diff --git a/README.md b/README.md index ede41bf..6f3172c 100644 --- a/README.md +++ b/README.md @@ -487,16 +487,18 @@ The **SimpleMetrics** workflow calculates regression metrics (RMSE, MSE, MAE, R #### Execution Flow 1. **Data Loading**: Loads prediction vs target data via a JOIN query -2. **Metrics Calculation**: Calculates configured regression metrics +2. **Metrics Calculation**: Calculates configured regression metrics; alerts on `simple_metrics_thresholds` breaches (one alert per model, the message names the window) unless the window has fewer rows than `simple_metrics_min_samples` 3. **Data Export**: Exports metrics to PostgreSQL +4. **Adaptive State**: when the schedule's `pipelines` doc has `frequency_alert`, escalates `adaptive_state` on a breach and recovers after the window is renewed without one (`min(12, ceil(interval_minutes / frequency_alert))` runs, at least 1; empty, under-sampled or threshold-less windows count as healthy). The orchestrator runs the schedule at `frequency_alert` while escalated. Runs on every schedule (it reads the `pipelines` doc to check for `frequency_alert`), so a MongoDB failure here fails the run; the metrics are already exported by then #### Architecture Diagram ```mermaid flowchart LR - A[1. load_custom_query] --> B[2. calculate_simple_metrics] --> C[3. export_data_to_postgres] + A[1. load_custom_query] --> B[2. calculate_simple_metrics] --> C[3. export_data_to_postgres] --> D[4. update_simple_metrics_adaptive_state] A -.-> Database[(Database)] C -.-> PostgreSQL[(PostgreSQL)] + D -.-> MongoDB[(MongoDB pipelines)] ``` #### Input Parameters @@ -510,7 +512,12 @@ flowchart LR "data_table_name": "laborious_data", "target_table_name": "simple_metrics", "interval_minutes": 60, - "model_config": { "target": "temperature", "alias": "production" }, + "model_config": { + "target": "temperature", + "alias": "production", + "simple_metrics_thresholds": { "rmse_max": 1.5 }, + "simple_metrics_min_samples": 10 + }, "metrics": ["rmse", "mse", "mae", "r2"] } ``` @@ -522,7 +529,7 @@ flowchart LR - PostgreSQL database - MLFlow server - MinIO object storage (for MLFlow artifacts) -- MongoDB server (for notifications) +- MongoDB server (notifications, prediction metadata, and read/write on the orchestrator's `pipelines` collection for the simple-metrics adaptive state) - OPC server(s) if using OPC export - PI Web API server if using PI Web API export @@ -898,7 +905,7 @@ See [OPC UA Communication](#opc-ua-communication) for semantics, concurrency, an | `MONGODB_URL` | MongoDB connection URI | `localhost:27018` | Yes | | `MONGODB_USERNAME` | MongoDB username | `root` | Yes | | `MONGODB_PASSWORD` | MongoDB password | `wKZDbMNU1c` | Yes | -| `MONGODB_DATABASE_NAME` | MongoDB database name | `sientia` | Yes | +| `MONGODB_DATABASE_NAME` | MongoDB database name; must be the same database as the orchestrator's `pipelines` | `sientia` | Yes | | `MONGODB_TTL_INDEX_HOURS` | MongoDB TTL index hours | `1` | No | | `LOG_LEVEL` | Application log level | `INFO` | No | | `PROJECT_NAME` | Project name for metrics | `laborious` | No | diff --git a/e2e/conftest.py b/e2e/conftest.py index b9c9591..ef2fd1f 100644 --- a/e2e/conftest.py +++ b/e2e/conftest.py @@ -460,6 +460,7 @@ def _simple_metrics_worker_activity_list(test_activities: Activities): test_activities.load_custom_query, test_activities.calculate_simple_metrics, test_activities.export_data_to_postgres, + test_activities.update_simple_metrics_adaptive_state, ] diff --git a/e2e/scenarios.md b/e2e/scenarios.md index 78fc0ae..032917b 100644 --- a/e2e/scenarios.md +++ b/e2e/scenarios.md @@ -413,6 +413,15 @@ four metrics with closed-form expected values. #### S.1.2 Subset metrics **Summary**: Requesting `metrics=['rmse']` writes only the rmse row. +#### S.1.3 Threshold breach escalates the adaptive state +**Summary**: With `simple_metrics_thresholds` breached and a `pipelines` doc +(same `schedule_name`) carrying `frequency_alert`, the workflow sets +`adaptive_state={escalated: true, healthy_streak: 0}` and stamps `updated_at`. + +**Expected Outcome**: +- `adaptive_state` written on the `pipelines` doc. +- `updated_at` is a BSON datetime (the orchestrator calls `.replace(tzinfo=UTC)` on it). + ### 6.2 Edge cases #### S.2.1 Zero-variance target returns r2=0 diff --git a/e2e/test_simple_metrics.py b/e2e/test_simple_metrics.py index c232e65..3edcfce 100644 --- a/e2e/test_simple_metrics.py +++ b/e2e/test_simple_metrics.py @@ -15,6 +15,7 @@ from decimal import Decimal import math import pytest +from pymongo import MongoClient from sqlalchemy import text from temporalio.testing import WorkflowEnvironment from temporalio.worker import Worker @@ -140,7 +141,8 @@ async def test_simple_metrics_happy_path_persists_all_metrics_and_columns( prediction/target pair set and written one row per metric. Every column expected by ``sientia_data.simple_metrics`` must be populated (except the nullable ``timestamp`` column) and the numerical values must match - closed-form expectations. + closed-form expectations, rounded to 2 decimals as ``RegressionMetrics`` + persists them (see ``sientia_model/metrics/regression.py``). """ client = temporal_test_env.client model_id = 511 @@ -154,13 +156,13 @@ async def test_simple_metrics_happy_path_persists_all_metrics_and_columns( ] diffs = [target - prediction for prediction, target in pairs] n = len(diffs) - expected_rmse = math.sqrt(sum(d * d for d in diffs) / n) - expected_mse = sum(d * d for d in diffs) / n - expected_mae = sum(abs(d) for d in diffs) / n + expected_rmse = round(math.sqrt(sum(d * d for d in diffs) / n), 2) + expected_mse = round(sum(d * d for d in diffs) / n, 2) + expected_mae = round(sum(abs(d) for d in diffs) / n, 2) target_mean = sum(t for _, t in pairs) / n ss_res = sum((target - prediction) ** 2 for prediction, target in pairs) ss_tot = sum((t - target_mean) ** 2 for _, t in pairs) - expected_r2 = 1.0 - (ss_res / ss_tot) + expected_r2 = round(1.0 - (ss_res / ss_tot), 2) _seed_predictions_and_targets(postgres_engine, model_id=model_id, pairs=pairs) @@ -331,3 +333,54 @@ async def test_simple_metrics_no_overlapping_data_short_circuits( {'m': str(model_id)}, ).scalar() assert count == 0, 'Empty target data must short-circuit and skip persistence' + + +@pytest.mark.asyncio +@pytest.mark.integration +async def test_simple_metrics_breach_escalates_adaptive_state( + temporal_test_env: WorkflowEnvironment, + temporal_worker_simple_metrics: Worker, + test_activities: Activities, + postgres_engine, + mongo_container, +): + """ + Scenario S.1.3: a threshold breach on a schedule whose pipelines doc has + ``frequency_alert`` escalates ``adaptive_state`` and stamps ``updated_at`` + as a BSON datetime (the type the orchestrator reads). + """ + model_id = 541 + schedule_name = 'test-simple-metrics-adaptive' + _seed_predictions_and_targets( + postgres_engine, model_id=model_id, pairs=[(1.0, 2.0), (2.0, 4.0), (3.0, 5.0)] + ) + + mongo = MongoClient(f'mongodb://localhost:{mongo_container.get_exposed_port(27017)}') + pipelines = mongo['test_db']['pipelines'] + try: + pipelines.delete_many({'schedule_name': schedule_name}) + pipelines.insert_one( + {'schedule_name': schedule_name, 'frequency': '1h', 'frequency_alert': '5m'} + ) + + input_data = _simple_metrics_input( + model_id, + schedule_name=schedule_name, + model_config={ + 'target': 'sensor_target', + 'simple_metrics_thresholds': {'rmse_max': 0.01}, + }, + ) + await start_and_await_workflow( + temporal_test_env.client, + SimpleMetrics.run, + input_data, + make_workflow_id('test-simple-metrics-adaptive'), + ) + + doc = pipelines.find_one({'schedule_name': schedule_name}) + finally: + mongo.close() + + assert doc['adaptive_state'] == {'escalated': True, 'healthy_streak': 0} + assert isinstance(doc['updated_at'], datetime) diff --git a/laborious/activities/activities.py b/laborious/activities/activities.py index 0d8ef09..cd75673 100644 --- a/laborious/activities/activities.py +++ b/laborious/activities/activities.py @@ -16,22 +16,22 @@ with workflow.unsafe.imports_passed_through(): from laborious.activities.gates import Gates from laborious.activities.mlflow import MLFlow from laborious.activities.model_metrics import ModelMetrics + from laborious.activities.mongo import MongoActivities from laborious.activities.opc import OPC - from laborious.activities.prediction_metadata import PredictionMetadata from laborious.activities.storage import Storage from laborious.utils.connectors_config import build_mlflow_config -class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, PredictionMetadata): +class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, MongoActivities): """ Central orchestrator for all Temporal activities used by Laborious workflows. 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, PI Web API writes, and PredictionMetadata (the single write point for - gate metadata — see `laborious.activities.prediction_metadata`). The worker constructs one - ``Activities`` instance per process and registers its callables on multiple workers bound to - different task queues. + drift/simple metrics, PI Web API writes, and MongoActivities (the single write point for + gate metadata and the simple metrics adaptive state — see `laborious.activities.mongo`). + The worker constructs one ``Activities`` instance per process and registers its callables on + multiple workers bound to different task queues. MLflow connectivity: unless ``mlflow_repository`` is injected (tests only), this class builds ``SientiaMLflowRepository`` from ``build_mlflow_config()`` so tracking credentials and URL @@ -73,7 +73,7 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, PredictionMetad - 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()``. - - mongo_repository: Optional ``MongoDBRepository`` (prediction metadata, best-effort) for + - mongo_repository: Optional ``MongoDBRepository`` (prediction metadata, adaptive state) for unit/e2e tests; in production leave unset so the repository is built from environment via ``build_mongodb_config()``, same pattern as ``mlflow_repository`` above. @@ -153,7 +153,7 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, PredictionMetad metrics_controller=mc, ) - PredictionMetadata.__init__( + MongoActivities.