SIENTIAPDE-1646

Update dependencies and refactor MLFlow activities - Replaced direct GitHub dependencies in `requirements.txt` with specific versioned packages for `sientia_do` and `sientia_model`. - Refactored imports in `activities.py` to streamline the code structure. - Enhanced the `MLFlow` class in `mlflow.py` by introducing a method to resolve model aliases, improving flexibility in model lookups. - Simplified shutdown logic in `worker.py` for better readability. - Added new tests for MLFlow activities and improved existing test coverage for data handling and model retraining processes.
2026-05-06 15:31:25 -03:00
parent 1ce8b9d3a7
commit 424be007ef
12 changed files with 625 additions and 29 deletions
--- a/laborious/activities/activities.py
+++ b/laborious/activities/activities.py
@@ -10,14 +10,13 @@ with workflow.unsafe.imports_passed_through():
    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
    from laborious.activities.mlflow import MLFlow
    from laborious.activities.model_metrics import ModelMetrics
    from laborious.activities.opc import OPC
    from laborious.activities.storage import Storage
+    from laborious.utils.connectors_config import build_mlflow_config


 class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
--- a/laborious/activities/mlflow.py
+++ b/laborious/activities/mlflow.py
@@ -12,7 +12,6 @@ with workflow.unsafe.imports_passed_through():
    import numpy as np
    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
@@ -27,6 +26,7 @@ with workflow.unsafe.imports_passed_through():
    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 sklearn.model_selection import train_test_split

    from laborious.utils.dataframe_debug import build_dataframe_debug_message
    from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
@@ -53,6 +53,7 @@ class MLFlow(MinioManager):
    """

    _MAX_DEBUG_DATAFRAME_ROWS = 100
+    _DEFAULT_MODEL_ALIAS = 'production'

    def __init__(
        self,
@@ -191,6 +192,21 @@ class MLFlow(MinioManager):
        latest = max(versions, key=lambda v: int(v.version))
        return str(latest.version)

+    def _resolve_model_alias(self, model_config: dict[str, Any] | None = None) -> str:
+        """
+        Resolve which MLflow alias should be used for model lookup/promotion.
+
+        Args:
+         - model_config: Optional model configuration that may include ``alias``.
+
+        Return:
+            str: Alias name trimmed and normalized; defaults to ``production``.
+        """
+        if not model_config:
+            return self._DEFAULT_MODEL_ALIAS
+        alias = str(model_config.get('alias', self._DEFAULT_MODEL_ALIAS)).strip()
+        return alias or self._DEFAULT_MODEL_ALIAS
+
    @activity.defn(name='request_transform')
    async def request_transform(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
        """
@@ -217,6 +233,7 @@ class MLFlow(MinioManager):

        model_name = input_data['model_name']
        model_config = input_data.get('model_config', {})
+        model_alias = self._resolve_model_alias(model_config)

        self._debug_dataframe('Raw input data:', data, metadata)

@@ -238,7 +255,7 @@ class MLFlow(MinioManager):
        try:
            wrapper = self.mlflow_repository.get_cached_model(
                model_name=model_name,
-                alias='production',
+                alias=model_alias,
                retention_minutes=model_config.get('retention_minutes', 0),
                metadata=metadata,
            )
@@ -317,6 +334,7 @@ class MLFlow(MinioManager):

        model_name = input_data['model_name']
        model_config = input_data.get('model_config', {})
+        model_alias = self._resolve_model_alias(model_config)

        self._debug_dataframe('Input data for prediction:', data, metadata)

@@ -332,7 +350,7 @@ class MLFlow(MinioManager):
        try:
            wrapper = self.mlflow_repository.get_cached_model(
                model_name=model_name,
-                alias='production',
+                alias=model_alias,
                retention_minutes=model_config.get('retention_minutes', 0),
                metadata=metadata,
            )
@@ -490,15 +508,16 @@ class MLFlow(MinioManager):
            }

        try:
+            model_alias = self._resolve_model_alias(model_config)
            mv_src = self.mlflow_repository._client.get_model_version_by_alias(
                name=model_name,
-                alias='production',
+                alias=model_alias,
            )
            source_run_id = mv_src.run_id

            wrapper = self.mlflow_repository.get_cached_model(
                model_name=model_name,
-                alias='production',
+                alias=model_alias,
                retention_minutes=0,
                metadata=metadata,
            )
@@ -595,10 +614,11 @@ class MLFlow(MinioManager):

            version = self._resolve_model_version_for_run(run_id)

