `STORE_BRANCH`, `STORE_USERNAME`, `STORE_PASSWORD`, `STORE_CACHE_TTL_SECONDS`, `PYPI_SERVER`, `PYPI_USERNAME`, and `PYPI_PASSWORD`.
- Updated `git-requirements-mapping.txt` to reflect changes in repository names.
- Modified `requirements-light.txt` and `requirements.txt` to upgrade `sientia-dataops-library` to version 1.12.0 and `sientia-mlops-library` to version 0.8.1.
- Updated `values.yaml` to include new environment variables for worker runtime and PluginStore configuration.
- Refactored E2E tests to utilize new MLflow repository stubs and PluginStore mocks for improved testing accuracy.

laborious/activities/activities.py

@@ -7,6 +7,10 @@ with workflow.unsafe.imports_passed_through():
     from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
     from sientia_do.observability.logger import Logger
     from sientia_do.repository.minio_repository import MinioRepository
+    from sientia_model.model_repository.mlflow_repository import SientiaMLflowRepository
+    from sientia_model.model_repository.plugin_store import PluginStore
+
+    from laborious.utils.connectors_config import build_mlflow_config
 
     from laborious.activities.api import API
     from laborious.activities.gates import Gates
@@ -18,61 +22,73 @@ with workflow.unsafe.imports_passed_through():
 
 class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
     """
-    Main activities orchestrator for the Laborious system.
+    Central orchestrator for all Temporal activities used by Laborious workflows.
 
-    This class combines functionality from multiple activity classes to provide
-    a unified interface for all workflow operations. It manages database connections,
-    MLFlow model interactions, data quality validation, and OPC server communications.
+    Composes Storage (Postgres + MinIO offload), MLFlow (wrapper-based inference and retrain
+    via ``SientiaMLflowRepository``), Gates (data quality and ML response filters), OPC exports,
+    drift/simple metrics, and PI Web API writes. The worker constructs one ``Activities`` instance
+    per process and registers its callables on multiple workers bound to different task queues.
 
-    The class implements multiple inheritance to combine specialized functionality:
-    - Storage: Database operations and data persistence
-    - MLFlow: Model inference and transformation operations
-    - Gates: Data quality validation and filtering mechanisms
-    - OPC: Real-time data export to OPC servers
-    - ModelMetrics: Model performance metrics and drift detection
-    - API: PI Web API export operations for industrial systems
+    MLflow connectivity: unless ``mlflow_repository`` is injected (tests only), this class builds
+    ``SientiaMLflowRepository`` from ``build_mlflow_config()`` so tracking credentials and URL
+    stay aligned with the rest of Laborious env-based configuration.
 
     Attributes:
-        postgres_config (dict): PostgreSQL connection configuration
-        mlflow_config (dict): MLFlow server configuration
-        opc_config (dict): OPC server configuration
-        pi_web_api_config (dict): PI Web API server configuration
-        logger (Logger): Logging and observability instance
-        notification_handler (NotificationHandler): Notification management instance
+        Inherits and exposes behaviour from mixins; the MLFlow mixin holds ``mlflow_repository``
+        and ``plugin_store`` after ``__init__``.
     """
 
     def __init__(
         self,
         postgres_config: dict[str, Any],
-        mlflow_config: dict[str, Any],
+        plugin_store: PluginStore,
         minio_config: dict[str, Any],
         opc_config: dict[str, Any],
         pi_web_api_config: dict[str, Any],
         logger: Logger,
         notification_handler: NotificationHandler,
+        metrics_controller: MetricsController | None = None,
+        mlflow_repository: SientiaMLflowRepository | None = None,
     ):
         """
-        Initialize the Activities orchestrator with all required configurations.
+        Wire Postgres, MinIO, MLflow, OPC, gates, metrics, and PI Web API into a single object.
 
-        This constructor initializes all parent classes with their respective
-        configurations and sets up the foundation for all activity operations.
+        A single ``MetricsController`` instance is created (or reused) and passed to MinIO,
+        MLflow repository, and all mixins so Prometheus and SDK metrics stay consistent.
 
         Args:
-            postgres_config: PostgreSQL connection configuration dictionary
-                Required keys: host, port, user, password, dbname, min_connections, max_connections
-            mlflow_config: MLFlow server configuration dictionary
-                Required keys: host, port, username, password
-            opc_config: OPC server configuration dictionary
-                Can contain multiple server configurations
-            pi_web_api_config: PI Web API server configuration dictionary
-                Required keys: base_url, auth_type, auth_token
-            logger: Logger instance for observability and debugging
-            notification_handler: Notification handler for alerts and monitoring
+         - postgres_config: Host, port, credentials, db name, and pool bounds for Storage.
+         - plugin_store: ``PluginStore`` instance; the worker must call ``install_runtime`` before
+           activities run so wrapper code is importable.
+         - minio_config: Endpoint, keys, bucket, retention, and TLS flag for object storage payloads.
+         - opc_config: Map of OPC server id to connection settings for ``OPC`` mixin.
+         - pi_web_api_config: Base URL and auth for ``API`` mixin.
+         - logger: Structured logger used across all activities.
+         - notification_handler: Handler for alerts and persisted notifications.
+         - metrics_controller: Optional shared controller; if ``None``, a new one is created.
+         - mlflow_repository: Optional ``SientiaMLflowRepository`` for unit/e2e tests; in production
+           leave unset so the repository is built from environment via ``build_mlflow_config()``.
 
         Raises:
-            Exception: If any parent class initialization fails
+            Exception: If any parent ``__init__`` fails (e.g. invalid config keys).
+
+        Return:
+            None
         """
-        metrics_controller = MetricsController(logger=logger)
+
+        mc = metrics_controller or MetricsController(logger=logger)
+
+        # Production path: one shared MLflow client for all model registry / tracking calls.
+        if mlflow_repository is None:
+            mlflow_cfg = build_mlflow_config()
+            mlflow_repository = SientiaMLflowRepository(
+                host=mlflow_cfg['url'],
+                username=mlflow_cfg['username'],
+                password=mlflow_cfg['password'],
+                logger=logger,
+                notification_handler=notification_handler,
+                metrics_controller=mc,
+            )
 
         minio_repository = MinioRepository(
             endpoint=minio_config['endpoint_url'],
@@ -81,11 +97,10 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
             bucket=minio_config['default_bucket'],
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
             secure=minio_config['secure'],
         )
 
-        # Initialize parent classes
         Storage.__init__(
             self,
             host=postgres_config['host'],
@@ -99,19 +114,17 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
             minio_repository=minio_repository,
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
         MLFlow.__init__(
             self,
-            mlflow_host=mlflow_config['host'],
-            mlflow_port=mlflow_config['port'],
-            mlflow_username=mlflow_config['username'],
-            mlflow_password=mlflow_config['password'],
+            mlflow_repository=mlflow_repository,
+            plugin_store=plugin_store,
             minio_repository=minio_repository,
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
         Gates.__init__(
@@ -119,7 +132,7 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
             minio_repository=minio_repository,
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
         OPC.__init__(
@@ -127,14 +140,14 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
             opc_servers=opc_config,
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
         ModelMetrics.__init__(
             self,
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
         API.__init__(
@@ -144,22 +157,18 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
             auth_token=pi_web_api_config['auth_token'],
             logger=logger,
             notification_handler=notification_handler,
-            metrics_controller=metrics_controller,
+            metrics_controller=mc,
         )
 
     async def shutdown(self):
         """
-        Gracefully shutdown all activities and clean up resources.
+        Close database pools, sync clients, and OPC sessions in a defined order.
 