__init__( self, mongo_repository=mongo_repository, logger=logger, @@ -202,4 +202,4 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, PredictionMetad OPC.close(self) ModelMetrics.close(self) API.close(self) - PredictionMetadata.close(self) + MongoActivities.close(self) diff --git a/laborious/activities/gates.py b/laborious/activities/gates.py index c1f91f3..4fb69c7 100644 --- a/laborious/activities/gates.py +++ b/laborious/activities/gates.py @@ -242,7 +242,7 @@ class Gates(SientiaMonitoring): 5. Handles errors gracefully with notification integration 6. Builds a metadata document describing which filter(s) triggered, with their policy and a counter, whenever one does — this activity only computes it, it never writes it - (see `PredictionMetadata.write_prediction_metadata` for the actual write) + (see `MongoActivities.write_prediction_metadata` for the actual write) Args: input_data: Configuration and data for input validation diff --git a/laborious/activities/model_metrics.py b/laborious/activities/model_metrics.py index cb44805..650ed94 100644 --- a/laborious/activities/model_metrics.py +++ b/laborious/activities/model_metrics.py @@ -4,9 +4,9 @@ with workflow.unsafe.imports_passed_through(): import time import traceback import warnings + from math import ceil from typing import Any - import numpy as np import pandas as pd from pandas import DataFrame, Index, Series, to_datetime from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler @@ -16,6 +16,7 @@ with workflow.unsafe.imports_passed_through(): from sientia_do.observability.sientia_monitoring import SientiaMonitoring from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ from sientia_model.analytics.drift_analysis import DriftAnalysis, DriftInsufficientDataError + from sientia_model.metrics.regression import RegressionMetrics from laborious import metrics from laborious.utils.dataframe_debug import build_dataframe_debug_message @@ -26,6 +27,73 @@ warnings.filterwarnings( ) +def threshold_breaches( + rows: list[dict[str, Any]], thresholds: dict[str, float] | None +) -> list[str]: + """ + Breach messages for metric rows outside their configured thresholds. + + Convention: `_max` breaches when value > threshold, `_min` when value < + threshold. Shared by the breach alert and the adaptive-state rule so both agree on what a + breach is. + """ + thresholds = thresholds or {} + breaches = [] + for row in rows: + metric_name, value = row['metric'], row['value'] + max_key, min_key = f'{metric_name}_max', f'{metric_name}_min' + if max_key in thresholds and value > thresholds[max_key]: + breaches.append(f'{metric_name}={value} exceeds {max_key}={thresholds[max_key]}') + elif min_key in thresholds and value < thresholds[min_key]: + breaches.append(f'{metric_name}={value} below {min_key}={thresholds[min_key]}') + return breaches + + +_MAX_RECOVERY_RUNS = 12 +_FREQUENCY_UNIT_SECONDS = {'s': 1, 'm': 60, 'h': 3600, 'd': 86400} + + +def frequency_to_seconds(frequency: str) -> int: + """`'5m'` -> 300. Same `''` format the orchestrator schedules with.""" + # ponytail: copy of orchestrator/utils/converters.py:parse_frequency (no shared helper in + # sientia_do), same int() parsing so both repos accept the same strings, plus a > 0 guard; + # move it to sientia_do if a third repo needs it. + unit = _FREQUENCY_UNIT_SECONDS.get(frequency[-1:]) + if unit is None: + raise ValueError(f'Invalid frequency: {frequency!r}') + value = int(frequency[:-1]) + if value <= 0: + raise ValueError(f'Invalid frequency: {frequency!r}') + return value * unit + + +def recovery_runs(interval_minutes: int, frequency_alert: str) -> int: + """ + N of the hysteresis rule: healthy runs needed to renew the whole window without a breach, + capped at 12 (SIENTIAPDE-1995 decision 3). + """ + runs = ceil(interval_minutes * 60 / frequency_to_seconds(frequency_alert)) + return min(_MAX_RECOVERY_RUNS, max(1, runs)) + + +def next_adaptive_state(state: dict[str, Any], breached: bool, runs: int) -> dict[str, Any]: + """ + Hysteresis rule of the adaptive frequency (SIENTIAPDE-1995): escalate on the first breach, + recover after `runs` consecutive non-breach runs. + """ + if breached: + return {'escalated': True, 'healthy_streak': 0} + if not state.get('escalated'): + return {'escalated': False, 'healthy_streak': 0} + try: + streak = int(state.get('healthy_streak', 0)) + 1 + except (TypeError, ValueError, OverflowError): + streak = 1 # hand-edited streak (null, text, inf): restart the count, stay escalated + if streak >= runs: + return {'escalated': False, 'healthy_streak': 0} + return {'escalated': True, 'healthy_streak': streak} + + class ModelMetrics(SientiaMonitoring): """ Metrics and statistical analysis activities for the Laborious pipeline. @@ -365,62 +433,62 @@ class ModelMetrics(SientiaMonitoring): @activity.defn(name='calculate_simple_metrics') def calculate_simple_metrics(self, input_data: dict[str, Any]) -> list[dict]: """ - Calculate simple metrics for a model. Metrics available are: - - rmse - - mse - - mae - - r2 - - accuracy - - precision - - recall - - f1 + Calculate simple regression metrics for a model using RegressionMetrics. + Args: - input_data (dict[str, Any]): Input data containing: + input_data: Input data containing: - metadata (dict): Workflow execution metadata - model_id (str): ID of the MLFlow model - - target_data (pd.DataFrame): Target data for calculating metrics, containing target and prediction columns - - metrics (list[str]): List of metrics to calculate + - target_data (list[dict]): Target data with target, prediction, timestamp columns + - metrics (list[str]): Metric names to calculate + - interval_minutes (int): Window interval in minutes + - model_type (str | None): Model algorithm type (for r2 lock) Returns: - dict[Hashable, Any]: Dictionary containing the calculated metrics + list[dict]: Records with metric, value, model_id, timestamp, data_size, interval_minutes """ - metadata = input_data['metadata'] model_id = input_data['model_id'] target_data = DataFrame(input_data['target_data']) - metric_names = input_data['metrics'] + metric_names = list(input_data['metrics']) interval_minutes = input_data['interval_minutes'] + model_type = input_data.get('model_type') data_size = target_data.shape[0] - output_data = [] + # Filter r2 when model_type is known and unsupported + if ( + model_type + and 'r2' in metric_names + and not RegressionMetrics.is_r2_supported(model_type) + ): + metric_names = [m for m in metric_names if m != 'r2'] + self.warning( + f'r2 excluded for model {model_id}: not supported for model_type={model_type}', + metadata, + ) + self.send_notification( + metadata=metadata, + notification_id='SIMPLE_METRICS_R2_UNSUPPORTED', + message=f'r2 excluded: not a valid metric for model_type={model_type}', + block='model_metrics', + level=NotificationLevel.WARNING, + ) - diff = target_data['target'] - target_data['prediction'] - diff_squared = diff**2 + if not metric_names: + self.info(f'No metrics to calculate for model {model_id} after filtering', metadata) + return [] self.info(f'Calculating simple metrics for model {model_id}: {metric_names}', metadata) - for metric in metric_names: - if metric == 'rmse': - output_data.append({'metric': 'rmse', 'value': np.sqrt(np.mean(diff_squared))}) - elif metric == 'mse': - output_data.append({'metric': 'mse', 'value': np.mean(diff_squared)}) - elif metric == 'mae': - output_data.append({'metric': 'mae', 'value': np.mean(np.abs(diff))}) - elif metric == 'r2': - y_true = target_data['target'] - y_mean = np.mean(y_true) + # Build Series with DatetimeIndex for RegressionMetrics + timestamps = pd.to_datetime(target_data['timestamp']) + real_data = Series(target_data['target'].values, index=timestamps, dtype=float) + predictions = Series(target_data['prediction'].values, index=timestamps, dtype=float) - ss_res = np.sum(diff_squared) - ss_tot = np.sum((y_true - y_mean) ** 2) - - # Evita divisão por zero - if ss_tot == 0: - r2_score = 0.0 - else: - r2_score = 1 - (ss_res / ss_tot) - - output_data.append({'metric': 'r2', 'value': r2_score}) + regression = RegressionMetrics(real_data, predictions) + output_data = regression.calculate(metric_names) + # Wrap with metadata columns matching the existing output schema data = DataFrame(output_data) data['model_id'] = model_id data['timestamp'] = target_data['timestamp'].max() @@ -429,4 +497,22 @@ class ModelMetrics(SientiaMonitoring): self._debug_dataframe(f'Simple metrics dataframe: Size {data.shape}', data, metadata) + # Threshold alerting (optional — no crash when absent) + for breach_msg in threshold_breaches(output_data, input_data.get('thresholds')): + message = ( + f'Threshold breach for model {model_id} ' + f'(window {interval_minutes} min): {breach_msg}' + ) + self.warning(message, metadata) + self.send_notification( + # Model-level trigger: the orchestrator dedups alerts on + # `:`, so one breach seen by several windows + # (one schedule each) becomes one alert (SIENTIAPDE-1995). + metadata={**metadata, 'schedule_name': f'{model_id}-simple_metrics'}, + notification_id='SIMPLE_METRICS_THRESHOLD_BREACH', + message=message, + block='model_metrics', + level=NotificationLevel.WARNING, + ) + return data.to_dict(orient='records') diff --git a/laborious/activities/mongo.py b/laborious/activities/mongo.py new file mode 100644 index 0000000..fc4d92f --- /dev/null +++ b/laborious/activities/mongo.py @@ -0,0 +1,227 @@ +from temporalio import activity, workflow + +with workflow.unsafe.imports_passed_through(): + from typing import Any + + from bson import ObjectId + from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler + from sientia_do.observability.logger import Logger + from sientia_do.observability.metrics_controller import MetricsController + from sientia_do.observability.sientia_monitoring import SientiaMonitoring + from sientia_do.repository.mongodb_repository_sync import MongoDBRepository + from sientia_do.temporal.constants import now + + from laborious.activities.model_metrics import ( + next_adaptive_state, + recovery_runs, + threshold_breaches, + ) + + +class MongoActivities(SientiaMonitoring): + """ + Activities that read from or write to MongoDB, sharing one `MongoDBRepository`. + + - `write_prediction_metadata`: centralized, best-effort writer for prediction gate metadata + (triggered filter, policy, counter). Gates only compute and return a metadata document; + they never write it themselves. Keeping the write here, instead of in each gate, puts the + one real write point in the subworkflow that's about to persist a `predictions` row — the + only place that reliably knows whether one is being written at all — and lets other + metadata sources reuse the same activity instead of duplicating write logic wherever they + originate. + - `update_simple_metrics_adaptive_state`: adaptive frequency state on the `pipelines` doc of + a simple metrics schedule (SIENTIAPDE-1995). + """ + + mongo_repository: MongoDBRepository | None = None + _PREDICTION_METADATA_COLLECTION = 'prediction_metadata' + _PIPELINES_COLLECTION = 'pipelines' + + def __init__( + self, + mongo_repository: MongoDBRepository | None = None, + logger: Logger | None = None, + notification_handler: NotificationHandler | None = None, + metrics_controller: MetricsController | None = None, + ): + """ + Initialize the Mongo activities. + + Args: + mongo_repository: MongoDB, where gate metadata (triggered filter, policy, counter) is + written best-effort and the simple metrics adaptive state is kept. `None` is a + valid state: both activities become no-ops. + + Raises: + Exception: If BaseActivity initialization fails + """ + self.mongo_repository = mongo_repository + SientiaMonitoring.__init__( + self, + logger=logger, + notification_handler=notification_handler, + metrics_controller=metrics_controller, + ) + + def close(self) -> None: + """ + Close the Mongo repository and clean up resources. + """ + if self.mongo_repository is not None: + try: + self.mongo_repository.close() + finally: + self.mongo_repository = None + SientiaMonitoring.shutdown(self) + + def __del__(self): + self.close() + + @activity.defn(name='write_prediction_metadata') + def write_prediction_metadata(self, input_data: dict[str, Any]) -> str | None: + """ + Best-effort write of one or more gate metadata documents to Mongo, combined into a single + `prediction_metadata` document. + + Implements both business rules for this feature in one place: + - The prediction write must never be blocked or failed by a metadata write. A failed + `insert(...)` is swallowed, never re-raised, logged here with call-site context (which + `mongo_repository.insert(...)` itself can't know) — matching how `storage.py`'s + `cleanup_minio_objects_expired` also logs again at the call site even though the + repository already logs internally and emits its own Prometheus metric + (`MONGODB_WRITE_ERROR_COUNT`); the generic library log is for the metric/trace, this + one is for whoever is reading logs for this specific call. + - Nothing is written when there's nothing relevant to report. An empty/missing + `documents` list (or one full of falsy/empty entries) skips Mongo entirely, so callers + can call this unconditionally and get `None` back for "no metadata this time" instead + of deciding upfront whether there's something worth saving. + + Takes a *list* of documents, not a single one: each caller contributes its own document + (shaped like `{'filters': [...]