+            promote_alias = self._resolve_model_alias(input_data.get('model_config'))
            self.mlflow_repository.promote_to_alias(
                model_name=model_name,
                version=version,
-                alias='production',
+                alias=promote_alias,
                metadata=metadata,
            )

@@ -643,9 +663,10 @@ class MLFlow(MinioManager):
        model_name = input_data['model_name']

        try:
+            model_alias = self._resolve_model_alias(input_data.get('model_config'))
            mv = self.mlflow_repository._client.get_model_version_by_alias(
                name=model_name,
-                alias='production',
+                alias=model_alias,
            )
            run_id = mv.run_id

--- a/laborious/worker/worker.py
+++ b/laborious/worker/worker.py
@@ -264,10 +264,8 @@ async def main():
        logger.custom_error(f'An unhandled exception occurred: {e}', metadata)
        exit_code = 1
    finally:
-        if notification_handler:
-            notification_handler.shutdown()
-        if activities:
-            await activities.shutdown()
+        notification_handler.shutdown()
+        await activities.shutdown()
        metrics.APP_UP.labels(pod_id=POD_ID).set(0)  # Mark app as DOWN
        sys.exit(exit_code)

--- a/requirements-local.txt
+++ b/requirements-local.txt
@@ -0,0 +1,19 @@
+temporalio
+psycopg2-binary
+sqlalchemy
+asyncua
+redis
+git+ssh://git@github.com/Aignosi/sientia-dataops-library.git@1.12.0
+#git+ssh://git@github.com/Aignosi/sientia-model-library.git@0.8.2
+/home/grezewave/Documents/projects/sientia/sientia-model-library
+prometheus-client
+botocore
+boto3
+s3fs
+pyarrow
+kaleido
+hyperopt
+shap
+pycurl
+scipy<1.14.0
+scikit-learn==1.5.2
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,8 +3,8 @@ psycopg2-binary
 sqlalchemy
 asyncua
 redis
-git+ssh://git@github.com/Aignosi/sientia-dataops-library.git@1.12.0
-git+ssh://git@github.com/Aignosi/sientia-model-library.git@0.8.1
+sientia_do==1.12.0
+sientia_model==0.8.2
 prometheus-client
 botocore
 boto3
--- a/tests/laborious/activities/test_activities.py
+++ b/tests/laborious/activities/test_activities.py
@@ -229,3 +229,74 @@ async def test_shutdown(
    mock_gates_init.close.assert_called_once()
    mock_model_metrics_init.close.assert_called_once()
    mock_api_init.close.assert_called_once()
+
+
+@patch('laborious.activities.activities.SientiaMLflowRepository')
+@patch('laborious.activities.activities.build_mlflow_config')
+@patch('laborious.activities.activities.Storage.__init__')
+@patch('laborious.activities.activities.MLFlow.__init__')
+@patch('laborious.activities.activities.OPC.__init__')
+@patch('laborious.activities.activities.Gates.__init__')
+@patch('laborious.activities.activities.ModelMetrics.__init__')
+@patch('laborious.activities.activities.API.__init__')
+@patch('laborious.activities.activities.MinioRepository')
+@patch('laborious.activities.activities.MetricsController')
+def test___init___builds_mlflow_repository_when_not_provided(
+    mock_metrics_controller,
+    mock_minio_repository,
+    _mock_api_init,
+    _mock_model_metrics_init,
+    _mock_gates_init,
+    _mock_opc_init,
+    _mock_mlflow_init,
+    _mock_storage_init,
+    mock_build_mlflow_config,
+    mock_mlflow_repository_cls,
+):
+    postgres_config = {
+        'host': 'localhost',
+        'port': 5432,
+        'user': 'postgres',
+        'password': 'postgres',
+        'dbname': 'postgres',
+        'min_connections': 1,
+        'max_connections': 10,
+    }
+    minio_config = {
+        'endpoint_url': 'localhost:9000',
+        'access_key': 'minio',
+        'secret_key': 'minio123',
+        'default_bucket': 'test',
+        'retention_hours': 24,
+        'secure': False,
+    }
+    opc_config = {'bootstrap_servers': 'localhost:9092', 'polling_time': 1000, 'group_id': 'test'}
+    pi_web_api_config = {'base_url': 'https://pi', 'auth_type': 'bearer', 'auth_token': 'token'}
+    logger = MagicMock()
+    notification_handler = MagicMock()
+    plugin_store = MagicMock()
+    mock_build_mlflow_config.return_value = {
+        'url': 'http://mlflow:80',
+        'username': 'u',
+        'password': 'p',
+    }
+
+    Activities(
+        postgres_config=postgres_config,
+        plugin_store=plugin_store,
+        minio_config=minio_config,
+        opc_config=opc_config,
+        pi_web_api_config=pi_web_api_config,
+        logger=logger,
+        notification_handler=notification_handler,
+    )
+
+    mock_build_mlflow_config.assert_called_once()
+    mock_mlflow_repository_cls.assert_called_once_with(
+        host='http://mlflow:80',
+        username='u',
+        password='p',
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=mock_metrics_controller.return_value,
+    )
--- a/tests/laborious/activities/test_api.py
+++ b/tests/laborious/activities/test_api.py
@@ -328,6 +328,27 @@ async def test_write_pi_web_api_data_empty_tags(mock_dataframe, api, base_input_
    }