-        This method ensures proper cleanup of all resources including:
-        - PostgreSQL connection pools
-        - OPC server connections
-        - PI Web API client connections
-        - MLFlow model repositories
-        - Any other resources that need explicit cleanup
+        Should be invoked on worker exit so connection pools and OPC sessions are released
+        cleanly before process termination.
 
-        The method should be called before the application terminates to ensure
-        proper resource cleanup and prevent resource leaks.
+        Return:
+            None
         """
         Storage.close(self)
         MLFlow.close(self)

laborious/activities/mlflow.py

@@ -1,11 +1,18 @@
 from temporalio import activity, workflow
 
 with workflow.unsafe.imports_passed_through():
+    import tempfile
     import traceback
+    from datetime import datetime
+    from pathlib import Path
+    from shutil import rmtree
     from typing import Any
 
+    import mlflow
     import numpy as np
-    from pandas import to_datetime
+    import pandas as pd
+    from pandas import DataFrame, to_datetime
+    from sklearn.model_selection import train_test_split
     from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
     from sientia_do.notifications.models import NotificationLevel
     from sientia_do.observability.logger import Logger
@@ -18,79 +25,71 @@ with workflow.unsafe.imports_passed_through():
         now,
     )
     from sientia_do.utils.formatters import create_sample_dict
+    from sientia_model.model_repository.mlflow_repository import SientiaMLflowRepository
+    from sientia_model.model_repository.plugin_store import PluginStore
 
     from laborious.utils.dataframe_debug import build_dataframe_debug_message
     from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
     from laborious.utils.repository.minio_manager import MinioManager
-    from laborious.utils.repository.model_repository import MLFlowRepository
 
 
 class MLFlow(MinioManager):
     """
-    MLFlow integration activities for model inference operations.
+    Temporal activities that talk to MLflow through ``SientiaMLflowRepository`` and ``SientiaModel`` wrappers.
 
-    This class provides activities for interacting with MLFlow models, including
-    data transformation and prediction operations. It handles authentication,
-    data preprocessing, and model management with configurable retention policies.
+    Models are resolved by registered name and the ``production`` alias (not by legacy stages or
+    separate transform/predict flavors). ``get_cached_model`` loads or reuses a wrapper; inference
+    uses ``wrapper.transform`` / ``wrapper.predict``; retrain uses ``wrapper.retrain`` or
+    ``wrapper.train`` plus ``store_model`` and registry promotion via ``promote_to_alias``.
 
-    The class implements comprehensive error handling and logging for all
-    MLFlow operations, ensuring reliable model inference in production environments.
+    Large inputs and outputs flow through ``MinioDataFramePayload`` when workflows offload parquet
+    to MinIO. On failure, transform/predict still return a payload with ``success: False`` and
+    error details for downstream gates.
 
     Attributes:
-        mlflow_host (str): MLFlow server hostname
-        mlflow_port (int): MLFlow server port
-        mlflow_username (str): MLFlow authentication username
-        mlflow_password (str): MLFlow authentication password
-        model_monitoring_repository (MLFlowRepository): Repository for MLFlow operations
+        mlflow_repository: Client for tracking, registry, artifact download, and run lifecycle.
+        plugin_store: Reference to the store (runtime is installed on the worker; reserved for
+            future store-backed helpers).
     """
 
     _MAX_DEBUG_DATAFRAME_ROWS = 100
 
     def __init__(
         self,
-        mlflow_host: str,
-        mlflow_port: int,
-        mlflow_username: str,
-        mlflow_password: str,
+        mlflow_repository: SientiaMLflowRepository,
+        plugin_store: PluginStore,
         minio_repository: MinioRepository | None = None,
         logger: Logger | None = None,
         notification_handler: NotificationHandler | None = None,
         metrics_controller: MetricsController | None = None,
     ):
         """
-        Initialize MLFlow activities with server configuration.
+        Attach shared MLflow and MinIO clients used by all ML activities in this mixin.
 
         Args:
-            mlflow_host: MLFlow server hostname or IP address
-            mlflow_port: MLFlow server port number
-            mlflow_username: Username for MLFlow authentication
-            mlflow_password: Password for MLFlow authentication
-            logger: Logger instance for observability and debugging
-            notification_handler: Notification handler for alerts and monitoring
+         - mlflow_repository: Repository built by ``Activities`` (or injected in tests).
+         - plugin_store: Plugin store instance from worker bootstrap.
+         - minio_repository: MinIO client for ``MinioDataFramePayload`` upload/download.
+         - logger: Structured logger.
+         - notification_handler: Notifications on hard failures where applicable.
+         - metrics_controller: Shared metrics controller.
 
-        Raises:
-            Exception: If MLFlowRepository initialization fails
+        Return:
+            None
         """
+
         MinioManager.__init__(
             self, minio_repository, logger, notification_handler, metrics_controller
         )
-        self.mlflow_host = mlflow_host
-        self.mlflow_port = mlflow_port
-        self.mlflow_username = mlflow_username
-        self.mlflow_password = mlflow_password
-
-        self.model_monitoring_repository = MLFlowRepository(
-            f'{mlflow_host}:{mlflow_port}',
-            mlflow_username,
-            mlflow_password,
-            logger,
-            notification_handler,
-            metrics_controller,
-        )
+        self.mlflow_repository = mlflow_repository
+        self.plugin_store = plugin_store
 
     def close(self) -> None:
         """
-        Close the MLFlow activity and clean up resources.
+        Release MinIO manager resources held by the mixin.
+
+        Return:
+            None
         """
         MinioManager.close(self)
 
@@ -115,35 +114,100 @@ class MLFlow(MinioManager):
             metadata,
         )
 