}`), and this flattens every `filters` entry across all of + them into one `prediction_metadata` document before the single insert. Today only one + gate ever contributes a document per prediction cycle (the gate chain short-circuits on + the first non-`None` `path_flag`), but the list shape means this activity won't need to + change again once more than one source contributes metadata for the same cycle. + + `insert(...)` doesn't hand back the inserted document's `_id`, so it is generated here, + client-side, before the insert, as a plain string (not a BSON ObjectId) — matching this + codebase's convention of string `_id`s (e.g. the `models` collection) and the 24-hex-char + size of the `predictions.metadata_id` Postgres column that will reference it. + + Args: + input_data: Dict with: + - documents (list[dict[str, Any]]): Metadata payloads to combine, each shaped + like `{'filters': [...]}`. Missing/empty, or entries with no `filters`, skip + the write entirely. + - metadata (dict): Workflow metadata for contextual logging. + + Return: + The generated id, or `None` when there's nothing to write, Mongo isn't configured, or + the insert failed. + """ + metadata = input_data['metadata'] + documents = input_data.get('documents') or [] + + filters = [ + entry for document in documents if document for entry in document.get('filters', []) + ] + + if not filters or self.mongo_repository is None: + return None + + generated_id = str(ObjectId()) + + try: + self.mongo_repository.insert( + self._PREDICTION_METADATA_COLLECTION, + {'filters': filters, '_id': generated_id}, + metadata, + ) + except Exception as e: + self.error(f'Failed to write prediction metadata: {e}', metadata) + return None + + return generated_id + + @activity.defn(name='update_simple_metrics_adaptive_state') + def update_simple_metrics_adaptive_state(self, input_data: dict[str, Any]) -> dict | None: + """ + Update the adaptive frequency state on the schedule's `pipelines` doc. + + The orchestrator derives the schedule frequency from `adaptive_state.escalated` and + `frequency_alert`. State is read from Mongo, not from the workflow input, because the + input only refreshes on the orchestrator tick and would be stale while escalated. + `updated_at` is bumped only when `escalated` flips: that is what makes the orchestrator + rewrite the schedule, and a streak change alone does not change the frequency. + + Args: + input_data: Dict with: + - metadata (dict): Workflow metadata; `schedule_name` identifies the doc. + - simple_metrics (list[dict]): Rows from `calculate_simple_metrics`; empty for an + empty or under-sampled window, which counts as a non-breach run. + - thresholds (dict): `simple_metrics_thresholds` from `model_config` (may be None: no breach). + - interval_minutes (int): Window size, used to compute the recovery runs. + + Return: + The new state, or `None` when adaptive mode is off, Mongo is not configured, the doc + is missing or `frequency_alert` is invalid. + + Raises: + Exception: Mongo read/write failures, so the workflow run fails explicitly. + """ + metadata = input_data['metadata'] + if self.mongo_repository is None: + return None + + # Mongo errors propagate on purpose: the run must fail explicitly so the problem shows + # up in the Temporal metrics (PR #59 review). Metrics were already exported by then. + query = {'schedule_name': metadata['schedule_name']} + pipeline = self.mongo_repository.find_one(self._PIPELINES_COLLECTION, query, metadata) + if not pipeline: + # Likely misconfig: MONGODB_DATABASE_NAME differs from the orchestrator's, or the + # schedule_name does not match its doc. + self.warning( + f'No pipelines doc for schedule {metadata["schedule_name"]}: ' + 'adaptive state skipped', + metadata, + ) + return None + frequency_alert = pipeline.get('frequency_alert') + if not frequency_alert: + return None + + try: + runs = recovery_runs(input_data['interval_minutes'], str(frequency_alert)) + except (TypeError, ValueError): + # Config error, not a failure: the orchestrator ignores an invalid `frequency_alert` + # too (keeps `frequency`) and already sends SIMPLE_METRICS_FREQUENCY_ALERT_BELOW_FLOOR + # for it. `interval_minutes` is untyped Mongo config as well (e.g. "60" or null). + self.warning( + f'Invalid frequency_alert={frequency_alert!r} or ' + f'interval_minutes={input_data["interval_minutes"]!r}: adaptive state skipped', + metadata, + ) + return None + + breached = bool( + threshold_breaches(input_data['simple_metrics'], input_data.get('thresholds')) + ) + state = pipeline.get('adaptive_state') + if not isinstance(state, dict): + # Missing or hand-edited garbage: not escalated, same as the orchestrator reads it. + state = {'escalated': False, 'healthy_streak': 0} + new_state = next_adaptive_state(state, breached, runs) + if new_state == state: + return new_state + + update: dict[str, Any] = {'adaptive_state': new_state} + if new_state['escalated'] != bool(state.get('escalated')): + # BSON datetime, never a string: the orchestrator normalizes `updated_at` with + # `.replace(tzinfo=UTC).strftime(...)` and a string would fail its whole tick. + update['updated_at'] = now() + self.mongo_repository.update_one( + self._PIPELINES_COLLECTION, query, {'$set': update}, metadata + ) + + self.info(f'Adaptive state of {query["schedule_name"]}: {state} -> {new_state}', metadata) + return new_state diff --git a/laborious/activities/prediction_metadata.py b/laborious/activities/prediction_metadata.py deleted file mode 100644 index 1fdb519..0000000 --- a/laborious/activities/prediction_metadata.py +++ /dev/null @@ -1,132 +0,0 @@ -from temporalio import activity, workflow - -with workflow.unsafe.imports_passed_through(): - from typing import Any - - from bson import ObjectId - from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler - from sientia_do.observability.logger import Logger - from sientia_do.observability.metrics_controller import MetricsController - from sientia_do.observability.sientia_monitoring import SientiaMonitoring - from sientia_do.repository.mongodb_repository_sync import MongoDBRepository - - -class PredictionMetadata(SientiaMonitoring): - """ - Centralized, best-effort writer for prediction gate metadata (triggered filter, policy, - counter) to MongoDB. - - Gates only compute and return a metadata document; they never write it themselves. Keeping - the write here, instead of in each gate, puts the one real write point in the subworkflow - that's about to persist a `predictions` row — the only place that reliably knows whether one - is being written at all — and lets other metadata sources reuse the same activity instead of - duplicating write logic wherever they originate. - """ - - mongo_repository: MongoDBRepository | None = None - _PREDICTION_METADATA_COLLECTION = 'prediction_metadata' - - def __init__( - self, - mongo_repository: MongoDBRepository | None = None, - logger: Logger | None = None, - notification_handler: NotificationHandler | None = None, - metrics_controller: MetricsController | None = None, - ): - """ - Initialize the prediction metadata writer. - - Args: - mongo_repository: MongoDB, where gate metadata (triggered filter, policy, counter) is - written best-effort. `None` is a valid state (see `write_prediction_metadata`). - - Raises: - Exception: If BaseActivity initialization fails - """ - self.mongo_repository = mongo_repository - SientiaMonitoring.__init__( - self, - logger=logger, - notification_handler=notification_handler, - metrics_controller=metrics_controller, - ) - - def close(self) -> None: - """ - Close the prediction metadata writer and clean up resources. - """ - if self.mongo_repository is not None: - try: - self.mongo_repository.close() - finally: - self.mongo_repository = None - SientiaMonitoring.shutdown(self) - - def __del__(self): - self.close() - - @activity.defn(name='write_prediction_metadata') - def write_prediction_metadata(self, input_data: dict[str, Any]) -> str | None: - """ - Best-effort write of one or more gate metadata documents to Mongo, combined into a single - `prediction_metadata` document. - - Implements both business rules for this feature in one place: - - The prediction write must never be blocked or failed by a metadata write. A failed - `insert(...)` is swallowed, never re-raised, logged here with call-site context (which - `mongo_repository.insert(...)` itself can't know) — matching how `storage.py`'s - `cleanup_minio_objects_expired` also logs again at the call site even though the - repository already logs internally and emits its own Prometheus metric - (`MONGODB_WRITE_ERROR_COUNT`); the generic library log is for the metric/trace, this - one is for whoever is reading logs for this specific call. - - Nothing is written when there's nothing relevant to report. An empty/missing - `documents` list (or one full of falsy/empty entries) skips Mongo entirely, so callers - can call this unconditionally and get `None` back for "no metadata this time" instead - of deciding upfront whether there's something worth saving. - - Takes a *list* of documents, not a single one: each caller contributes its own document - (shaped like `{'filters': [...]}`), and this flattens every `filters` entry across all of - them into one `prediction_metadata` document before the single insert. Today only one - gate ever contributes a document per prediction cycle (the gate chain short-circuits on - the first non-`None` `path_flag`), but the list shape means this activity won't need to - change again once more than one source contributes metadata for the same cycle. - - `insert(...)` doesn't hand back the inserted document's `_id`, so it is generated here, - client-side, before the insert, as a plain string (not a BSON ObjectId) — matching this - codebase's convention of string `_id`s (e.g. the `models` collection) and the 24-hex-char - size of the `predictions.metadata_id` Postgres column that will reference it. - - Args: - input_data: Dict with: - - documents (list[dict[str, Any]]): Metadata payloads to combine, each shaped - like `{'filters': [...]}`. Missing/empty, or entries with no `filters`, skip - the write entirely. - - metadata (dict): Workflow metadata for contextual logging. - - Return: - The generated id, or `None` when there's nothing to write, Mongo isn't configured, or - the insert failed. - """ - metadata = input_data['metadata'] - documents = input_data.get('documents') or [] - - filters = [ - entry for document in documents if document for entry in document.get('filters', []) - ] - - if not filters or self.mongo_repository is None: - return None - - generated_id = str(ObjectId()) - - try: - self.mongo_repository.insert( - self._PREDICTION_METADATA_COLLECTION, - {'filters': filters, '_id': generated_id}, - metadata, - ) - except Exception as e: - self.error(f'Failed to write prediction metadata: {e}', metadata) - return None - - return generated_id diff --git a/laborious/activities/storage.py b/laborious/activities/storage.py index b4ae983..b59ddd8 100644 --- a/laborious/activities/storage.py +++ b/laborious/activities/storage.py @@ -184,7 +184,7 @@ class Storage(Postgres, SientiaMonitoring): REPEAT persists a new row (unlike STOP, which persists nothing), reusing the previous row's `prediction`/`response_time`/`prediction_status`/`prediction_confidence` values with a fresh `timestamp`/`created_at`. `metadata_id` references the `prediction_metadata` - document for this cycle, if any (see `PredictionMetadata.write_prediction_metadata`); + document for this cycle, if any (see `MongoActivities.write_prediction_metadata`); `None` is a valid value — the column is nullable. Args (query_items): diff --git a/laborious/worker/worker.py b/laborious/worker/worker.py index b44d829..c33fe9b 100644 --- a/laborious/worker/worker.py +++ b/laborious/worker/worker.py @@ -217,6 +217,7 @@ async def main(): activities.load_custom_query, activities.calculate_simple_metrics, activities.export_data_to_postgres, + activities.update_simple_metrics_adaptive_state, ], logger=logger, runtime='core', diff --git a/laborious/workflows/simple_metrics.py b/laborious/workflows/simple_metrics.py index 657964d..61531c1 100644 --- a/laborious/workflows/simple_metrics.py +++ b/laborious/workflows/simple_metrics.py @@ -31,6 +31,17 @@ class SimpleMetrics: model_config = input_data['model_config'] target_name = model_config['target'] + model_type = model_config.get('model_type') + thresholds = model_config.get('simple_metrics_thresholds') + min_samples = model_config.get('simple_metrics_min_samples') + # Untyped Mongo config: anything but a positive int (bool excluded) turns the guard off + # instead of raising in workflow code; falsy values mean "unset" and stay silent. + if type(min_samples) is not int or min_samples <= 0: + if min_samples: + workflow.logger.warning( + f'Invalid simple_metrics_min_samples={min_samples!r}: min samples guard off' + ) + min_samples = None # Built from internal Temporal workflow config, not user input (bandit B608). query = f""" @@ -60,37 +71,63 @@ class SimpleMetrics: start_to_close_timeout=timedelta(seconds=300), ) - if not target_data: - return + # Too few rows for a reliable value: metrics are still persisted, but a noisy short + # window must neither alert nor escalate (SIENTIAPDE-1995). + enough_samples = bool(target_data) and (not min_samples or len(target_data) >= min_samples) + if target_data and not enough_samples: + workflow.logger.info( + f'{len(target_data)} rows < simple_metrics_min_samples={min_samples}: ' + 'threshold alerting and escalation skipped' + ) - simple_metrics = await workflow.execute_local_activity_method( - Activities.calculate_simple_metrics, + simple_metrics = None + if target_data: + simple_metrics = await workflow.execute_local_activity_method( + Activities.calculate_simple_metrics, + { + **metadata, + 'model_id': model_id, + 'target_data': target_data, + 'metrics': input_data.get('metrics', ['rmse', 'mse', 'mae', 'r2']), + 'interval_minutes': interval_minutes, + 'model_type': model_type, + 'thresholds': thresholds if enough_samples else None, + }, + retry_policy=retry_policy, + start_to_close_timeout=timedelta(seconds=300), + ) + + if simple_metrics: + await workflow.execute_activity_method( + Activities.export_data_to_postgres, + { + **metadata, + 'data': simple_metrics, + 'schema': input_data['schema'], + 'table_name': input_data['target_table_name'], + 'timestamp_conversion': { + 'column': 'timestamp', + 'format': DATETIME_FORMAT_WITH_TZ, + }, + }, + retry_policy=retry_policy, + start_to_close_timeout=timedelta(seconds=300), + ) + + # No try/except: a failure here must fail the run so it shows up in the Temporal metrics + # (metrics are already persisted above). + await workflow.execute_activity_method( + Activities.update_simple_metrics_adaptive_state, { **metadata, - 'model_id': model_id, - 'target_data': target_data, - 'metrics': input_data.get('metrics', ['rmse', 'mse', 'mae', 'r2']), + # Only a valid breach escalates: an empty or under-sampled window is sent as no + # rows, which counts as a healthy run (spec decision 8). Always called, even + # without thresholds (no breach), so an escalated doc recovers after they are + # removed. + 'simple_metrics': (simple_metrics or []) if enough_samples else [], + 'thresholds': thresholds, 'interval_minutes': interval_minutes, }, retry_policy=retry_policy, - start_to_close_timeout=timedelta(seconds=300), - ) - - if not simple_metrics: - return - - await workflow.execute_activity_method( - Activities.export_data_to_postgres, - { - **metadata, - 'data': simple_metrics, - 'schema': input_data['schema'], - 'table_name': input_data['target_table_name'], - 'timestamp_conversion': { - 'column': 'timestamp', - 'format': DATETIME_FORMAT_WITH_TZ, - }, - }, - retry_policy=retry_policy, - start_to_close_timeout=timedelta(seconds=300), + start_to_close_timeout=timedelta(minutes=1), ) diff --git a/laborious/workflows/sub_workflows/prediction_process.py b/laborious/workflows/sub_workflows/prediction_process.py index 34999f4..d410a9f 100644 --- a/laborious/workflows/sub_workflows/prediction_process.py +++ b/laborious/workflows/sub_workflows/prediction_process.py @@ -266,7 +266,7 @@ class PredictionProcess: Return: The generated id, or `None` when there's nothing to write, including when the - activity call itself fails (see `PredictionMetadata.write_prediction_metadata`). + activity call itself fails (see `MongoActivities.write_prediction_metadata`). """ if not document: # Skip the remote activity call entirely rather than dispatching it just to have diff --git a/requirements-local.txt b/requirements-local.txt index 77aabf1..bb61155 100644 --- a/requirements-local.txt +++ b/requirements-local.txt @@ -6,7 +6,7 @@ sqlalchemy<2.1 asyncua==1.0.6 redis git+ssh://git@github.com/Aignosi/sientia-dataops-library.git@1.12.2 -git+ssh://git@github.com/Aignosi/sientia-model-library.git@0.10.0 +git+ssh://git@github.com/Aignosi/sientia-model-library.git@0.13.1 prometheus-client # fsspec/s3fs pinned as a pair: s3fs pins fsspec exactly, so a free `s3fs` can make the # resolver fail against evidently's fsspec>=2024.2.0. Versions match the @@ -22,4 +22,4 @@ hyperopt shap pycurl scipy<1.14.0 -scikit-learn==1.5.2 +scikit-learn==1.8.0 diff --git a/requirements.txt b/requirements.txt index e2cdfcb..f83c61b 100644 --- a/requirements.txt +++ b/requirements.txt @@ -6,7 +6,7 @@ sqlalchemy<2.1 asyncua==1.0.6 redis sientia_do>=1.12.2 -sientia_model>=0.8.2 +sientia_model>=0.12.0 prometheus-client # fsspec/s3fs pinned as a pair: s3fs pins fsspec exactly, so a free `s3fs` can make the # resolver fail against evidently's fsspec>=2024.2.0. Versions match the diff --git a/tests/laborious/activities/test_activities.py b/tests/laborious/activities/test_activities.py index f902b09..7dee4c4 100644 --- a/tests/laborious/activities/test_activities.py +++ b/tests/laborious/activities/test_activities.py @@ -5,8 +5,8 @@ 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.mongo import MongoActivities from laborious.activities.opc import OPC -from laborious.activities.prediction_metadata import PredictionMetadata from laborious.activities.storage import Storage @@ -14,7 +14,7 @@ from laborious.activities.storage import Storage @patch('laborious.activities.activities.MLFlow.__init__') @patch('laborious.activities.activities.OPC.__init__') @patch('laborious.activities.activities.Gates.__init__') -@patch('laborious.activities.activities.PredictionMetadata.__init__') +@patch('laborious.activities.activities.MongoActivities.__init__') @patch('laborious.activities.activities.ModelMetrics.__init__') @patch('laborious.activities.activities.API.__init__') @patch('laborious.activities.activities.MinioRepository') @@ -24,7 +24,7 @@ def test___init__( mock_minio_repository, mock_api_init, mock_model_metrics_init, - mock_prediction_metadata_init, + mock_mongo_activities_init, mock_gates_init, mock_opc_init, mock_mlflow_init, @@ -87,7 +87,7 @@ def test___init__( assert isinstance(activities, Gates) assert isinstance(activities, ModelMetrics) assert isinstance(activities, API) - assert isinstance(activities, PredictionMetadata) + assert isinstance(activities, MongoActivities) mock_storage_init.assert_called_once_with( ANY, @@ -131,7 +131,7 @@ def test___init__( metrics_controller=mock_metrics_controller.return_value, ) - mock_prediction_metadata_init.assert_called_once_with( + mock_mongo_activities_init.assert_called_once_with( ANY, mongo_repository=mongo_repository, logger=logger, @@ -172,7 +172,7 @@ def test___init__( @patch('laborious.activities.activities.MLFlow') @patch('laborious.activities.activities.OPC') @patch('laborious.activities.activities.Gates') -@patch('laborious.activities.activities.PredictionMetadata') +@patch('laborious.activities.activities.MongoActivities') @patch('laborious.activities.activities.ModelMetrics') @patch('laborious.activities.activities.API') @patch('laborious.activities.activities.MinioRepository') @@ -180,7 +180,7 @@ def test_shutdown( _mock_minio_repository, mock_api_init, mock_model_metrics_init, - mock_prediction_metadata_init, + mock_mongo_activities_init, mock_gates_init, mock_opc_init, mock_mlflow_init, @@ -243,7 +243,7 @@ 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() - mock_prediction_metadata_init.close.assert_called_once() + mock_mongo_activities_init.close.assert_called_once() @patch('laborious.activities.activities.SientiaMLflowRepository') @@ -252,7 +252,7 @@ def test_shutdown( @patch('laborious.activities.activities.MLFlow.__init__') @patch('laborious.activities.activities.OPC.__init__') @patch('laborious.activities.activities.Gates.__init__') -@patch('laborious.activities.activities.PredictionMetadata.__init__') +@patch('laborious.activities.activities.MongoActivities.__init__') @patch('laborious.activities.activities.ModelMetrics.__init__') @patch('laborious.activities.activities.API.__init__') @patch('laborious.activities.activities.MinioRepository') @@ -262,7 +262,7 @@ def test___init___builds_mlflow_repository_when_not_provided( mock_minio_repository, _mock_api_init, _mock_model_metrics_init, - _mock_prediction_metadata_init, + _mock_mongo_activities_init, _mock_gates_init, _mock_opc_init, _mock_mlflow_init, @@ -326,7 +326,7 @@ def test___init___builds_mlflow_repository_when_not_provided( @patch('laborious.activities.activities.MLFlow.__init__') @patch('laborious.activities.activities.OPC.__init__') @patch('laborious.activities.activities.Gates.__init__') -@patch('laborious.activities.activities.PredictionMetadata.__init__') +@patch('laborious.activities.activities.MongoActivities.__init__') @patch('laborious.activities.activities.ModelMetrics.__init__') @patch('laborious.activities.activities.API.__init__') @patch('laborious.activities.activities.MinioRepository') @@ -336,7 +336,7 @@ def test___init___builds_mongo_repository_when_not_provided( mock_minio_repository, _mock_api_init, _mock_model_metrics_init, - mock_prediction_metadata_init, + mock_mongo_activities_init, mock_gates_init, _mock_opc_init, _mock_mlflow_init, @@ -398,9 +398,9 @@ def test___init___builds_mongo_repository_when_not_provided( notification_handler=notification_handler, metrics_controller=mock_metrics_controller.return_value, ) - # The same repository built here is the one handed to PredictionMetadata, not a second + # The same repository built here is the one handed to MongoActivities, not a second # instance. - mock_prediction_metadata_init.assert_called_once_with( + mock_mongo_activities_init.assert_called_once_with( ANY, mongo_repository=mock_mongo_repository_cls.return_value, logger=logger, diff --git a/tests/laborious/activities/test_gates.py b/tests/laborious/activities/test_gates.py index b3aebe1..a7f95d7 100644 --- a/tests/laborious/activities/test_gates.py +++ b/tests/laborious/activities/test_gates.py @@ -153,8 +153,8 @@ def test_input_gate_with_filter(gates_activity): # Assert: the gate only computes the document now — it never writes anywhere, and never # checks path_flag to decide whether to build one (that decision moved to the caller, see - # `PredictionMetadata.write_prediction_metadata` and `prediction_process.py`), so even a STOP - # resolution still gets its document built. + # `MongoActivities.write_prediction_metadata` and `prediction_process.py`), so even a + # STOP resolution still gets its document built. assert (path_flag, confidence, comment) == ('STOP', -1, 'Input data with bad quality') assert document == {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]} gates_activity.debug.assert_called() diff --git a/tests/laborious/activities/test_model_metrics.py b/tests/laborious/activities/test_model_metrics.py index 7f5bdcd..ed7ce56 100644 --- a/tests/laborious/activities/test_model_metrics.py +++ b/tests/laborious/activities/test_model_metrics.py @@ -7,7 +7,13 @@ from sientia_do.observability.metrics_controller import MetricsController