+@mark.asyncio
+@patch('laborious.activities.api.DataFrame')
+async def test_write_pi_web_api_data_updates_confidence_and_comments(
+    mock_dataframe, api, base_input_data
+):
+    mock_dataframe.return_value = _create_mock_dataframe()
+    api.pi_web_api_client.write_value.side_effect = [
+        [{'WebId': 'web_id_1', 'Errors': []}],
+        [{'WebId': 'web_id_2', 'Errors': []}],
+    ]
+    with patch.object(
+        api,
+        'process_pi_web_api_response',
+        new=AsyncMock(side_effect=[(0.33, 'PI warning'), (0, '')]),
+    ) as process_mock:
+        result = await api.write_pi_web_api_data(base_input_data)
+
+    assert process_mock.await_count == 2
+    assert result is not None
+
+
@mark.asyncio
 async def test_close(api):
    api.close()
--- a/tests/laborious/activities/test_mlflow.py
+++ b/tests/laborious/activities/test_mlflow.py
@@ -1,3 +1,4 @@
+from datetime import datetime
 from unittest.mock import ANY, AsyncMock, MagicMock, patch

 import numpy as np
@@ -108,6 +109,47 @@ metadata = {
 }


+def test_detect_and_parse_datetime_index_empty(mlflow):
+    df = pd.DataFrame()
+    out = mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+    assert out.empty
+
+
+def test_detect_and_parse_datetime_index_mixed_types_error(mlflow):
+    idx = pd.Index([pd.Timestamp('2020-01-01', tz='UTC'), 'x'])
+    df = pd.DataFrame({'a': [1, 2]}, index=idx)
+    with raises(ValueError):
+        mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+
+
+def test_detect_and_parse_datetime_index_invalid_string_error(mlflow):
+    idx = pd.Index(['bad-format'])
+    df = pd.DataFrame({'a': [1]}, index=idx)
+    with raises(ValueError):
+        mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+
+
+def test_detect_and_parse_datetime_index_unsupported_type_error(mlflow):
+    idx = pd.Index([pd.Period('2020-01', freq='M')])
+    df = pd.DataFrame({'a': [1]}, index=idx)
+    with raises(ValueError):
+        mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+
+
+def test_detect_and_parse_datetime_index_datetime_success(mlflow):
+    idx = pd.Index([datetime(2020, 1, 1, 0, 0, 0)])
+    df = pd.DataFrame({'a': [1]}, index=idx)
+    out = mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+    assert out.index[0].endswith('+0000')
+
+
+def test_detect_and_parse_datetime_index_timestamp_with_tz_success(mlflow):
+    idx = pd.DatetimeIndex([pd.Timestamp('2020-01-01 00:00:00', tz='UTC')])
+    df = pd.DataFrame({'a': [1]}, index=idx)
+    out = mlflow._detect_and_parse_datetime_index(df, metadata['metadata'])
+    assert out.index[0].endswith('+0000')
+
+
@mark.asyncio
@patch(
    'laborious.activities.mlflow.MinioDataFramePayload.from_dataframe',
@@ -160,6 +202,41 @@ async def test_request_transform_success(mock_from_dataframe, mlflow):
    assert response_data == mock_from_dataframe.return_value