+    def _detect_and_parse_datetime_index(self, data: pd.DataFrame, metadata: dict) -> pd.DataFrame:
+        """
+        Ensure the transform output index is homogeneous and encoded as ``DATETIME_FORMAT_WITH_TZ`` strings.
+
+        Accepts an all-string index (validated against the format), or all-``datetime`` /
+        ``Timestamp`` (naive timestamps are localized to UTC before formatting). Mixed element types
+        or unsupported types raise ``ValueError`` with a message logged at info level.
+
+        Args:
+         - data: DataFrame whose index carries the time dimension after transform.
+         - metadata: Workflow metadata for log correlation.
+
+        Return:
+            ``pd.DataFrame``: Same frame with a normalized string index; empty frames are returned as-is.
+        """
+
+        if data.empty:
+            self.info('Data is empty, skipping datetime index detection and parsing', metadata)
+            return data
+
+        index = data.index
+        index_type = type(index[0])
+
+        self.info(f'Index type: {index_type}', metadata)
+
+        message = (
+            f'Index must be all timestamp like column. Valid formats are: pandas Timestamp, datetime, '
+            f'string in format {DATETIME_FORMAT_WITH_TZ}.'
+        )
+
+        if not all(isinstance(i, index_type) for i in index):
+            types = map(str, map(type, index))
+            raise ValueError(f'{message}. Elements are {",".join(types)}')
+
+        if index_type is str:
+            try:
+                pd.to_datetime(data.index, format=DATETIME_FORMAT_WITH_TZ)
+            except ValueError as e:
+                raise ValueError(f'{message}. Unable to parse given date format: {e}') from e
+
+        elif index_type is datetime or index_type is pd.Timestamp:
+            idx = data.index
+            if hasattr(idx, 'tz') and idx.tz is None:
+                data.index = idx.tz_localize('UTC')  # type: ignore[attr-defined]
+
+            data.index = data.index.strftime(DATETIME_FORMAT_WITH_TZ)  # type: ignore[attr-defined]
+        else:
+            raise ValueError(f'{message}. Got {index_type}.')
+
+        return data
+
+    def _resolve_model_version_for_run(self, run_id: str) -> str:
+        """
+        Map an MLflow ``run_id`` to the latest registered model version that produced that run.
+
+        ``search_model_versions`` may return multiple versions if the model was registered more than
+        once for the same run; the highest numeric ``version`` wins so promotion targets the newest
+        artifact set.
+
+        Args:
+         - run_id: Run UUID from ``retrain_model`` / experiment payload.
+
+        Return:
+            str: Registry version string acceptable by ``promote_to_alias``.
+
+        Raises:
+            ValueError: If the filter returns no versions (model not registered for this run).
+        """
+
+        versions = self.mlflow_repository._client.search_model_versions(
+            filter_string=f"run_id='{run_id}'"
+        )
+        if not versions:
+            raise ValueError(f'No registered model version found for run_id={run_id}')
+        latest = max(versions, key=lambda v: int(v.version))
+        return str(latest.version)
+
     @activity.defn(name='request_transform')
     async def request_transform(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
         """
-        Transform input data using MLFlow models.
+        Pivot long-format sensor rows, load the production wrapper, and run ``wrapper.transform``.
 
-        This activity processes input data through MLFlow model transformation,
-        including data preprocessing, format conversion, and validation. It handles
-        data deduplication, pivoting, and cleanup to ensure optimal model performance.
-
-        The transformation process includes:
-        1. Data deduplication based on variable and timestamp
-        2. Data pivoting for model input format
-        3. Null value handling and cleanup
-        4. MLFlow model transformation request
-        5. Response validation and logging
+        Expected tabular shape after load: columns including ``variable``, ``timestamp``, ``value``,
+        and ``created_at`` for deduplication. Data are sorted by ``created_at``, de-duplicated per
+        ``(variable, timestamp)``, pivoted wide, then passed to the model. ``model_config`` may
+        include ``retention_minutes`` for wrapper cache TTL.
 
         Args:
-            input_data: Configuration and data for transformation
-                Required keys:
-                    - metadata (dict): Workflow execution metadata
-                    - data (dict): Input data for transformation
-                    - model_name (str): Name of the MLFlow model to use
-                    - model_retention (int): Model retention period in minutes
+         - input_data: Dict with ``metadata``, ``model_name``, ``data`` (``MinioDataFramePayload``
+           dict or inline dataframe dict), and optional ``model_config``.
 
-        Returns:
-            dict: Transformed data from MLFlow model
-
-        Raises:
-            Exception: If transformation fails or MLFlow model is unavailable
+        Return:
+            ``MinioDataFramePayload`` with transformed frame and ``success: True``, or a payload
+            with ``success: False`` and exception details in ``status`` if transform fails.
         """
         metadata = input_data['metadata']
         self.info('Transforming data...', metadata)
@@ -156,12 +220,11 @@ class MLFlow(MinioManager):
 
         self._debug_dataframe('Raw input data:', data, metadata)
 
-        # Sort by created_at in descending order and keep first occurrence of each variable/timestamp pair
+        # Long → wide: keep newest row per (variable, timestamp), then pivot for the wrapper API.
         data = data.sort_values('created_at', ascending=False).drop_duplicates(
             subset=['variable', 'timestamp'], keep='first'
         )
 
-        # Pivot data for model input format
         data = data.pivot(index='timestamp', columns='variable', values='value')
         data.fillna(np.nan, inplace=True)
 
@@ -172,10 +235,25 @@ class MLFlow(MinioManager):
 
         self._debug_dataframe('Processed input data:', data, metadata)
 
-        # Request transformation from MLFlow model
-        response_data = await self.model_monitoring_repository.transform(
-            model_name, data, model_config, metadata
-        )
+        try:
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias='production',
+                retention_minutes=model_config.get('retention_minutes', 0),
+                metadata=metadata,
+            )
+            transformed_df, transform_meta = wrapper.transform(data)
+            if transform_meta:
+                self.info(f'Wrapper transform metadata: {transform_meta}', metadata)
+
+            transformed_df = self._detect_and_parse_datetime_index(transformed_df, metadata)
+
+            response_data: dict[str, Any] = {'success': True, 'content': transformed_df}
+        except Exception as e:
+            response_data = {
+                'success': False,
+                'content': {'message': str(e), 'traceback': traceback.format_exc()},
+            }
 
         self.debug(
             f'Transform raw response data: \n {create_sample_dict(response_data, max_items=5, max_depth=5)}',
@@ -217,32 +295,19 @@ class MLFlow(MinioManager):
     @activity.defn(name='request_predict')
     async def request_predict(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
         """
-        Execute predictions using MLFlow models.
+        Load the production wrapper and call ``wrapper.predict`` on the prepared feature frame.
 
-        This activity performs ML model inference using MLFlow models with the
-        transformed data. It handles data format conversion, null value processing,
-        and model prediction requests with comprehensive error handling.
-
-        The prediction process includes:
-        1. Data format validation and cleanup
-        2. Null value handling for model compatibility
-        3. MLFlow model prediction request
-        4. Response validation and logging
-        5. Performance monitoring and metrics
+        The activity normalizes ``NaN`` to ``None`` for JSON-friendly columns, rebuilds a
+        ``timestamp`` column in the internal string format, preserves the original index for
+        alignment, and records ``response_time`` seconds on the output frame. Non-DataFrame
+        predictions are coerced to a single ``prediction`` column.
 