from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ from sientia_model.analytics.drift_analysis import DriftInsufficientDataError -from laborious.activities.model_metrics import ModelMetrics +from laborious.activities.model_metrics import ( + ModelMetrics, + frequency_to_seconds, + next_adaptive_state, + recovery_runs, + threshold_breaches, +) @fixture @@ -46,6 +52,23 @@ metadata = { } +def _mock_regression_metrics_class(calculate_return): + """Returns a patch context manager that mocks RegressionMetrics.""" + mock_instance = MagicMock() + mock_instance.calculate.return_value = calculate_return + mock_class = MagicMock(return_value=mock_instance) + mock_class.is_r2_supported = MagicMock(return_value=True) + mock_class.supported_metrics = MagicMock(return_value=['rmse', 'mse', 'mae', 'r2']) + return ( + patch( + 'laborious.activities.model_metrics.RegressionMetrics', + mock_class, + ), + mock_class, + mock_instance, + ) + + def test_calculate_drift_invalid_chunk_period(model_metrics_activity): # Arrange input_data = { @@ -694,7 +717,6 @@ def test_get_drift_metrics_dataframe_error( def test_calculate_simple_metrics_success_all_metrics(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28', '2023-05-26 11:12:29'], @@ -711,28 +733,28 @@ def test_calculate_simple_metrics_success_all_metrics(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) - - # Assert - assert len(result['metric']) == 4 - assert 'rmse' in result['metric'].values - assert 'mse' in result['metric'].values - assert 'mae' in result['metric'].values - assert 'r2' in result['metric'].values - assert all(model_id == 'test_model_id' for model_id in result['model_id'].values) - assert all(timestamp == '2023-05-26 11:12:29' for timestamp in result['timestamp'].values) - assert all(data_size == 3 for data_size in result['data_size'].values) - assert all(interval_minutes == 5 for interval_minutes in result['interval_minutes'].values) - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['rmse', 'mse', 'mae', 'r2']", - metadata['metadata'], + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'rmse', 'value': 0.1}, + {'metric': 'mse', 'value': 0.01}, + {'metric': 'mae', 'value': 0.1}, + {'metric': 'r2', 'value': 0.99}, + ] ) - model_metrics_activity.debug.assert_called_once() + + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 4 + assert set(result['metric'].values) == {'rmse', 'mse', 'mae', 'r2'} + assert all(mid == 'test_model_id' for mid in result['model_id'].values) + assert all(ts == '2023-05-26 11:12:29' for ts in result['timestamp'].values) + assert all(ds == 3 for ds in result['data_size'].values) + assert all(im == 5 for im in result['interval_minutes'].values) + mock_instance.calculate.assert_called_once_with(['rmse', 'mse', 'mae', 'r2']) def test_calculate_simple_metrics_success_rmse_only(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -749,23 +771,23 @@ def test_calculate_simple_metrics_success_rmse_only(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'rmse', 'value': 0.1}] + ) - # Assert - assert len(result['metric']) == 1 + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 assert result['metric'].values[0] == 'rmse' assert result['model_id'].values[0] == 'test_model_id' assert result['timestamp'].values[0] == '2023-05-26 11:12:28' assert result['data_size'].values[0] == 2 assert result['interval_minutes'].values[0] == 5 - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['rmse']", metadata['metadata'] - ) + mock_instance.calculate.assert_called_once_with(['rmse']) def test_calculate_simple_metrics_success_mse_only(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -782,23 +804,23 @@ def test_calculate_simple_metrics_success_mse_only(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'mse', 'value': 0.01}] + ) - # Assert - assert len(result['metric']) == 1 + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 assert result['metric'].values[0] == 'mse' assert result['model_id'].values[0] == 'test_model_id' assert result['timestamp'].values[0] == '2023-05-26 11:12:28' assert result['data_size'].values[0] == 2 assert result['interval_minutes'].values[0] == 5 - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['mse']", metadata['metadata'] - ) + mock_instance.calculate.assert_called_once_with(['mse']) def test_calculate_simple_metrics_success_mae_only(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -815,23 +837,23 @@ def test_calculate_simple_metrics_success_mae_only(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'mae', 'value': 0.1}] + ) - # Assert - assert len(result['metric']) == 1 + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 assert result['metric'].values[0] == 'mae' assert result['model_id'].values[0] == 'test_model_id' assert result['timestamp'].values[0] == '2023-05-26 11:12:28' assert result['data_size'].values[0] == 2 assert result['interval_minutes'].values[0] == 5 - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['mae']", metadata['metadata'] - ) + mock_instance.calculate.assert_called_once_with(['mae']) def test_calculate_simple_metrics_success_r2_only(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -848,24 +870,24 @@ def test_calculate_simple_metrics_success_r2_only(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'r2', 'value': 0.95}] + ) - # Assert - assert len(result['metric']) == 1 + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 assert result['metric'].values[0] == 'r2' assert result['model_id'].values[0] == 'test_model_id' assert result['timestamp'].values[0] == '2023-05-26 11:12:28' assert result['data_size'].values[0] == 2 assert result['interval_minutes'].values[0] == 5 - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['r2']", metadata['metadata'] - ) + mock_instance.calculate.assert_called_once_with(['r2']) -def test_calculate_simple_metrics_r2_zero_ss_tot(model_metrics_activity): - # Arrange - # All target values are the same, so ss_tot will be 0 +def test_calculate_simple_metrics_r2_zero_variance_delegates_to_lib(model_metrics_activity): + """r2 zero-variance is now the lib's responsibility. Activity just passes through.""" target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -882,24 +904,18 @@ def test_calculate_simple_metrics_r2_zero_ss_tot(model_metrics_activity): 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) - - # Assert - assert len(result['metric']) == 1 - assert result['metric'].values[0] == 'r2' - assert result['value'].values[0] == 0.0 # Should return 0.0 when ss_tot == 0 - assert result['model_id'].values[0] == 'test_model_id' - assert result['timestamp'].values[0] == '2023-05-26 11:12:28' - assert result['data_size'].values[0] == 2 - assert result['interval_minutes'].values[0] == 5 - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['r2']", metadata['metadata'] + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'r2', 'value': 0.0}] ) + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert result['value'].values[0] == 0.0 + mock_instance.calculate.assert_called_once_with(['r2']) + def test_calculate_simple_metrics_success_multiple_metrics_subset(model_metrics_activity): - # Arrange target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28', '2023-05-26 11:12:29'], @@ -916,23 +932,26 @@ def test_calculate_simple_metrics_success_multiple_metrics_subset(model_metrics_ 'interval_minutes': 5, } - # Act - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) - - # Assert - assert len(result['metric']) == 2 - assert 'rmse' in result['metric'].values - assert 'mae' in result['metric'].values - assert all(model_id == 'test_model_id' for model_id in result['model_id'].values) - assert all(timestamp == '2023-05-26 11:12:29' for timestamp in result['timestamp'].values) - assert all(data_size == 3 for data_size in result['data_size'].values) - assert all(interval_minutes == 5 for interval_minutes in result['interval_minutes'].values) - model_metrics_activity.info.assert_called_once_with( - "Calculating simple metrics for model test_model_id: ['rmse', 'mae']", metadata['metadata'] + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'rmse', 'value': 0.1}, + {'metric': 'mae', 'value': 0.1}, + ] ) + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) -def test_calculate_simple_metrics_unknown_metric_ignored(model_metrics_activity): + assert len(result) == 2 + assert set(result['metric'].values) == {'rmse', 'mae'} + assert all(mid == 'test_model_id' for mid in result['model_id'].values) + assert all(ts == '2023-05-26 11:12:29' for ts in result['timestamp'].values) + assert all(ds == 3 for ds in result['data_size'].values) + assert all(im == 5 for im in result['interval_minutes'].values) + mock_instance.calculate.assert_called_once_with(['rmse', 'mae']) + + +def test_calculate_simple_metrics_unknown_metric_raises(model_metrics_activity): target_data = DataFrame( { 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], @@ -949,10 +968,218 @@ def test_calculate_simple_metrics_unknown_metric_ignored(model_metrics_activity) 'interval_minutes': 5, } - result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + mock_instance = MagicMock() + mock_instance.calculate.side_effect = ValueError('Unknown metric: unknown_metric') + mock_class = MagicMock(return_value=mock_instance) + mock_class.is_r2_supported = MagicMock(return_value=True) - assert len(result['metric']) == 1 + with patch('laborious.activities.model_metrics.RegressionMetrics', mock_class): + with raises(ValueError, match='Unknown metric'): + model_metrics_activity.calculate_simple_metrics(input_data) + + +def test_calculate_simple_metrics_r2_skipped_for_nonlinear_model(model_metrics_activity): + """When model_type is non-linear, r2 is excluded and a warning notification fires.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['rmse', 'r2'], + 'interval_minutes': 5, + 'model_type': 'XGBoost', + } + + mock_instance = MagicMock() + mock_instance.calculate.return_value = [{'metric': 'rmse', 'value': 0.1}] + mock_class = MagicMock(return_value=mock_instance) + mock_class.is_r2_supported = MagicMock(return_value=False) + + with patch('laborious.activities.model_metrics.RegressionMetrics', mock_class): + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 assert result['metric'].values[0] == 'rmse' + mock_class.is_r2_supported.assert_called_once_with('XGBoost') + mock_instance.calculate.assert_called_once_with(['rmse']) + model_metrics_activity.warning.assert_called_once() + model_metrics_activity.send_notification.assert_called_once() + call_kwargs = model_metrics_activity.send_notification.call_args.kwargs + assert call_kwargs['notification_id'] == 'SIMPLE_METRICS_R2_UNSUPPORTED' + assert call_kwargs['level'] == NotificationLevel.WARNING + + +def test_calculate_simple_metrics_no_model_type_includes_r2(model_metrics_activity): + """When model_type is None (legacy input), r2 is included without check.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['r2'], + 'interval_minutes': 5, + # no model_type key + } + + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [{'metric': 'r2', 'value': 0.95}] + ) + + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert result['metric'].values[0] == 'r2' + mock_class.is_r2_supported.assert_not_called() + + +def test_calculate_simple_metrics_threshold_breach_rmse(model_metrics_activity): + """When rmse exceeds rmse_max, a WARNING notification fires.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['rmse'], + 'interval_minutes': 5, + 'thresholds': {'rmse_max': 0.05}, + } + + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'rmse', 'value': 0.1}, + ] + ) + + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 + model_metrics_activity.send_notification.assert_called_once() + call_kwargs = model_metrics_activity.send_notification.call_args.kwargs + assert call_kwargs['notification_id'] == 'SIMPLE_METRICS_THRESHOLD_BREACH' + assert call_kwargs['level'] == NotificationLevel.WARNING + assert 'rmse' in call_kwargs['message'] + assert '(window 5 min)' in call_kwargs['message'] + # Model-level trigger so the orchestrator dedup merges windows (spec decision 9). + assert call_kwargs['metadata']['schedule_name'] == 'test_model_id-simple_metrics' + assert call_kwargs['metadata']['model_name'] == 'test_model' + + +def test_calculate_simple_metrics_threshold_no_breach(model_metrics_activity): + """When rmse is below rmse_max, no notification fires.