+@mark.asyncio
+@patch(
+    'laborious.activities.mlflow.MinioDataFramePayload.from_dataframe',
+    new_callable=AsyncMock,
+)
+async def test_request_transform_success_without_transform_meta(mock_from_dataframe, mlflow):
+    ts = pd.Timestamp('2020-01-01', tz='UTC')
+    raw = pd.DataFrame(
+        {
+            'variable': ['v1'],
+            'timestamp': [ts],
+            'value': [1.0],
+            'created_at': [ts],
+        }
+    )
+    out_idx = pd.Index([ts.strftime(DATETIME_FORMAT_WITH_TZ)], name=None)
+    out_df = pd.DataFrame({'v1': [1.0]}, index=out_idx)
+    wrapper = MagicMock()
+    wrapper.transform.return_value = (out_df, {})
+    mlflow.mlflow_repository.get_cached_model.return_value = wrapper
+
+    payload = AsyncMock()
+    payload.retrieve = AsyncMock(return_value=raw)
+
+    input_data = {
+        **metadata,
+        'data': payload,
+        'model_name': 'test_model',
+        'model_config': {},
+    }
+
+    await mlflow.request_transform(input_data)
+    mock_from_dataframe.assert_called_once()
+
+
@mark.asyncio
@patch(
    'laborious.activities.mlflow.MinioDataFramePayload.from_dataframe',
@@ -233,6 +310,39 @@ async def test_request_predict(mock_to_datetime, mock_from_dataframe, mlflow):
    assert response_data == mock_from_dataframe.return_value


+@mark.asyncio
+@patch(
+    'laborious.activities.mlflow.MinioDataFramePayload.from_dataframe',
+    new_callable=AsyncMock,
+)
+@patch('laborious.activities.mlflow.to_datetime')
+async def test_request_predict_success_dataframe_and_meta(
+    mock_to_datetime, mock_from_dataframe, mlflow
+):
+    wrapper = MagicMock()
+    pred_df = pd.DataFrame({'raw': [0.3]})
+    wrapper.predict.return_value = (pred_df, {'m': 1})
+    mlflow.mlflow_repository.get_cached_model.return_value = wrapper
+
+    data_mock = MagicMock()
+    data_mock.index = pd.DatetimeIndex([pd.Timestamp('2020-01-01', tz='UTC')])
+    payload = AsyncMock()
+    payload.retrieve = AsyncMock(return_value=data_mock)
+
+    input_data = {
+        **metadata,
+        'data': payload,
+        'model_name': 'test_model',
+        'model_config': {},
+    }
+
+    await mlflow.request_predict(input_data)
+
+    assert list(pred_df.columns) == ['prediction', 'response_time']
+    mlflow.info.assert_any_call("Wrapper predict metadata: {'m': 1}", metadata['metadata'])
+    mock_from_dataframe.assert_called_once()
+
+
@mark.asyncio
@patch(
    'laborious.activities.mlflow.MinioDataFramePayload.from_dataframe',
@@ -275,7 +385,7 @@ async def test_request_predict_failure(mock_to_datetime, mock_from_dataframe, ml
 async def test_retrain_model_success_data_success_retrain(
    mock_to_datetime, mock_rmtree, mock_mkdtemp, mock_log_artifact, mlflow
 ):
-    mock_mkdtemp.return_value = '/tmp/x'
+    mock_mkdtemp.return_value = 'tmp'

    mv_alias = MagicMock()
    mv_alias.run_id = 'source-run'
@@ -326,7 +436,7 @@ async def test_retrain_model_success_data_success_retrain(
 async def test_retrain_model_success_with_payload_data(
    mock_to_datetime, mock_rmtree, mock_mkdtemp, mock_log_artifact, mlflow
 ):
-    mock_mkdtemp.return_value = '/tmp/x'
+    mock_mkdtemp.return_value = 'tmp'
    mv_alias = MagicMock(run_id='src')
    mlflow.mlflow_repository._client.get_model_version_by_alias.return_value = mv_alias
    wrapper = MagicMock()
@@ -362,6 +472,48 @@ async def test_retrain_model_success_with_payload_data(
    assert response['success'] is True