         Args:
-            input_data: Configuration and data for prediction
-                Required keys:
-                    - metadata (dict): Workflow execution metadata
-                    - data (dict): Transformed data for prediction
-                    - model_name (str): Name of the MLFlow model to use
-                    - model_retention (int): Model retention period in minutes
+         - input_data: Same envelope as ``request_transform`` (``metadata``, ``model_name``,
+           ``data``, optional ``model_config`` with ``retention_minutes``).
 
-        Returns:
-            dict: Prediction results from MLFlow model
-
-        Raises:
-            Exception: If prediction fails or MLFlow model is unavailable
+        Return:
+            ``MinioDataFramePayload`` with predictions or error status mirroring transform behaviour.
         """
         metadata = input_data['metadata']
         self.info('Predicting data...', metadata)
@@ -255,7 +320,8 @@ class MLFlow(MinioManager):
 
         self._debug_dataframe('Input data for prediction:', data, metadata)
 
-        # Convert numpy.nan to None for model compatibility
+        input_index = data.index
+
         data.replace(np.nan, None, inplace=True)
 
         data['timestamp'] = data.index
@@ -263,10 +329,41 @@ class MLFlow(MinioManager):
             data['timestamp'], format=DATETIME_FORMAT_WITH_TZ
         ).dt.strftime(DATETIME_FORMAT)
 
-        # Request prediction from MLFlow model
-        response_data = await self.model_monitoring_repository.predict(
-            model_name, data, model_config, metadata
-        )
+        try:
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias='production',
+                retention_minutes=model_config.get('retention_minutes', 0),
+                metadata=metadata,
+            )
+            start_time = datetime.now()
+            predict_data, pred_meta = wrapper.predict({}, data)
+            end_time = datetime.now()
+
+            if pred_meta:
+                self.info(f'Wrapper predict metadata: {pred_meta}', metadata)
+
+            if isinstance(predict_data, DataFrame):
+                self._debug_dataframe(
+                    'Data received from model prediction:', predict_data, metadata
+                )
+                predict_data.columns = pd.Index(['prediction'])
+            else:
+                self.debug(
+                    f'Data received from model prediction (not a DataFrame): {predict_data}',
+                    metadata,
+                )
+                predict_data = pd.DataFrame(predict_data, columns=['prediction'])
+
+            predict_data.index = input_index
+            predict_data['response_time'] = (end_time - start_time).total_seconds()
+
+            response_data: dict[str, Any] = {'success': True, 'content': predict_data}
+        except Exception as e:
+            response_data = {
+                'success': False,
+                'content': {'message': str(e), 'traceback': traceback.format_exc()},
+            }
 
         self.debug(
             f'Prediction response data: \n {create_sample_dict(response_data, max_items=5, max_depth=5)}',
@@ -303,34 +400,22 @@ class MLFlow(MinioManager):
     @activity.defn(name='retrain_model')
     async def retrain_model(self, input_data: dict[str, Any]) -> dict[str, Any]:
         """
-        Retrain MLFlow models with updated training data.
+        Fit an updated wrapper from historical data, log a new run, and register a model version.
 
-        This activity orchestrates the complete model retraining process,
-        including data preparation, model retraining execution, and result
-        validation. It handles data preprocessing, column cleanup, and
-        comprehensive error handling for production model management.
-
-        The retraining process includes:
-        1. Data timestamp extraction and validation
-        2. Column cleanup and data preparation
-        3. Data pivoting for model input format
-        4. MLFlow model retraining execution
-        5. Result validation and error handling
+        Flow: load long-format data from MinIO → dedupe/pivot like inference prep → require
+        ``model_config['target']`` → read current ``production`` version for ``source_run_id`` tag →
+        ``start_run`` with retrain tags → ``wrapper.retrain`` or ``wrapper.train`` when
+        ``full_retrain`` is set (optional ``validation_fraction``) → log input CSV artifact →
+        ``store_model`` and ``log_params``. Does not promote; the workflow calls
+        ``update_production_model`` after validation.
 
         Args:
-            input_data (dict): Input data containing:
-                - metadata (dict): Workflow execution metadata
-                - data (dict[str, Any]): Training data for model retraining
-                - model_name (str): Name of the MLFlow model to retrain
+         - input_data: Must include ``metadata``, ``model_name``, ``data`` (payload), and
+           ``model_config`` with at least ``target``; optional ``full_retrain``, ``validation_fraction``.
 
-        Returns:
-            dict: Retraining results containing:
-                - status (str): Retraining operation status
-                - timestamp (str): Timestamp of the retraining operation
-                - experiment (str): MLFlow experiment identifier
-
-        Raises:
-            Exception: If retraining fails or encounters critical errors
+        Return:
+            On success: ``success``, ``experiment`` (``run_id``, ``experiment_id``, ``experiment_name``),
+            ``message``, ``timestamp``. On failure: ``success: False``, error fields, and optional trace.
         """
 
         if self.minio_repository is None:
@@ -339,7 +424,6 @@ class MLFlow(MinioManager):
         metadata = input_data['metadata']
 
         try:
-            # Payload-based retrain input (inline dict or MinIO offloaded).
             payload = MinioDataFramePayload.from_dict(input_data['data'])
             data = await payload.retrieve(self.minio_repository, metadata)
 
@@ -371,7 +455,6 @@ class MLFlow(MinioManager):
         timestamp = data['timestamp'].max()
         self.debug(f'Timestamp: {timestamp}', metadata)
 
-        # Sort by created_at in descending order and keep first occurrence of each variable/timestamp pair
         if 'created_at' in data.columns:
             data = data.sort_values('created_at', ascending=False).drop_duplicates(
                 subset=['variable', 'timestamp'], keep='first'
@@ -382,10 +465,8 @@ class MLFlow(MinioManager):
         data.drop(columns=['model_id'], inplace=True, errors='ignore')
         data.drop(columns=['created_at'], inplace=True, errors='ignore')
 
-        # Pivot data for model input format
         data = data.pivot(index='timestamp', columns='variable', values='value')
         data.fillna(np.nan, inplace=True)
-        # data.reset_index(inplace=True)
         data.columns.name = None
 
         data['timestamp'] = data.index
@@ -396,60 +477,110 @@ class MLFlow(MinioManager):
 
         data.columns.name = None
 
-        retrain_output = await self.model_monitoring_repository.retrain_model(
-            data=data, model_name=model_name, model_config=model_config, metadata=metadata
-        )
+        target = model_config.get('target')
+        if target is None:
+            msg = 'model_config must include "target" for retraining'
+            self.info(msg, metadata)
+            return {
+                'success': False,
+                'experiment': None,
+                'message': msg,
+                'traceback': '',
+                'timestamp': str(timestamp),
+            }
 