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['rmse'], + 'interval_minutes': 5, + 'thresholds': {'rmse_max': 1.0}, + } + + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'rmse', 'value': 0.1}, + ] + ) + + with patcher: + model_metrics_activity.calculate_simple_metrics(input_data) + + model_metrics_activity.send_notification.assert_not_called() + + +def test_calculate_simple_metrics_threshold_r2_below_min(model_metrics_activity): + """When r2 drops below r2_min, a WARNING notification fires.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['r2'], + 'interval_minutes': 5, + 'thresholds': {'r2_min': 0.95}, + } + + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'r2', 'value': 0.8}, + ] + ) + + with patcher: + model_metrics_activity.calculate_simple_metrics(input_data) + + model_metrics_activity.send_notification.assert_called_once() + call_kwargs = model_metrics_activity.send_notification.call_args.kwargs + assert 'r2' in call_kwargs['message'] + + +def test_calculate_simple_metrics_no_thresholds_no_alert(model_metrics_activity): + """When thresholds is None (not configured), no alerting, no crash.""" + target_data = DataFrame( + { + 'timestamp': ['2023-05-26 11:12:27', '2023-05-26 11:12:28'], + 'target': [1.0, 2.0], + 'prediction': [1.1, 2.1], + } + ) + + input_data = { + **metadata, + 'model_id': 'test_model_id', + 'target_data': target_data.to_dict(), + 'metrics': ['rmse'], + 'interval_minutes': 5, + # no thresholds key + } + + patcher, mock_class, mock_instance = _mock_regression_metrics_class( + [ + {'metric': 'rmse', 'value': 999.0}, + ] + ) + + with patcher: + result = DataFrame(model_metrics_activity.calculate_simple_metrics(input_data)) + + assert len(result) == 1 + model_metrics_activity.send_notification.assert_not_called() @mark.parametrize( @@ -1000,3 +1227,79 @@ def test_to_naive_utc_keeps_naive_timestamps(): assert result.dt.tz is None assert result.tolist() == [Timestamp('2024-01-01 06:00:00')] + + +@mark.parametrize( + 'rows, thresholds, expected_count', + [ + ([{'metric': 'rmse', 'value': 2.0}], {'rmse_max': 1.0}, 1), + ([{'metric': 'rmse', 'value': 0.5}], {'rmse_max': 1.0}, 0), + ([{'metric': 'r2', 'value': 0.5}], {'r2_min': 0.9}, 1), + ([{'metric': 'r2', 'value': 0.95}], {'r2_min': 0.9}, 0), + ([{'metric': 'rmse', 'value': 1.0}], {'rmse_max': 1.0}, 0), # equal is not a breach + ([{'metric': 'r2', 'value': 0.9}], {'r2_min': 0.9}, 0), + ([{'metric': 'mae', 'value': 99.0}], {'rmse_max': 1.0}, 0), + ([{'metric': 'rmse', 'value': 99.0}], None, 0), + ([], {'rmse_max': 1.0}, 0), + ( + [{'metric': 'rmse', 'value': 2.0}, {'metric': 'r2', 'value': 0.1}], + {'rmse_max': 1.0, 'r2_min': 0.9}, + 2, + ), + ], +) +def test_threshold_breaches(rows, thresholds, expected_count): + assert len(threshold_breaches(rows, thresholds)) == expected_count + + +def _state(escalated: bool, streak: int) -> dict: + return {'escalated': escalated, 'healthy_streak': streak} + + +@mark.parametrize( + 'state, breached, runs, expected', + [ + ({}, True, 3, _state(True, 0)), # first breach escalates + ({}, False, 3, _state(False, 0)), # missing state = healthy + (_state(False, 0), False, 3, _state(False, 0)), # healthy stays healthy + (_state(True, 0), False, 3, _state(True, 1)), # healthy run while escalated + (_state(True, 1), False, 3, _state(True, 2)), + (_state(True, 2), False, 3, _state(False, 0)), # N-th healthy run recovers + (_state(True, 2), True, 3, _state(True, 0)), # breach resets the streak + (_state(True, 0), False, 1, _state(False, 0)), # N=1 recovers on first healthy run + (_state(True, None), False, 3, _state(True, 1)), # hand-edited streak restarts + (_state(True, 'x'), False, 3, _state(True, 1)), + (_state(True, float('inf')), False, 3, _state(True, 1)), + ], +) +def test_next_adaptive_state(state, breached, runs, expected): + assert next_adaptive_state(state, breached, runs) == expected + + +@mark.parametrize( + 'frequency, seconds', + [('30s', 30), ('5m', 300), ('2h', 7200), ('1d', 86400)], +) +def test_frequency_to_seconds(frequency, seconds): + assert frequency_to_seconds(frequency) == seconds + + +@mark.parametrize('frequency', ['', '5', 'm', '5x', 'abc', '0m']) +def test_frequency_to_seconds_rejects_invalid(frequency): + with raises(ValueError): + frequency_to_seconds(frequency) + + +@mark.parametrize( + 'interval_minutes, frequency_alert, expected', + [ + (60, '5m', 12), # 60/5 = 12, at the cap + (60, '15m', 4), + (60, '7m', 9), # ceil(8.57) + (60, '30s', 12), # 120 capped at 12 + (1440, '1h', 12), # 24 capped at 12 + (5, '1h', 1), # window shorter than one run: at least 1 + ], +) +def test_recovery_runs(interval_minutes, frequency_alert, expected): + assert recovery_runs(interval_minutes, frequency_alert) == expected diff --git a/tests/laborious/activities/test_mongo.py b/tests/laborious/activities/test_mongo.py new file mode 100644 index 0000000..837e410 --- /dev/null +++ b/tests/laborious/activities/test_mongo.py @@ -0,0 +1,357 @@ +from datetime import datetime +from unittest.mock import MagicMock, call, patch + +from pytest import fixture, mark, raises +from sientia_do.observability.metrics_controller import MetricsController +from sientia_do.observability.sientia_monitoring import SientiaMonitoring + +from laborious.activities.mongo import MongoActivities + + +@fixture +def mongo() -> MagicMock: + return MagicMock() + + +@fixture +def prediction_metadata(mongo: MagicMock): + activity = MongoActivities( + mongo_repository=mongo, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + activity.error = MagicMock() + return activity + + +metadata = { + 'metadata': { + 'model_id': 'test_model', + 'model_name': 'test_model', + 'workflow_name': 'test_workflow', + 'schema_name': 'test_schedule', + }, +} + + +def test_write_prediction_metadata_no_documents_skips_mongo(prediction_metadata, mongo: MagicMock): + # Act + result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': []}) + + # Assert + assert result is None + mongo.insert.assert_not_called() + + +def test_write_prediction_metadata_missing_documents_key_skips_mongo( + prediction_metadata, mongo: MagicMock +): + # Act + result = prediction_metadata.write_prediction_metadata({**metadata}) + + # Assert + assert result is None + mongo.insert.assert_not_called() + + +def test_write_prediction_metadata_documents_with_no_filters_skips_mongo( + prediction_metadata, mongo: MagicMock +): + # Arrange: a falsy document (None) and one with an empty `filters` list — neither contributes + # anything to write. + documents = [None, {'filters': []}] + + # Act + result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents}) + + # Assert + assert result is None + mongo.insert.assert_not_called() + + +def test_write_prediction_metadata_not_configured_returns_none(): + # Arrange + activity = MongoActivities( + mongo_repository=None, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + activity.error = MagicMock() + document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]} + + # Act + result = activity.write_prediction_metadata({**metadata, 'documents': [document]}) + + # Assert + assert result is None + activity.error.assert_not_called() + + +def test_write_prediction_metadata_success_generates_id_and_inserts( + prediction_metadata, mongo: MagicMock +): + # Arrange + document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 3}]} + + # Act + result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]}) + + # Assert: a plain 24-hex-char string id (matches predictions.metadata_id's varchar(24)), + # not a BSON ObjectId — `insert(...)` never hands one back, so it's generated client-side. + assert result is not None + assert len(result) == 24 + int(result, 16) # raises ValueError if not hex + + collection, inserted_document, inserted_metadata = mongo.insert.call_args.args + assert collection == MongoActivities._PREDICTION_METADATA_COLLECTION + assert inserted_document == {'filters': document['filters'], '_id': result} + assert inserted_metadata == metadata['metadata'] + + +def test_write_prediction_metadata_combines_multiple_documents( + prediction_metadata, mongo: MagicMock +): + # Arrange: two documents (e.g. from two different sources contributing metadata for the same + # prediction cycle) — their `filters` entries are flattened into one combined document. + documents = [ + {'filters': [{'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}]}, + { + 'filters': [ + {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2} + ] + }, + ] + + # Act + result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents}) + + # Assert + assert result is not None + _, inserted_document, _ = mongo.insert.call_args.args + assert inserted_document['filters'] == [ + {'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}, + {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2}, + ] + + +def test_write_prediction_metadata_insert_failure_swallowed(prediction_metadata, mongo: MagicMock): + # Arrange + mongo.insert.side_effect = RuntimeError('connection reset') + document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]} + + # Act + result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]}) + + # Assert: the prediction write must never be blocked or failed by this + assert result is None + prediction_metadata.error.assert_called_once() + + +@patch.object(SientiaMonitoring, 'shutdown') +def test_close_disposes_mongo_repository(mock_shutdown): + """``MongoActivities.close`` should close the Mongo client and clear the reference.""" + mongo = MagicMock() + activity = MongoActivities( + mongo_repository=mongo, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + activity.close() + mongo.close.assert_called_once() + assert activity.mongo_repository is None + # Other instances collected by GC during the test also hit the patched shutdown. + assert mock_shutdown.call_args_list.count(call(activity)) == 1 + + +@patch.object(SientiaMonitoring, 'shutdown') +def test_close_without_mongo_repository(mock_shutdown): + """When no Mongo repository is configured, ``close`` only shuts down monitoring.""" + activity = MongoActivities( + mongo_repository=None, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + activity.close() + # Other instances collected by GC during the test also hit the patched shutdown. + assert mock_shutdown.call_args_list.count(call(activity)) == 1 + + +_SM_METADATA = {**metadata['metadata'], 'schedule_name': 'm1-simple_metrics-60m'} +_BREACH_INPUT = { + 'metadata': _SM_METADATA, + 'simple_metrics': [{'metric': 'rmse', 'value': 2.0}], + 'thresholds': {'rmse_max': 1.0}, + 'interval_minutes': 60, +} +_HEALTHY_INPUT = {**_BREACH_INPUT, 'simple_metrics': [{'metric': 'rmse', 'value': 0.5}]} +_EMPTY_INPUT = {**_BREACH_INPUT, 'simple_metrics': []} + + +@fixture +def mongo_activities(mongo: MagicMock): + activity = MongoActivities( + mongo_repository=mongo, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + activity.info = MagicMock() + activity.warning = MagicMock() + return activity + + +def _pipeline(**fields) -> dict: + return {'schedule_name': 'm1-simple_metrics-60m', 'frequency': '1h', **fields} + + +def _set_of(mongo: MagicMock) -> dict: + collection, query, update, _metadata = mongo.update_one.call_args.args + assert collection == 'pipelines' + assert query == {'schedule_name': 'm1-simple_metrics-60m'} + return update['$set'] + + +def test_adaptive_state_no_mongo_is_noop(): + activity = MongoActivities( + mongo_repository=None, + logger=MagicMock(), + notification_handler=MagicMock(), + metrics_controller=MagicMock(spec=MetricsController), + ) + assert