+@mark.asyncio
+@patch('laborious.activities.mlflow.mlflow.log_artifact')
+@patch('laborious.activities.mlflow.tempfile.mkdtemp')
+@patch('laborious.activities.mlflow.rmtree')
+@patch('laborious.activities.mlflow.to_datetime')
+async def test_retrain_model_success_full_retrain_branch(
+    mock_to_datetime, mock_rmtree, mock_mkdtemp, mock_log_artifact, mlflow
+):
+    mock_mkdtemp.return_value = 'tmp'
+    mv_alias = MagicMock(run_id='src')
+    mlflow.mlflow_repository._client.get_model_version_by_alias.return_value = mv_alias
+    wrapper = MagicMock()
+    mlflow.mlflow_repository.get_cached_model.return_value = wrapper
+    mock_cm = MagicMock()
+    mock_cm.__enter__.return_value = MagicMock(run_id='r', experiment_id='e')
+    mock_cm.__exit__.return_value = False
+    mlflow.mlflow_repository.start_run.return_value = mock_cm
+
+    ts = pd.Timestamp('2020-01-01', tz='UTC')
+    raw_data = pd.DataFrame(
+        {
+            'variable': ['target', 'f1', 'target', 'f1'],
+            'timestamp': [ts, ts, ts + pd.Timedelta(days=1), ts + pd.Timedelta(days=1)],
+            'value': [1.0, 2.0, 3.0, 4.0],
+        }
+    )
+    payload = AsyncMock()
+    payload.retrieve = AsyncMock(return_value=raw_data)
+
+    response = await mlflow.retrain_model(
+        {
+            **metadata,
+            'data': payload,
+            'model_name': 'test_model',
+            'model_config': {'target': 'target', 'full_retrain': True, 'validation_fraction': 0.5},
+        }
+    )
+
+    wrapper.train.assert_called_once()
+    assert response['success'] is True
+
+
@mark.asyncio
@patch('laborious.activities.mlflow.to_datetime')
 async def test_retrain_model_success_data_fail_retrain(mock_to_datetime, mlflow):
@@ -393,7 +545,6 @@ async def test_retrain_model_success_data_fail_retrain(mock_to_datetime, mlflow)
    )

    assert response['success'] is False
-    mlflow.send_notification_async.assert_called_once()
    assert 'retrain failed' in response['message']


@@ -415,19 +566,17 @@ async def test_retrain_model_data_error(mlflow):

@mark.asyncio
 async def test_retrain_model_missing_target(mlflow):
-    raw_data = MagicMock(columns=['variable', 'timestamp', 'value'])
-    raw_data.__getitem__.return_value.max.return_value = 'ts'
+    ts = pd.Timestamp('2020-01-01', tz='UTC')
+    raw_data = pd.DataFrame(
+        {
+            'variable': ['f1', 'f2'],
+            'timestamp': [ts, ts],
+            'value': [1.0, 2.0],
+        }
+    )
    payload = AsyncMock()
    payload.retrieve = AsyncMock(return_value=raw_data)

-    pivoted = MagicMock()
-    raw_data.drop_duplicates.return_value = raw_data
-    raw_data.pivot.return_value = pivoted
-    pivoted.fillna = MagicMock()
-    pivoted.columns.name = None
-    pivoted.index = MagicMock()
-    pivoted.__setitem__ = MagicMock()
-
    response = await mlflow.retrain_model(
        {
            **metadata,
@@ -561,6 +710,24 @@ async def test_get_reference_data_not_found(mlflow):
    assert result is None


+@mark.asyncio
+async def test_get_reference_data_missing_csv_file_returns_none(mlflow):
+    input_data = {
+        **metadata,
+        'model_name': 'test_model',
+    }
+    mv = MagicMock(run_id='run1')
+    mlflow.mlflow_repository._client.get_model_version_by_alias.return_value = mv
+
+    with patch('laborious.activities.mlflow.tempfile.mkdtemp', return_value='tmp'):
+        with patch('laborious.activities.mlflow.rmtree'):
+            with patch('laborious.activities.mlflow.Path') as mp:
+                mp.return_value.rglob.return_value = []
+                result = await mlflow.get_reference_data(input_data)
+
+    assert result is None
+
+
@mark.asyncio
 async def test_get_reference_data_exception(mlflow):
    input_data = {
--- a/tests/laborious/utils/models/test_minio_dataframe_payload.py
+++ b/tests/laborious/utils/models/test_minio_dataframe_payload.py
@@ -190,6 +190,21 @@ async def test_from_dataframe_inline():
    assert result.last_timestamp == '2024-01-01'