-        if not retrain_output['success']:
-            trace = retrain_output['traceback']
-            await self.send_notification_async(
-                metadata=metadata,
-                notification_id='RETRAIN_MODEL_ERROR',
-                message=f'Error retraining model {model_name}: {retrain_output["message"]}',
-                block='retrain_model',
-                level=NotificationLevel.ERROR,
-                attachment_content=trace,
+        try:
+            mv_src = self.mlflow_repository._client.get_model_version_by_alias(
+                name=model_name,
+                alias='production',
             )
-            self.error(trace, metadata=metadata)
+            source_run_id = mv_src.run_id
 
-        return {**retrain_output, 'timestamp': timestamp}
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias='production',
+                retention_minutes=0,
+                metadata=metadata,
+            )
+
+            run_name = f'{model_name}-retrain-{datetime.now().strftime("%Y%m%d%H%M%S")}'
+
+            with self.mlflow_repository.start_run(
+                model_name=model_name,
+                run_name=run_name,
+                experiment_name=model_name,
+                tags={'retrain': 'true', 'source_run_id': source_run_id},
+                metadata=metadata,
+            ) as run_info:
+                if model_config.get('full_retrain'):
+                    val_frac = float(model_config.get('validation_fraction', 0.2))
+                    train_df, val_df = train_test_split(data, test_size=val_frac, random_state=42)
+                    wrapper.train(
+                        train_data=train_df,
+                        val_data=val_df,
+                        target=target,
+                    )
+                else:
+                    wrapper.retrain(data)
+
+                tmp_dir = tempfile.mkdtemp(prefix='laborious_retrain_')
+                try:
+                    raw_csv = Path(tmp_dir) / 'retrain_input.csv'
+                    data.to_csv(raw_csv, index=False)
+                    mlflow.log_artifact(str(raw_csv))
+                finally:
+                    rmtree(tmp_dir, ignore_errors=True)
+
+                wrapper.store_model(name=model_name)
+
+                self.mlflow_repository.log_params(
+                    {
+                        'retrain': 'true',
+                        'retrain_date': datetime.now().isoformat(),
+                        'source_run_id': source_run_id,
+                        'retrain_samples': str(data.shape),
+                    }
+                )
+
+                experiment_payload = {
+                    'run_id': run_info.run_id,
+                    'experiment_id': run_info.experiment_id,
+                    'experiment_name': model_name,
+                }
+
+            return {
+                'success': True,
+                'experiment': experiment_payload,
+                'message': 'Model retrained successfully.',
+                'timestamp': str(timestamp),
+            }
+
+        except Exception as e:
+            error_msg = f'Error retraining model {model_name}: {e}'
+            self.info(error_msg, metadata)
+            return {
+                'success': False,
+                'experiment': None,
+                'message': error_msg,
+                'traceback': traceback.format_exc(),
+                'timestamp': str(timestamp),
+            }
 
     @activity.defn(name='update_production_model')
     async def update_production_model(self, input_data: dict[str, Any]) -> dict[Any, Any]:
         """
-        Update production model with newly trained model version.
+        Point the ``production`` alias at the model version registered for the retrain run.
 
-        This activity manages the critical process of updating production
-        models with newly trained versions. It handles model deployment,
-        status tracking, and comprehensive reporting for operational
-        visibility and audit trails.
-
-        The update process includes:
-        1. Production model update execution
-        2. Status and metadata tracking
-        3. Comprehensive reporting and logging
-        4. Error handling and notification
-        5. Audit trail maintenance
+        Resolves the highest numeric registry version whose ``run_id`` matches
+        ``experiment['run_id']``, then calls ``promote_to_alias``. On failure, sends a notification
+        and re-raises so the workflow can surface the error.
 
         Args:
-            input_data (dict): Input data containing:
-                - metadata (dict): Workflow execution metadata
-                - model_name (str): Name of the MLFlow model to update
-                - experiment (str): MLFlow experiment identifier
-                - model_id (str): Unique identifier for the model version
-                - timestamp (str): Timestamp of the update operation
-                - status (str): Current status of the model update
+         - input_data: ``metadata``, ``model_name``, and ``experiment`` with ``run_id`` and
+           ``experiment_id`` (as returned from ``retrain_model``).
 
-        Returns:
-            dict[Any, Any]: Comprehensive update report containing:
-                - model_id (str): Model version identifier
-                - model_name (str): Name of the updated model
-                - timestamp (str): Update operation timestamp
-                - status (str): Update operation status
-                - Additional MLFlow response metadata
-
-        Raises:
-            Exception: If production model update fails
+        Return:
+            Dict with ``model_name``, promoted ``version``, ``mlflow_run_id``, ``mlflow_experiment_id``.
         """
         metadata = input_data['metadata']
         model_name = input_data['model_name']
@@ -459,12 +590,25 @@ class MLFlow(MinioManager):
         )
 
         try:
-            response = await self.model_monitoring_repository.update_production_model(
-                experiment=experiment, model_name=model_name, metadata=metadata
+            run_id = experiment['run_id']
+            experiment_id = experiment['experiment_id']
+
+            version = self._resolve_model_version_for_run(run_id)
+
+            self.mlflow_repository.promote_to_alias(
+                model_name=model_name,
+                version=version,
+                alias='production',
+                metadata=metadata,
             )
 
             self.info(f'Production model {model_name} updated successfully', metadata)
-            return response
+            return {
+                'model_name': model_name,
+                'version': version,
+                'mlflow_run_id': run_id,
+                'mlflow_experiment_id': experiment_id,
+            }
 
         except Exception as e:
             trace = traceback.format_exc()
@@ -482,47 +626,51 @@ class MLFlow(MinioManager):
     @activity.defn(name='get_reference_data')
     async def get_reference_data(self, input_data: dict[str, Any]) -> list[dict] | None:
         """
-        Get reference data from the MLflow Model Registry.
+        Download ``evaluation_data.csv`` from the MLflow run linked to ``production`` and parse it.
 
-        This method retrieves evaluation reference data stored as artifacts in the
-        MLflow Model Registry. The reference data is typically used for model
-        drift detection, performance comparison, and quality validation. The method
-        loads the data from a CSV artifact file and formats timestamps for
-        consistent processing.
-
-        The method handles:
-        1. Loading evaluation data artifact from MLflow Model Registry
-        2. Timestamp parsing and formatting for consistency
-        3. Data conversion to dictionary format for workflow consumption
-        4. Graceful handling of missing reference data
+        Used by drift workflows to compare live data against the reference distribution logged with
+        the model. Artifacts are downloaded to a temp directory, discovered via ``rglob`` (nested
+        layout-safe), then timestamps are normalized to ``DATETIME_FORMAT`` string columns before
+        returning record-oriented dicts.
 
         Args:
-            input_data (dict): Input data containing:
-                - metadata (dict): Workflow execution metadata
-                - model_name (str): Name of the MLFlow model to get reference data from
+         - input_data: ``metadata`` and ``model_name`` for registry lookup.
 