activity.update_simple_metrics_adaptive_state(_BREACH_INPUT) is None + + +def test_adaptive_state_missing_pipeline_doc_is_noop(mongo_activities, mongo): + mongo.find_one.return_value = None + assert mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) is None + mongo.update_one.assert_not_called() + # Likely misconfig (other MONGODB_DATABASE_NAME or schedule_name mismatch): not silent. + mongo_activities.warning.assert_called_once() + assert 'm1-simple_metrics-60m' in mongo_activities.warning.call_args.args[0] + + +def test_adaptive_state_without_frequency_alert_is_noop(mongo_activities, mongo): + mongo.find_one.return_value = _pipeline() + assert mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) is None + mongo.update_one.assert_not_called() + mongo_activities.warning.assert_not_called() + + +def test_adaptive_state_breach_escalates_and_bumps_updated_at(mongo_activities, mongo): + mongo.find_one.return_value = _pipeline(frequency_alert='5m') + + result = mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) + + assert result == {'escalated': True, 'healthy_streak': 0} + update = _set_of(mongo) + assert update['adaptive_state'] == {'escalated': True, 'healthy_streak': 0} + # BSON datetime, never a string: the orchestrator calls .replace(tzinfo=...) on it. + assert isinstance(update['updated_at'], datetime) + assert update['updated_at'].tzinfo is not None + + +def test_adaptive_state_healthy_run_increments_streak_without_updated_at(mongo_activities, mongo): + # 60 min window / 5m = N 12, so one healthy run only moves the streak. + mongo.find_one.return_value = _pipeline( + frequency_alert='5m', adaptive_state={'escalated': True, 'healthy_streak': 0} + ) + + result = mongo_activities.update_simple_metrics_adaptive_state(_HEALTHY_INPUT) + + assert result == {'escalated': True, 'healthy_streak': 1} + assert _set_of(mongo) == {'adaptive_state': {'escalated': True, 'healthy_streak': 1}} + + +def test_adaptive_state_recovers_after_window_renewed(mongo_activities, mongo): + # 60 min window / 30m = N 2: second consecutive healthy run recovers. + mongo.find_one.return_value = _pipeline( + frequency_alert='30m', adaptive_state={'escalated': True, 'healthy_streak': 1} + ) + + result = mongo_activities.update_simple_metrics_adaptive_state(_HEALTHY_INPUT) + + assert result == {'escalated': False, 'healthy_streak': 0} + update = _set_of(mongo) + assert update['adaptive_state'] == {'escalated': False, 'healthy_streak': 0} + assert isinstance(update['updated_at'], datetime) + + +def test_adaptive_state_empty_window_counts_as_healthy_run(mongo_activities, mongo): + # Spec decision 8: an empty/under-sampled window never escalates but counts toward recovery. + mongo.find_one.return_value = _pipeline( + frequency_alert='5m', adaptive_state={'escalated': True, 'healthy_streak': 3} + ) + + result = mongo_activities.update_simple_metrics_adaptive_state(_EMPTY_INPUT) + + assert result == {'escalated': True, 'healthy_streak': 4} + + +def test_adaptive_state_empty_window_never_escalates(mongo_activities, mongo): + mongo.find_one.return_value = _pipeline(frequency_alert='5m') + + result = mongo_activities.update_simple_metrics_adaptive_state(_EMPTY_INPUT) + + assert result == {'escalated': False, 'healthy_streak': 0} + mongo.update_one.assert_not_called() + + +def test_adaptive_state_unchanged_does_not_write(mongo_activities, mongo): + mongo.find_one.return_value = _pipeline( + frequency_alert='5m', adaptive_state={'escalated': False, 'healthy_streak': 0} + ) + + result = mongo_activities.update_simple_metrics_adaptive_state(_HEALTHY_INPUT) + + assert result == {'escalated': False, 'healthy_streak': 0} + mongo.update_one.assert_not_called() + + +@mark.parametrize('frequency_alert', ['5min', '0m', 5]) +def test_adaptive_state_invalid_frequency_alert_is_skipped( + mongo_activities, mongo, frequency_alert +): + mongo.find_one.return_value = _pipeline(frequency_alert=frequency_alert) + + assert mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) is None + mongo.update_one.assert_not_called() + mongo_activities.warning.assert_called_once() + + +@mark.parametrize('interval_minutes', ['60', None]) +def test_adaptive_state_invalid_interval_minutes_is_skipped( + mongo_activities, mongo, interval_minutes +): + # Untyped Mongo config: "60" * 60 is a string, None * 60 a TypeError. + mongo.find_one.return_value = _pipeline(frequency_alert='5m') + + result = mongo_activities.update_simple_metrics_adaptive_state( + {**_BREACH_INPUT, 'interval_minutes': interval_minutes} + ) + + assert result is None + mongo.update_one.assert_not_called() + mongo_activities.warning.assert_called_once() + + +@mark.parametrize('adaptive_state', ['garbage', ['escalated']]) +def test_adaptive_state_malformed_state_is_not_escalated(mongo_activities, mongo, adaptive_state): + mongo.find_one.return_value = _pipeline(frequency_alert='5m', adaptive_state=adaptive_state) + + result = mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) + + assert result == {'escalated': True, 'healthy_streak': 0} + mongo.update_one.assert_called_once() + + +@mark.parametrize('failing_call', ['find_one', 'update_one']) +def test_adaptive_state_mongo_error_propagates(mongo_activities, mongo, failing_call): + mongo.find_one.return_value = _pipeline(frequency_alert='5m') + getattr(mongo, failing_call).side_effect = RuntimeError('mongo down') + + with raises(RuntimeError, match='mongo down'): + mongo_activities.update_simple_metrics_adaptive_state(_BREACH_INPUT) diff --git a/tests/laborious/activities/test_prediction_metadata.py b/tests/laborious/activities/test_prediction_metadata.py deleted file mode 100644 index b7c1143..0000000 --- a/tests/laborious/activities/test_prediction_metadata.py +++ /dev/null @@ -1,179 +0,0 @@ -from unittest.mock import MagicMock, call, patch - -from pytest import fixture -from sientia_do.observability.metrics_controller import MetricsController -from sientia_do.observability.sientia_monitoring import SientiaMonitoring - -from laborious.activities.prediction_metadata import PredictionMetadata - - -@fixture -def mongo() -> MagicMock: - return MagicMock() - - -@fixture -def prediction_metadata(mongo: MagicMock): - activity = PredictionMetadata( - mongo_repository=mongo, - logger=MagicMock(), - notification_handler=MagicMock(), - metrics_controller=MagicMock(spec=MetricsController), - ) - activity.error = MagicMock() - return activity - - -metadata = { - 'metadata': { - 'model_id': 'test_model', - 'model_name': 'test_model', - 'workflow_name': 'test_workflow', - 'schema_name': 'test_schedule', - }, -} - - -def test_write_prediction_metadata_no_documents_skips_mongo(prediction_metadata, mongo: MagicMock): - # Act - result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': []}) - - # Assert - assert result is None - mongo.insert.assert_not_called() - - -def test_write_prediction_metadata_missing_documents_key_skips_mongo( - prediction_metadata, mongo: MagicMock -): - # Act - result = prediction_metadata.write_prediction_metadata({**metadata}) - - # Assert - assert result is None - mongo.insert.assert_not_called() - - -def test_write_prediction_metadata_documents_with_no_filters_skips_mongo( - prediction_metadata, mongo: MagicMock -): - # Arrange: a falsy document (None) and one with an empty `filters` list — neither contributes - # anything to write. - documents = [None, {'filters': []}] - - # Act - result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents}) - - # Assert - assert result is None - mongo.insert.assert_not_called() - - -def test_write_prediction_metadata_not_configured_returns_none(): - # Arrange - activity = PredictionMetadata( - mongo_repository=None, - logger=MagicMock(), - notification_handler=MagicMock(), - metrics_controller=MagicMock(spec=MetricsController), - ) - activity.error = MagicMock() - document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]} - - # Act - result = activity.write_prediction_metadata({**metadata, 'documents': [document]}) - - # Assert - assert result is None - activity.error.assert_not_called() - - -def test_write_prediction_metadata_success_generates_id_and_inserts( - prediction_metadata, mongo: MagicMock -): - # Arrange - document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 3}]} - - # Act - result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]}) - - # Assert: a plain 24-hex-char string id (matches predictions.metadata_id's varchar(24)), - # not a BSON ObjectId — `insert(...)` never hands one back, so it's generated client-side. - assert result is not None - assert len(result) == 24 - int(result, 16) # raises ValueError if not hex - - collection, inserted_document, inserted_metadata = mongo.insert.call_args.args - assert collection == PredictionMetadata._PREDICTION_METADATA_COLLECTION - assert inserted_document == {'filters': document['filters'], '_id': result} - assert inserted_metadata == metadata['metadata'] - - -def test_write_prediction_metadata_combines_multiple_documents( - prediction_metadata, mongo: MagicMock -): - # Arrange: two documents (e.g. from two different sources contributing metadata for the same - # prediction cycle) — their `filters` entries are flattened into one combined document. - documents = [ - {'filters': [{'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}]}, - { - 'filters': [ - {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2} - ] - }, - ] - - # Act - result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents}) - - # Assert - assert result is not None - _, inserted_document, _ = mongo.insert.call_args.args - assert inserted_document['filters'] == [ - {'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}, - {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2}, - ] - - -def test_write_prediction_metadata_insert_failure_swallowed(prediction_metadata, mongo: MagicMock): - # Arrange - mongo.insert.side_effect = RuntimeError('connection reset') - document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]} - - # Act - result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]}) - - # Assert: the prediction write must never be blocked or failed by this - assert result is None - prediction_metadata.error.assert_called_once() - - -@patch.object(SientiaMonitoring, 'shutdown') -def test_close_disposes_mongo_repository(mock_shutdown): - """``PredictionMetadata.close`` should close the Mongo client and clear the reference.""" - mongo = MagicMock() - activity = PredictionMetadata( - mongo_repository=mongo, - logger=MagicMock(), - notification_handler=MagicMock(), - metrics_controller=MagicMock(spec=MetricsController), - ) - activity.close() - mongo.close.assert_called_once() - assert activity.mongo_repository is None - # Other instances collected by GC during the test also hit the patched shutdown. - assert mock_shutdown.call_args_list.count(call(activity)) == 1 - - -@patch.object(SientiaMonitoring, 'shutdown') -def test_close_without_mongo_repository(mock_shutdown): - """When no Mongo repository is configured, ``close`` only shuts down monitoring.""" - activity = PredictionMetadata( - mongo_repository=None, - logger=MagicMock(), - notification_handler=MagicMock(), - metrics_controller=MagicMock(spec=MetricsController), - ) - activity.close() - # Other instances collected by GC during the test also hit the patched shutdown. - assert mock_shutdown.call_args_list.count(call(activity)) == 1 diff --git a/tests/laborious/worker/test_worker.py b/tests/laborious/worker/test_worker.py index a4c1898..c9ded87 100644 --- a/tests/laborious/worker/test_worker.py +++ b/tests/laborious/worker/test_worker.py @@ -16,6 +16,7 @@ def _build_fake_activities(): inst.export_data_to_postgres = MagicMock() inst.load_custom_query = MagicMock() inst.calculate_simple_metrics = MagicMock() + inst.update_simple_metrics_adaptive_state = MagicMock() inst.get_reference_data = MagicMock() inst.calculate_drift = MagicMock() inst.request_predict = MagicMock() @@ -188,6 +189,11 @@ async def test_main_success_exit_zero(monkeypatch): # Drift and SimpleMetrics load no model code, so every runtime shares the `core` queues. assert prepare_calls[1].kwargs['runtime'] == 'core' assert prepare_calls[2].kwargs['runtime'] == 'core' + # prepare_calls[1] is the SimpleMetrics worker (order: MinimalRetrain, SimpleMetrics, Drift, PredictionsBatch). + assert ( + fake_activities.update_simple_metrics_adaptive_state + in prepare_calls[1].kwargs['activities'] + ) @mark.asyncio diff --git a/tests/laborious/workflows/test_simple_metrics.py b/tests/laborious/workflows/test_simple_metrics.py index dbda735..a3c8632 100644 --- a/tests/laborious/workflows/test_simple_metrics.py +++ b/tests/laborious/workflows/test_simple_metrics.py @@ -1,6 +1,6 @@ -from unittest.mock import ANY, AsyncMock, call, patch +from unittest.mock import ANY, AsyncMock, MagicMock, call, patch -from pytest import fixture, mark +from pytest import fixture, mark, raises from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ from laborious.activities.activities import Activities @@ -42,12 +42,17 @@ async def test_run(workflow_mock: AsyncMock, simple_metrics: SimpleMetrics): target_data = {'data': 'test_target_data'} simple_metrics_data = {'metrics': 'test_simple_metrics_data'} - workflow_mock.execute_activity_method = AsyncMock(side_effect=[target_data, None]) + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[target_data, None, None]) workflow_mock.execute_local_activity_method = AsyncMock(return_value=simple_metrics_data) # Act await simple_metrics.run(input_data) + # Assert - load, export and adaptive state (no thresholds still updates it) + assert workflow_mock.execute_activity_method.call_count == 3 + workflow_mock.logger.warning.assert_not_called() + # Assert - Check load_custom_query call # Query format matches psycopg2.sql output (identifiers with double quotes, literals with single quotes) expected_query = f""" @@ -103,6 +108,8 @@ async def test_run(workflow_mock: AsyncMock, simple_metrics: SimpleMetrics): 'target_data': target_data, 'metrics': input_data['metrics'], 'interval_minutes': input_data['interval_minutes'], + 'model_type': None, + 'thresholds': None, }, retry_policy=ANY, start_to_close_timeout=ANY, @@ -128,13 +135,22 @@ async def test_run_empty_target_data(workflow_mock: AsyncMock, simple_metrics: S target_data = None + workflow_mock.logger = MagicMock() workflow_mock.execute_activity_method = AsyncMock(return_value=target_data) + workflow_mock.execute_local_activity_method = AsyncMock() # Act await simple_metrics.run(input_data) - # Assert - Should not call calculate_simple_metrics or export - assert workflow_mock.execute_activity_method.call_count == 1 + # Assert - No calculate or export; load + adaptive state (empty window = healthy run) + workflow_mock.execute_local_activity_method.assert_not_called() + calls = workflow_mock.execute_activity_method.call_args_list + assert [c.args[0] for c in calls] == [ + Activities.load_custom_query, + Activities.update_simple_metrics_adaptive_state, + ] + assert calls[1].args[1]['simple_metrics'] == [] + workflow_mock.logger.warning.assert_not_called() @mark.asyncio @@ -157,15 +173,22 @@ async def test_run_empty_simple_metrics(workflow_mock: AsyncMock, simple_metrics target_data = {'data': 'test_target_data'} simple_metrics_data = None + workflow_mock.logger = MagicMock() workflow_mock.execute_activity_method = AsyncMock(return_value=target_data) workflow_mock.execute_local_activity_method = AsyncMock(return_value=simple_metrics_data) # Act await simple_metrics.run(input_data) - # Assert - Should call calculate_simple_metrics but not export - workflow_mock.execute_activity_method.assert_called_once() + # Assert - calculate returns None: no export, adaptive state gets [] (the `or []` is live) workflow_mock.execute_local_activity_method.assert_called_once() + calls = workflow_mock.execute_activity_method.call_args_list + assert [c.args[0] for c in calls] == [ + Activities.load_custom_query, + Activities.update_simple_metrics_adaptive_state, + ] + assert calls[1].args[1]['simple_metrics'] == [] + workflow_mock.logger.warning.assert_not_called() @mark.asyncio @@ -188,12 +211,17 @@ async def test_run_default_metrics(workflow_mock: AsyncMock, simple_metrics: Sim target_data = {'data': 'test_target_data'} simple_metrics_data = {'metrics': 'test_simple_metrics_data'} - workflow_mock.execute_activity_method = AsyncMock(side_effect=[target_data, None]) + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[target_data, None, None]) workflow_mock.execute_local_activity_method = AsyncMock(return_value=simple_metrics_data) # Act await simple_metrics.run(input_data) + # Assert - all three activities ran + assert workflow_mock.execute_activity_method.call_count == 3 + workflow_mock.logger.warning.assert_not_called() + # Assert - Check calculate_simple_metrics call with default metrics workflow_mock.execute_activity_method.assert_any_call( Activities.load_custom_query, @@ -209,7 +237,182 @@ async def test_run_default_metrics(workflow_mock: AsyncMock, simple_metrics: Sim 'target_data': target_data, 'metrics': ['rmse', 'mse', 'mae', 'r2'], # Default value 'interval_minutes': input_data['interval_minutes'], + 'model_type': None, + 'thresholds': None, }, retry_policy=ANY, start_to_close_timeout=ANY, ) + + +def _input(**model_config) -> dict: + return { + 'schedule_name': 'test_schedule', + 'model_name': 'test_model', + 'model_id': 'test_model_id', + 'interval_minutes': 60, + 'model_config': {'target': 'test_target', **model_config}, + 'schema': 'test_schema', + 'predictions_table_name': 'test_predictions_table', + 'data_table_name': 'test_data_table', + 'target_table_name': 'test_target_table', + 'metrics': ['rmse'], + } + + +_ROW = {'timestamp': '2026-10-06 10:00:00+0000', 'prediction': 1.0, 'target': 2.0} + + +_METRICS = [{'metric': 'rmse', 'value': 9.0}] + + +@mark.asyncio +@mark.parametrize('n_rows', [2, 4]) +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_below_min_samples_drops_thresholds(workflow_mock, simple_metrics, n_rows): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[_ROW] * n_rows, None, None]) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_METRICS) + + await simple_metrics.run( + _input(simple_metrics_thresholds={'rmse_max': 1.0}, simple_metrics_min_samples=5) + ) + + activity_input = workflow_mock.execute_local_activity_method.call_args.args[1] + assert activity_input['thresholds'] is None + # Persist, don't alert: load, export, adaptive state. + assert workflow_mock.execute_activity_method.call_count == 3 + export_call = workflow_mock.execute_activity_method.call_args_list[1] + assert export_call.args[0] == Activities.export_data_to_postgres + assert export_call.args[1]['data'] == _METRICS + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_at_min_samples_keeps_thresholds(workflow_mock, simple_metrics): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[_ROW] * 5, None, None]) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_METRICS) + + await simple_metrics.run( + _input(simple_metrics_thresholds={'rmse_max': 1.0}, simple_metrics_min_samples=5) + ) + + activity_input = workflow_mock.execute_local_activity_method.call_args.args[1] + assert activity_input['thresholds'] == {'rmse_max': 1.0} + assert workflow_mock.execute_activity_method.call_count == 3 + + +_BREACH_ROWS = [{'metric': 'rmse', 'value': 9.0}] +_THRESHOLDS = {'rmse_max': 1.0} + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_updates_adaptive_state_after_export(workflow_mock, simple_metrics): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock( + side_effect=[[_ROW] * 3, None, {'escalated': True, 'healthy_streak': 0}] + ) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_BREACH_ROWS) + + await simple_metrics.run(_input(simple_metrics_thresholds=_THRESHOLDS)) + + calls = workflow_mock.execute_activity_method.call_args_list + assert calls[1].args[0] is Activities.export_data_to_postgres + assert calls[2].args[0] is Activities.update_simple_metrics_adaptive_state + assert calls[2].args[1] == { + **metadata, + 'simple_metrics': _BREACH_ROWS, + 'thresholds': _THRESHOLDS, + 'interval_minutes': 60, + } + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_without_thresholds_still_updates_adaptive_state(workflow_mock, simple_metrics): + # No thresholds = never a breach; still calling lets an escalated doc recover after the + # thresholds are removed. + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[_ROW] * 3, None, None]) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_BREACH_ROWS) + + await simple_metrics.run(_input()) + + calls = workflow_mock.execute_activity_method.call_args_list + assert len(calls) == 3 + assert calls[2].args[0] is Activities.update_simple_metrics_adaptive_state + assert calls[2].args[1]['thresholds'] is None + assert calls[2].args[1]['simple_metrics'] == _BREACH_ROWS + workflow_mock.logger.warning.assert_not_called() + + +@mark.asyncio +@mark.parametrize( + ('min_samples', 'warns'), [(None, False), (0, False), ('5', True), (5.0, True), (True, True)] +) +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_guard_off_keeps_thresholds(workflow_mock, simple_metrics, min_samples, warns): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[_ROW] * 2, None, None]) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_METRICS) + + await simple_metrics.run( + _input(simple_metrics_thresholds={'rmse_max': 1.0}, simple_metrics_min_samples=min_samples) + ) + + activity_input = workflow_mock.execute_local_activity_method.call_args.args[1] + assert activity_input['thresholds'] == {'rmse_max': 1.0} + assert workflow_mock.logger.warning.call_count == int(warns) + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_empty_window_counts_as_healthy_run(workflow_mock, simple_metrics): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[], None]) + workflow_mock.execute_local_activity_method = AsyncMock() + + await simple_metrics.run(_input(simple_metrics_thresholds=_THRESHOLDS)) + + workflow_mock.execute_local_activity_method.assert_not_called() + calls = workflow_mock.execute_activity_method.call_args_list + assert len(calls) == 2 # load + adaptive state, no export + assert calls[1].args[0] is Activities.update_simple_metrics_adaptive_state + assert calls[1].args[1]['simple_metrics'] == [] + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_below_min_samples_sends_no_rows_to_adaptive_state(workflow_mock, simple_metrics): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock(side_effect=[[_ROW] * 2, None, None]) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_BREACH_ROWS) + + await simple_metrics.run( + _input(simple_metrics_thresholds=_THRESHOLDS, simple_metrics_min_samples=5) + ) + + calls = workflow_mock.execute_activity_method.call_args_list + assert calls[1].args[0] is Activities.export_data_to_postgres # metrics still persisted + assert calls[2].args[0] is Activities.update_simple_metrics_adaptive_state + assert calls[2].args[1]['simple_metrics'] == [] # never a breach + + +@mark.asyncio +@patch('laborious.workflows.simple_metrics.workflow', new_callable=AsyncMock) +async def test_run_adaptive_state_failure_fails_workflow(workflow_mock, simple_metrics): + workflow_mock.logger = MagicMock() + workflow_mock.execute_activity_method = AsyncMock( + side_effect=[[_ROW] * 3, None, RuntimeError('dispatch failed')] + ) + workflow_mock.execute_local_activity_method = AsyncMock(return_value=_BREACH_ROWS) + input_data = _input(simple_metrics_thresholds=_THRESHOLDS) + + with raises(RuntimeError, match='dispatch failed'): + await simple_metrics.run(input_data) + + # Metrics were exported before the failing call + calls = workflow_mock.execute_activity_method.call_args_list + assert calls[1].args[0] is Activities.export_data_to_postgres