+@pytest.mark.asyncio
+@patch('laborious.utils.models.minio_dataframe_payload.OFFLOAD_THRESHOLD_BYTES', 10**9)
+async def test_from_dataframe_inline_uses_provided_last_timestamp():
+    minio = AsyncMock()
+    df = _mock_dataframe({'timestamp': ['2024-01-01'], 'value': [42]})
+    result = await MinioDataFramePayload.from_dataframe(
+        dataframe=df,
+        minio_repo=minio,
+        model_name='m',
+        operation='initial',
+        last_timestamp='2024-01-02',
+    )
+    assert result.last_timestamp == '2024-01-02'
+
+
@pytest.mark.asyncio
@patch('laborious.utils.models.minio_dataframe_payload.now')
@patch('laborious.utils.models.minio_dataframe_payload.OFFLOAD_THRESHOLD_BYTES', 0)
@@ -264,3 +279,9 @@ def test_from_dict_passthrough_existing_instance():
    original = MinioDataFramePayload(last_timestamp='2024-01-01', data={'a': 1}, bucket='b')
    result = MinioDataFramePayload.from_dict(original)
    assert result is original
+
+
+def test_debug_with_logger_calls_custom_debug():
+    logger = MagicMock()
+    MinioDataFramePayload._debug(logger, 'msg', {'a': 1})
+    logger.custom_debug.assert_called_once_with('msg', {'a': 1})
--- a/tests/laborious/utils/test_dataframe_debug.py
+++ b/tests/laborious/utils/test_dataframe_debug.py
@@ -0,0 +1,12 @@
+from pandas import DataFrame
+
+from laborious.utils.dataframe_debug import build_dataframe_debug_message
+
+
+def test_build_dataframe_debug_message_skips_large_dataframe():
+    df = DataFrame({'a': [1, 2, 3]})
+
+    msg = build_dataframe_debug_message('payload', df, max_rows=1)
+
+    assert 'skipped because dataframe has 3 rows' in msg
+    assert '(max: 1)' in msg
--- a/tests/laborious/worker/test_worker.py
+++ b/tests/laborious/worker/test_worker.py
@@ -0,0 +1,243 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from pytest import mark, raises
+
+from laborious.worker import worker
+
+
+def _build_fake_activities():
+    inst = MagicMock()
+    inst.init_opc = AsyncMock()
+    inst.shutdown = AsyncMock()
+    inst.load_query_with_minio_offload = MagicMock()
+    inst.retrain_model = MagicMock()
+    inst.update_production_model = MagicMock()
+    inst.format_retrain_report = MagicMock()
+    inst.export_data_to_postgres = MagicMock()
+    inst.load_custom_query = MagicMock()
+    inst.calculate_simple_metrics = MagicMock()
+    inst.get_reference_data = MagicMock()
+    inst.calculate_drift = MagicMock()
+    inst.request_predict = MagicMock()
+    inst.request_transform = MagicMock()
+    inst.input_gate = MagicMock()
+    inst.mlflow_response_gate = MagicMock()
+    inst.mlflow_content_gate = MagicMock()
+    inst.format_transformed_data = MagicMock()
+    inst.format_prediction = MagicMock()
+    inst.format_default_prediction = MagicMock()
+    inst.write_opc_data = MagicMock()
+    inst.cleanup_minio_objects_expired = MagicMock()
+    inst.repeat_last_prediction = MagicMock()
+    inst.write_metrics = MagicMock()
+    inst.write_pi_web_api_data = MagicMock()
+    return inst
+
+
+def _build_fake_worker(async_result=None, async_error: Exception | None = None):
+    w = MagicMock()
+
+    async def _run():
+        if async_error is not None:
+            raise async_error
+        return async_result
+
+    w.run = MagicMock(side_effect=_run)
+    return w
+
+
+@patch('laborious.worker.worker.start_http_server')
+def test_start_prometheus_server_success(mock_start_http):
+    with patch.object(worker.metrics.APP_UP, 'labels') as labels:
+        gauge = MagicMock()
+        labels.return_value = gauge
+        with patch('laborious.worker.worker.os.getenv', return_value='9090'):
+            worker.start_prometheus_server()
+    mock_start_http.assert_called_once_with(9090)
+    gauge.set.assert_called_once_with(1)
+
+
+@patch('laborious.worker.worker.start_http_server', side_effect=RuntimeError('nope'))
+def test_start_prometheus_server_error_exits(_mock_start_http):
+    with patch('laborious.worker.worker.os._exit', side_effect=SystemExit(1)) as m_exit:
+        with raises(SystemExit):
+            worker.start_prometheus_server()
+    m_exit.assert_called_once_with(1)
+
+
+@mark.asyncio
+async def test_main_missing_runtime_exits_fast(monkeypatch):
+    monkeypatch.setenv('RUNTIME', '')
+    with (
+        patch('laborious.worker.worker.start_prometheus_server'),
+        patch('laborious.worker.worker.get_logger') as m_logger,
+        patch('laborious.worker.worker.NotificationHandler'),