-        Returns:
-            list[dict[Hashable, Any]] | None: Reference data from the MLflow Model Registry
-                as a list of dictionaries. Returns None if reference data is not found
-                or if the artifact does not exist.
-
-        Raises:
-            Exception: If artifact loading fails or encounters errors during processing
+        Return:
+            List of row dicts, or ``None`` if the artifact path is missing or any step fails.
         """
-
         metadata = input_data['metadata']
         model_name = input_data['model_name']
-        artifact = 'evaluation_data.csv'
 
-        reference_data = await self.model_monitoring_repository.load_artifact_dataframe(
-            model_name=model_name, artifact_path=artifact, metadata=metadata
-        )
+        try:
+            mv = self.mlflow_repository._client.get_model_version_by_alias(
+                name=model_name,
+                alias='production',
+            )
+            run_id = mv.run_id
 
-        if reference_data is None:
-            self.warning(f'Reference data not found for model {model_name}', metadata)
+            tmpdir = tempfile.mkdtemp(prefix='laborious_eval_')
+            try:
+                self.mlflow_repository.download_artifacts(
+                    run_id=run_id,
+                    artifact_path='evaluation_data.csv',
+                    dst_path=tmpdir,
+                    metadata=metadata,
+                )
+                csv_candidates = list(Path(tmpdir).rglob('evaluation_data.csv'))
+                if not csv_candidates:
+                    self.warning(f'Reference data not found for model {model_name}', metadata)
+                    return None
+
+                reference_data = pd.read_csv(csv_candidates[0])
+            finally:
+                rmtree(tmpdir, ignore_errors=True)
+
+            reference_data['timestamp'] = to_datetime(reference_data['timestamp'])
+            reference_data['timestamp'] = reference_data['timestamp'].dt.strftime(DATETIME_FORMAT)
+
+            return reference_data.to_dict(orient='records')
+
+        except Exception as e:
+            self.warning(f'Reference data not found for model {model_name}: {e}', metadata)
             return None
-
-        reference_data['timestamp'] = to_datetime(reference_data['timestamp'])
-        reference_data['timestamp'] = reference_data['timestamp'].dt.strftime(DATETIME_FORMAT)
-
-        return reference_data.to_dict(orient='records')

laborious/utils/connectors_config.py

@@ -3,31 +3,88 @@ from os import getenv
 from typing import Any
 
 
+def _build_mlflow_tracking_url() -> str:
+    """
+    Compose a single tracking URI for ``SientiaMLflowRepository`` from host and port env vars.
+
+    If ``MLFLOW_HOST`` already contains a port in the authority (e.g. ``http://tracker:80``),
+    it is returned unchanged so operators can override port logic explicitly.
+
+    Return:
+        str: Full tracking URL (scheme + host [+ port]).
+    """
+
+    mlflow_host = getenv('MLFLOW_HOST', 'http://localhost').rstrip('/')
+    mlflow_port = getenv('MLFLOW_PORT', '5080')
+
+    # Host already includes an explicit port (e.g. http://tracker:80)
+    host_after_scheme = mlflow_host.split('://', 1)[-1]
+    if ':' in host_after_scheme:
+        return mlflow_host
+
+    return f'{mlflow_host}:{mlflow_port}'
+
+
 def build_mlflow_config() -> dict[str, Any]:
     """
-    Build MLFlow server configuration from environment variables.
+    Read MLflow tracking and registry credentials from the environment.
 
-    This function constructs an MLFlow configuration dictionary from
-    environment variables with sensible defaults for local development.
-    It handles server connection and authentication parameters.
+    Used by ``Activities`` when constructing ``SientiaMLflowRepository``. The ``url`` value is the
+    same string workers and notebooks should use for ``MLFLOW_TRACKING_URI``-style clients.
 
     Environment Variables:
-        MLFLOW_HOST: MLFlow server hostname (default: http://localhost)
-        MLFLOW_PORT: MLFlow server port (default: 5080)
-        MLFLOW_USERNAME: MLFlow username (default: aignosi)
-        MLFLOW_PASSWORD: MLFlow password (default: aignosi)
+        MLFLOW_HOST: Host with scheme; port optional if MLFLOW_PORT is set (default: http://localhost)
+        MLFLOW_PORT: Appended when host has no explicit port (default: 5080)
+        MLFLOW_USERNAME: Basic-auth or service user (default: aignosi)
+        MLFLOW_PASSWORD: Password or token (default: aignosi)
 
-    Returns:
-        dict: MLFlow configuration dictionary with all required parameters
+    Return:
+        dict[str, Any]: ``url``, ``username``, ``password``.
     """
     return {
-        'host': getenv('MLFLOW_HOST', 'http://localhost'),
-        'port': int(getenv('MLFLOW_PORT', '5080')),
+        'url': _build_mlflow_tracking_url(),
         'username': getenv('MLFLOW_USERNAME', 'aignosi'),
         'password': getenv('MLFLOW_PASSWORD', 'aignosi'),
     }
 
 
+def build_plugin_store_config() -> dict[str, Any]:
+    """
+    Collect settings for ``PluginStore`` (Git-backed catalog + runtime install via pip).
+
+    Mirrors the model-manager service: the worker passes these kwargs into ``PluginStore`` after
+    ``install_runtime`` resolves wheels from the configured PyPI index. Missing optional env vars
+    become ``None`` so the store can run without auth in local dev.
+
+    Environment Variables:
+        STORE_BASE_URL: Git HTTP(S) server (e.g. Gitea) base URL (default: http://localhost:3000)
+        STORE_OWNER: Namespace or org owning the store repo (default: sientia)
+        STORE_REPO: Repository name (default: model-library-store)
+        STORE_BRANCH: Checkout branch; unset lets the client use default
+        STORE_USERNAME / STORE_PASSWORD: HTTP basic credentials for Git fetch
+        STORE_CACHE_TTL_SECONDS: Optional integer seconds for metadata cache TTL
+        PYPI_SERVER: Index URL for ``pip install`` during runtime install (default: http://localhost:5000)
+        PYPI_USERNAME / PYPI_PASSWORD: Optional index authentication
+
+    Return:
+        dict[str, Any]: Keys aligned with ``PluginStore`` constructor parameter names.
+    """
+
+    cache_ttl_seconds = getenv('STORE_CACHE_TTL_SECONDS')
+    return {
+        'base_url': getenv('STORE_BASE_URL', 'http://localhost:3000'),
+        'owner': getenv('STORE_OWNER', 'sientia'),
+        'repo': getenv('STORE_REPO', 'model-library-store'),
+        'username': getenv('STORE_USERNAME'),
+        'password': getenv('STORE_PASSWORD'),
+        'branch': getenv('STORE_BRANCH'),
+        'cache_ttl_seconds': int(cache_ttl_seconds) if cache_ttl_seconds else None,
+        'pypi_index_url': getenv('PYPI_SERVER', 'http://localhost:5000'),
+        'pypi_username': getenv('PYPI_USERNAME'),
+        'pypi_password': getenv('PYPI_PASSWORD'),
+    }
+
+
 def build_opc_config() -> dict[str, Any]:
     """
     Build OPC server configuration from environment variables.
@@ -73,14 +130,15 @@ def build_minio_config() -> dict[str, Any]:
     Build MinIO (S3-compatible) configuration from environment variables.
 
