gitea_admin/sientia-dataops-laborious-legacy
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(simple_metrics): extract threshold_breaches helper (SIENTIAPDE-2191)

Browse Source
This commit is contained in:
PedroHMCosme
2026-10-07 15:33:15 -03:00
parent 9f8237a46e
commit f503fd0610
2 changed files with 60 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
laborious/activities/model_metrics.py
View File

@@ -26,6 +26,28 @@ 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: `<metric>_max` breaches when value > threshold, `<metric>_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
class ModelMetrics(SientiaMonitoring):
"""
Metrics and statistical analysis activities for the Laborious pipeline.
@@ -430,37 +452,21 @@ class ModelMetrics(SientiaMonitoring):
self._debug_dataframe(f'Simple metrics dataframe: Size {data.shape}', data, metadata)
# Threshold alerting (optional — no crash when absent)
thresholds = input_data.get('thresholds')
if thresholds:
# Convention: _max thresholds breach when value > threshold,
# _min thresholds breach when value < threshold.
for row in output_data:
metric_name = row['metric']
value = row['value']
max_key = f'{metric_name}_max'
min_key = f'{metric_name}_min'
breach_msg = None
if max_key in thresholds and value > thresholds[max_key]:
breach_msg = f'{metric_name}={value} exceeds {max_key}={thresholds[max_key]}'
elif min_key in thresholds and value < thresholds[min_key]:
breach_msg = f'{metric_name}={value} below {min_key}={thresholds[min_key]}'
if breach_msg:
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
# `<trigger>:<notification_id>`, 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,
)
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
# `<trigger>:<notification_id>`, 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')