+        patch('laborious.worker.worker.MetricsController'),
+        patch.object(worker.metrics.APP_UP, 'labels') as labels,
+        patch('laborious.worker.worker.sys.exit', side_effect=SystemExit(1)),
+    ):
+        labels.return_value = MagicMock()
+        with raises(SystemExit):
+            await worker.main()
+    assert m_logger.return_value.custom_critical.called
+
+
+@mark.asyncio
+async def test_main_plugin_install_failure(monkeypatch):
+    monkeypatch.setenv('RUNTIME', 'single')
+    fake_activities = _build_fake_activities()
+    fake_plugin = MagicMock()
+    fake_plugin.install_runtime = AsyncMock(side_effect=RuntimeError('install failed'))
+    with (
+        patch('laborious.worker.worker.start_prometheus_server'),
+        patch('laborious.worker.worker.get_logger'),
+        patch(
+            'laborious.worker.worker.build_mongodb_config',
+            return_value={'connection_string': 'cs', 'database_name': 'db'},
+        ),
+        patch('laborious.worker.worker.NotificationHandler'),
+        patch('laborious.worker.worker.MetricsController'),
+        patch(
+            'laborious.worker.worker.build_plugin_store_config',
+            return_value={
+                'base_url': '',
+                'owner': '',
+                'repo': '',
+                'username': None,
+                'password': None,
+                'branch': None,
+                'cache_ttl_seconds': None,
+                'pypi_index_url': '',
+                'pypi_username': None,
+                'pypi_password': None,
+            },
+        ),
+        patch('laborious.worker.worker.PluginStore', return_value=fake_plugin),
+        patch('laborious.worker.worker.Activities', return_value=fake_activities),
+        patch.object(worker.metrics.APP_UP, 'labels') as labels,
+        patch('laborious.worker.worker.sys.exit', side_effect=SystemExit(1)),
+    ):
+        labels.return_value = MagicMock()
+        with raises(SystemExit):
+            await worker.main()
+
+
+@mark.asyncio
+async def test_main_success_exit_zero(monkeypatch):
+    monkeypatch.setenv('RUNTIME', 'single')
+    fake_activities = _build_fake_activities()
+    fake_plugin = MagicMock()
+    fake_plugin.install_runtime = AsyncMock(return_value=None)
+    fake_workers = [_build_fake_worker() for _ in range(4)]
+
+    with (
+        patch('laborious.worker.worker.start_prometheus_server'),
+        patch('laborious.worker.worker.get_logger'),
+        patch(
+            'laborious.worker.worker.build_mongodb_config',
+            return_value={'connection_string': 'cs', 'database_name': 'db'},
+        ),
+        patch('laborious.worker.worker.NotificationHandler') as m_notif_cls,
+        patch('laborious.worker.worker.MetricsController'),
+        patch(
+            'laborious.worker.worker.build_plugin_store_config',
+            return_value={
+                'base_url': '',
+                'owner': '',
+                'repo': '',
+                'username': None,
+                'password': None,
+                'branch': None,
+                'cache_ttl_seconds': None,
+                'pypi_index_url': '',
+                'pypi_username': None,
+                'pypi_password': None,
+            },
+        ),
+        patch('laborious.worker.worker.PluginStore', return_value=fake_plugin),
+        patch('laborious.worker.worker.Activities', return_value=fake_activities),
+        patch('laborious.worker.worker.build_postgres_config', return_value={}),
+        patch('laborious.worker.worker.build_minio_config', return_value={}),
+        patch('laborious.worker.worker.build_opc_config', return_value={}),
+        patch('laborious.worker.worker.build_api_config', return_value={}),
+        patch('laborious.worker.worker.PrometheusConfig', return_value=MagicMock()),
+        patch('laborious.worker.worker.TelemetryConfig', return_value=MagicMock()),
+        patch('laborious.worker.worker.Runtime', return_value=MagicMock()),
+        patch(
+            'laborious.worker.worker.client.Client.connect', new=AsyncMock(return_value=MagicMock())
+        ),
+        patch('laborious.worker.worker.prepare_worker', side_effect=fake_workers) as m_prepare,
+        patch.object(worker.metrics.APP_UP, 'labels') as labels,
+        patch('laborious.worker.worker.sys.exit', side_effect=SystemExit(0)),
+    ):
+        labels.return_value = MagicMock()
+        with raises(SystemExit):