     Environment Variables:
-        MINIO_ENDPOINT: MinIO endpoint including scheme (default: http://localhost:9000)
+        MINIO_ENDPOINT_URL: Host:port or URL for the S3 API (default: http://localhost:9000)
         MINIO_ACCESS_KEY: Access key (default: minioadmin)
         MINIO_SECRET_KEY: Secret key (default: minioadmin)
-        MINIO_REGION: Region name for S3 client (default: us-east-1)
-        MINIO_BUCKET_DEFAULT: Default bucket for uploads (default: laborious)
-        MINIO_SECURE: Whether to use HTTPS (default: false)
-    Returns:
-        dict: MinIO configuration dictionary
+        MINIO_DEFAULT_BUCKET: Default bucket for Laborious payloads (default: laborious)
+        MINIO_RETENTION_HOURS: Offloaded object retention window (default: 24)
+        MINIO_SECURE: If ``true``, use HTTPS (default: false)
+
+    Return:
+        dict[str, Any]: Keys consumed by ``Activities`` / ``MinioRepository``.
     """
     return {
         'endpoint_url': getenv('MINIO_ENDPOINT_URL', 'http://localhost:9000'),

laborious/worker/prepare_worker.py

@@ -1,73 +0,0 @@
-import os
-import re
-from collections.abc import Sequence
-from typing import Any
-
-from sientia_do.observability.logger import Logger
-from temporalio.client import Client
-from temporalio.worker import PollerBehaviorAutoscaling, Worker
-
-parameters = [
-    ('MAX_CONCURRENT_WORKFLOW_TASKS', '200'),
-    ('MAX_CONCURRENT_ACTIVITIES', '200'),
-    ('MAX_CONCURRENT_LOCAL_ACTIVITIES', '200'),
-    ('MAX_CACHED_WORKFLOWS', '200'),
-    ('WORKFLOW_POLLER_BEHAVIOUR_MINIMUM', '10'),
-    ('WORKFLOW_POLLER_BEHAVIOUR_INITIAL', '100'),
-    ('WORKFLOW_POLLER_BEHAVIOUR_MAXIMUM', '200'),
-    ('ACTIVITY_POLLER_BEHAVIOUR_MINIMUM', '10'),
-    ('ACTIVITY_POLLER_BEHAVIOUR_INITIAL', '100'),
-    ('ACTIVITY_POLLER_BEHAVIOUR_MAXIMUM', '200'),
-]
-
-
-def camel_to_snake(text: str) -> str:
-    """Convert camelCase or PascalCase to snake_case."""
-    text = re.sub('(.)([A-Z][a-z]+)', r'\1_\2', text)
-    text = re.sub('([a-z0-9])([A-Z])', r'\1_\2', text)
-    return text.lower()
-
-
-def prepare_worker(
-    main_workflow: type,
-    other_workflows: Sequence[type],
-    activities: Sequence[Any],
-    temporal_client: Client,
-    logger: Logger,
-) -> Worker:
-    main_workflow_name = main_workflow.__name__.upper()
-
-    queue_name = f'{camel_to_snake(main_workflow.__name__)}-queue'
-
-    local_workflow_parameters = {}
-
-    for parameter in parameters:
-        local_workflow_parameters[parameter[0]] = int(
-            os.getenv(main_workflow_name + '_' + parameter[0], parameter[1])
-        )
-
-    logger.info(f'Preparing worker for {main_workflow_name} with queue {queue_name}')
-    logger.info(f'Worker runtime config: {local_workflow_parameters}')
-
-    return Worker(
-        temporal_client,
-        task_queue=queue_name,
-        workflows=[main_workflow, *other_workflows],
-        activities=[*activities],
-        max_concurrent_workflow_tasks=local_workflow_parameters['MAX_CONCURRENT_WORKFLOW_TASKS'],
-        max_concurrent_activities=local_workflow_parameters['MAX_CONCURRENT_ACTIVITIES'],
-        max_concurrent_local_activities=local_workflow_parameters[
-            'MAX_CONCURRENT_LOCAL_ACTIVITIES'
-        ],
-        max_cached_workflows=local_workflow_parameters['MAX_CACHED_WORKFLOWS'],
-        workflow_task_poller_behavior=PollerBehaviorAutoscaling(
-            minimum=local_workflow_parameters['WORKFLOW_POLLER_BEHAVIOUR_MINIMUM'],
-            initial=local_workflow_parameters['WORKFLOW_POLLER_BEHAVIOUR_INITIAL'],
-            maximum=local_workflow_parameters['WORKFLOW_POLLER_BEHAVIOUR_MAXIMUM'],
-        ),
-        activity_task_poller_behavior=PollerBehaviorAutoscaling(
-            minimum=local_workflow_parameters['ACTIVITY_POLLER_BEHAVIOUR_MINIMUM'],
-            initial=local_workflow_parameters['ACTIVITY_POLLER_BEHAVIOUR_INITIAL'],
-            maximum=local_workflow_parameters['ACTIVITY_POLLER_BEHAVIOUR_MAXIMUM'],
-        ),
-    )

laborious/worker/worker.py

@@ -1,32 +1,32 @@
 """
 Laborious Worker Module
 
-This module provides the main worker implementation for the Sientia DataOps Laborious system.
-It orchestrates Temporal workers, manages task queues, and handles the lifecycle of
-prediction and retraining workflows.
+Entry process that connects to Temporal, registers Laborious activities, and runs four workers in
+parallel. Each worker shares the same ``Activities`` instance (single Postgres pool, single MLflow
+repository, single PluginStore handle) but polls a different task queue.
 
-The worker supports multiple task queues:
-- predictions_batch-queue: Handles batch prediction workflows (heavy workload)
-  Includes activities for MLFlow, data quality gates, OPC export, PI Web API export, and PostgreSQL
-- minimal_retrain-queue: Handles model retraining workflows
-- drift-queue: Handles drift detection workflows
-- simple_metrics-queue: Handles simple metrics calculation workflows
+Task queues (see ``sientia_do.temporal.worker.prepare_worker``):
+- ``predictions_batch-{runtime}-queue`` + sub-workflows on the same queue (ML-heavy path).
+- ``minimal_retrain-{runtime}-queue`` (retrain + promote + export).
+- ``drift-queue`` and ``simple_metrics-queue`` without a runtime suffix so existing schedulers
+  keep stable queue names.
 
-Key Features:
-- Resource-based scaling with WorkerTuner (CPU and memory aware)
-- Automatic polling scaling with PollerBehaviorAutoscaling
-- Prometheus metrics integration
-- Comprehensive error handling and logging
-- Graceful shutdown with cleanup
-- Multiple worker instances for different workflow types
+Bootstrap order:
+1. Prometheus app metrics and Mongo-backed notification handler.
+2. ``RUNTIME`` validation and ``PluginStore.install_runtime`` so ``SientiaModel`` code is importable.
+3. ``Activities`` construction (builds ``SientiaMLflowRepository`` internally from env).
+4. OPC client initialization inside activities.
+5. Temporal ``Runtime`` with SDK Prometheus bind, client connect, then ``prepare_worker`` per workflow.
+
+Shutdown closes workers, notifications, activities (pools + OPC), and clears ``app_up``.
 
 Environment Variables:
-- TEMPORAL_HOST: Temporal server address (default: localhost:7233)
-- TEMPORAL_NAMESPACE: Temporal namespace (default: laborious)
-- POD_ID: Kubernetes pod identifier for metrics
-- HTTP_METRICS_PORT: Prometheus metrics server port (default: 9090)
-- HTTP_SDK_METRICS_PORT: Temporal SDK metrics port (default: 9091)
-- PROJECT_NAME: Project name for notifications (default: laborious)
+- RUNTIME: Required non-empty string passed to ``install_runtime``.
+- STORE_* / PYPI_*: Plugin store and private index (see ``build_plugin_store_config``).
+- TEMPORAL_HOST, TEMPORAL_NAMESPACE: Cluster connection.
+- POD_ID, HTTP_METRICS_PORT, HTTP_SDK_METRICS_PORT: Observability.
+- PROJECT_NAME, MONGODB_*: Notifications (via ``build_mongodb_config`` in handler).
+- POSTGRES_*, MINIO_*, OPC_*, PI_WEB_API_*, MLFLOW_*: Passed through ``Activities`` helpers.
 """
 
 from temporalio import client, workflow
@@ -40,20 +40,22 @@ with workflow.unsafe.imports_passed_through():
     from prometheus_client import start_http_server
     from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
     from sientia_do.observability.logger import get_logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.temporal.worker.prepare_worker import prepare_worker
     from sientia_do.utils.connectors_config import (
         build_api_config,
         build_mongodb_config,
         build_postgres_config,
     )
+    from sientia_model.model_repository.plugin_store import PluginStore
 
     from laborious import metrics
     from laborious.activities.activities import Activities
     from laborious.utils.connectors_config import (
         build_minio_config,
-        build_mlflow_config,
         build_opc_config,
+        build_plugin_store_config,
     )
-    from laborious.worker.prepare_worker import prepare_worker
     from laborious.workflows.drift import Drift
     from laborious.workflows.minimal_retrain import MinimalRetrain
     from laborious.workflows.predictions_batch import PredictionsBatch
@@ -69,23 +71,18 @@ SDK_METRICS_PORT = int(os.getenv('HTTP_SDK_METRICS_PORT', '9091'))
 
 async def main():
     """
-    Main entry point for the Laborious worker application.
+    Run the full worker lifecycle: metrics, notifications, runtime install, workers, gather.
 
-    This function initializes and starts all components of the worker:
-    1. Sets up logging and metadata
-    2. Starts Prometheus metrics server
-    3. Initializes notification handler
-    4. Creates and configures activities
-    5. Initializes OPC connections
-    6. Starts Temporal client and workers
-    7. Manages worker lifecycle and graceful shutdown
-
-    The function runs indefinitely until interrupted or an error occurs.
-    On error, it performs cleanup and exits with a non-zero status code.
+    Exits the process with code 0 on normal completion of all worker tasks, or 1 after logging
+    if any worker raises. ``finally`` always shuts down notifications and activities and sets
+    ``app_up`` to 0 before ``sys.exit``.
 
     Raises:
-        Exception: Any unhandled exception during worker execution
-        SystemExit: On graceful shutdown or error conditions
+        Exception: Propagated from ``asyncio.gather`` only before ``finally`` handling; typically
+            workers run until cancelled.
+
+    Return:
+        None (process terminates via ``sys.exit`` from the ``finally`` block).
     """
     host = os.getenv('TEMPORAL_HOST', 'localhost:7233')
     logger = get_logger(__name__)
@@ -113,16 +110,55 @@ async def main():
         project_name=os.getenv('PROJECT_NAME', 'laborious'),
     )
 
+    metrics_controller = MetricsController(logger=logger)
+
+    runtime = os.getenv('RUNTIME', '').strip()
+    if not runtime:
+        logger.custom_critical(
+            'RUNTIME environment variable is required and must be non-empty',
+            metadata,
+        )
+        metrics.APP_UP.labels(pod_id=POD_ID).set(0)
+        sys.exit(1)
+
+    metadata_runtime = {**metadata, 'runtime': runtime}
+    logger.custom_info(f'Installing PluginStore runtime: {runtime}', metadata_runtime)
+
+    ps_cfg = build_plugin_store_config()
+    plugin_store = PluginStore(
+        base_url=ps_cfg['base_url'],
+        owner=ps_cfg['owner'],
+        repo=ps_cfg['repo'],
+        username=ps_cfg['username'],
+        password=ps_cfg['password'],
+        branch=ps_cfg['branch'],
+        cache_ttl_seconds=ps_cfg['cache_ttl_seconds'],
+        pypi_index_url=ps_cfg['pypi_index_url'],
+        pypi_username=ps_cfg['pypi_username'],
+        pypi_password=ps_cfg['pypi_password'],
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=metrics_controller,
+    )
+
+    try:
+        await plugin_store.install_runtime(runtime_name=runtime, metadata=metadata_runtime)
+    except Exception as exc:
+        logger.custom_critical(f'Failed to install runtime {runtime}: {exc}', metadata_runtime)
+        metrics.APP_UP.labels(pod_id=POD_ID).set(0)
+        sys.exit(1)
+
     logger.custom_info('Starting Activities...', metadata)
 
     activities = Activities(
         postgres_config=build_postgres_config(),
-        mlflow_config=build_mlflow_config(),
+        plugin_store=plugin_store,
         minio_config=build_minio_config(),
         opc_config=build_opc_config(),
         pi_web_api_config=build_api_config(),
         logger=logger,
         notification_handler=notification_handler,
+        metrics_controller=metrics_controller,
     )
 
     logger.custom_info('Initializing OPC...', metadata)
@@ -159,6 +195,7 @@ async def main():
                 activities.export_data_to_postgres,
             ],
             logger=logger,
+            runtime=runtime,
         ),
         prepare_worker(
             temporal_client=temporal_client,
@@ -210,6 +247,7 @@ async def main():
                 activities.write_pi_web_api_data,
             ],
             logger=logger,
+            runtime=runtime,
         ),
     ]
 
@@ -221,8 +259,6 @@ async def main():
 
     exit_code = 0
     try:
-        # This will run the workers and wait for them to complete.
-        # If an exception occurs in any of the worker handlers, it will be propagated here.
         await asyncio.gather(*handlers)
     except BaseException as e:  # NOSONAR
         logger.custom_error(f'An unhandled exception occurred: {e}', metadata)