+            await worker.main()
+
+    notif = m_notif_cls.return_value
+    notif.shutdown.assert_called_once()
+    fake_activities.shutdown.assert_awaited_once()
+    assert m_prepare.call_count == 4
+    prepare_calls = m_prepare.call_args_list
+    assert prepare_calls[0].kwargs['runtime'] == 'single'
+    assert prepare_calls[3].kwargs['runtime'] == 'single'
+
+
+@mark.asyncio
+async def test_main_worker_gather_error_exits_one(monkeypatch):
+    monkeypatch.setenv('RUNTIME', 'single')
+    fake_activities = _build_fake_activities()
+    fake_plugin = MagicMock()
+    fake_plugin.install_runtime = AsyncMock(return_value=None)
+    fake_workers = [
+        _build_fake_worker(async_error=RuntimeError('boom')),
+        _build_fake_worker(),
+        _build_fake_worker(),
+        _build_fake_worker(),
+    ]
+
+    with (
+        patch('laborious.worker.worker.start_prometheus_server'),
+        patch('laborious.worker.worker.get_logger') as m_logger,
+        patch(
+            'laborious.worker.worker.build_mongodb_config',
+            return_value={'connection_string': 'cs', 'database_name': 'db'},
+        ),
+        patch('laborious.worker.worker.NotificationHandler'),
+        patch('laborious.worker.worker.MetricsController'),
+        patch(
+            'laborious.worker.worker.build_plugin_store_config',
+            return_value={
+                'base_url': '',
+                'owner': '',
+                'repo': '',
+                'username': None,
+                'password': None,
+                'branch': None,
+                'cache_ttl_seconds': None,
+                'pypi_index_url': '',
+                'pypi_username': None,
+                'pypi_password': None,
+            },
+        ),
+        patch('laborious.worker.worker.PluginStore', return_value=fake_plugin),
+        patch('laborious.worker.worker.Activities', return_value=fake_activities),
+        patch('laborious.worker.worker.build_postgres_config', return_value={}),
+        patch('laborious.worker.worker.build_minio_config', return_value={}),
+        patch('laborious.worker.worker.build_opc_config', return_value={}),
+        patch('laborious.worker.worker.build_api_config', return_value={}),
+        patch('laborious.worker.worker.PrometheusConfig', return_value=MagicMock()),
+        patch('laborious.worker.worker.TelemetryConfig', return_value=MagicMock()),
+        patch('laborious.worker.worker.Runtime', return_value=MagicMock()),
+        patch(
+            'laborious.worker.worker.client.Client.connect', new=AsyncMock(return_value=MagicMock())
+        ),
+        patch('laborious.worker.worker.prepare_worker', side_effect=fake_workers),
+        patch.object(worker.metrics.APP_UP, 'labels') as labels,
+        patch('laborious.worker.worker.sys.exit', side_effect=SystemExit(1)),
+    ):
+        labels.return_value = MagicMock()
+        with raises(SystemExit):
+            await worker.main()
+
+    assert m_logger.return_value.custom_error.called
--- a/tests/laborious/workflows/subworkflows/test_prediction_process.py
+++ b/tests/laborious/workflows/subworkflows/test_prediction_process.py
@@ -1,6 +1,6 @@
 from unittest.mock import ANY, AsyncMock, MagicMock, call, patch

-from pytest import fixture, mark
+from pytest import fixture, mark, raises

 from laborious.activities.activities import Activities
 from laborious.workflows.sub_workflows.prediction_process import PredictionProcess
@@ -840,3 +840,27 @@ async def test_run_with_cleanup_prefixes(workflow_mock, prediction_process):
        retry_policy=ANY,
        start_to_close_timeout=ANY,
    )
+
+
+@mark.asyncio
+@patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
+async def test_run_always_cleans_up_on_pipeline_exception(workflow_mock, prediction_process):
+    input_data = {
+        'metadata': metadata,
+        'data': {'last_timestamp': '2024-01-01'},
+        'model_id': 1,
+        'model_name': 'm',
+        'model_config': {},
+        'save_transform': False,
+    }
+    prediction_process._run_prediction_pipeline = AsyncMock(side_effect=RuntimeError('boom'))
+
+    with raises(RuntimeError):
+        await prediction_process.run(input_data)
+
+    workflow_mock.execute_activity_method.assert_called_once_with(
+        Activities.cleanup_minio_objects_expired,
+        {**metadata, 'data': input_data['data']},
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )