chore(anomaly-detection): Return meaningful errors for client side er…

…rors

src/seer/anomaly_detection/anomaly_detection.py

@@ -23,6 +23,7 @@
  TimeSeriesWithHistory,
 )
 from seer.dependency_injection import inject, injected
+from seer.exceptions import ClientError, ServerError
 
 anomaly_detection_module.enable()
 logger = logging.getLogger(__name__)
@@ -107,10 +108,23 @@ def _online_detect(
  # Retrieve historic data
  historic = alert_data_accessor.query(alert.id)
  if historic is None:
- raise Exception(f"Invalid alert id {alert.id}")
+
+ logger.error(
+ "no_stored_history_data",
+ extra={
+ "alert_id": alert.id,
+ },
+ )
+ raise ClientError("No timeseries data found for alert")
 
  if not isinstance(historic.anomalies, MPTimeSeriesAnomalies):
- raise Exception("Invalid state")
+ logger.error(
+ "invalid_state",
+ extra={
+ "note": f"Expecting object of type MPTimeSeriesAnomalies but found {type(historic.anomalies)}"
+ },
+ )
+ raise ServerError("Invalid state")
  anomalies: MPTimeSeriesAnomalies = historic.anomalies
 
  # TODO: Need to check the time gap between historic data and the new datapoint against the alert configuration
@@ -168,7 +182,7 @@ def _combo_detect(
  "minimum_required": min_len,
  },
  )
- raise Exception("Insufficient history data")
+ raise ClientError("Insufficient history data")
 
  logger.info(
  f"Detecting anomalies for time series with {len(ts_with_history.current)} datapoints and history of {len(ts_with_history.history)} datapoints"
@@ -194,8 +208,6 @@ def _combo_detect(
  return ts_external, streamed_anomalies
 
  def _update_anomalies(self, ts_external: List[TimeSeriesPoint], anomalies: TimeSeriesAnomalies):
- if anomalies is None:
- raise Exception("No anomalies available for the timeseries.")
  for i, point in enumerate(ts_external):
  point.anomaly = Anomaly(
  anomaly_score=anomalies.scores[i],
@@ -227,7 +239,7 @@ def detect_anomalies(self, request: DetectAnomaliesRequest) -> DetectAnomaliesRe
  else:
  ts, anomalies = self._batch_detect(request.context, request.config)
  self._update_anomalies(ts, anomalies)
- return DetectAnomaliesResponse(timeseries=ts)
+ return DetectAnomaliesResponse(success=True, timeseries=ts)
 
  @inject
  def store_data(
@@ -253,9 +265,7 @@ def store_data(
  "minimum_required": min_len,
  },
  )
- return StoreDataResponse(
- success=False, message=f"Insufficient time series data for alert {request.alert.id}"
- )
+ raise ClientError("Insufficient time series data for alert")
 
  logger.info(
  "store_alert_request",
@@ -266,27 +276,14 @@ def store_data(
  "num_datapoints": len(request.timeseries),
  },
  )
- try:
- ts, anomalies = self._batch_detect(request.timeseries, request.config)
- alert_data_accessor.save_alert(
- organization_id=request.organization_id,
- project_id=request.project_id,
- external_alert_id=request.alert.id,
- config=request.config,
- timeseries=ts,
- anomalies=anomalies,
- anomaly_algo_data={"window_size": anomalies.window_size},
- )
- return StoreDataResponse(success=True)
- except Exception as e:
- sentry_sdk.capture_exception(e)
- logger.error(
- "error_saving_alert",
- extra={
- "organization_id": request.organization_id,
- "project_id": request.project_id,
- "external_alert_id": request.alert.id,
- },
- exc_info=e,
- )
- return StoreDataResponse(success=False, message=str(e))
+ ts, anomalies = self._batch_detect(request.timeseries, request.config)
+ alert_data_accessor.save_alert(
+ organization_id=request.organization_id,
+ project_id=request.project_id,
+ external_alert_id=request.alert.id,
+ config=request.config,
+ timeseries=ts,
+ anomalies=anomalies,
+ anomaly_algo_data={"window_size": anomalies.window_size},
+ )
+ return StoreDataResponse(success=True)

src/seer/anomaly_detection/detectors/anomaly_detectors.py

@@ -18,6 +18,7 @@
  TimeSeriesAnomalies,
 )
 from seer.dependency_injection import inject, injected
+from seer.exceptions import ServerError
 
 logger = logging.getLogger(__name__)
 
@@ -85,7 +86,7 @@ def _compute_matrix_profile(
  logger.debug(f"window_size: {window_size}")
  if window_size <= 0:
  # TODO: Add sentry logging of this error
- raise Exception("Invalid window size")
+ raise ServerError("Invalid window size")
  # Get the matrix profile for the time series
  mp = stumpy.stump(
  ts_values,

src/seer/anomaly_detection/detectors/mp_utils.py

@@ -5,6 +5,7 @@
 from seer.anomaly_detection.detectors.mp_config import MPConfig
 from seer.anomaly_detection.detectors.normalizers import Normalizer
 from seer.dependency_injection import inject, injected
+from seer.exceptions import ServerError
 
 
 class MPUtils(BaseModel):
@@ -38,12 +39,12 @@ def get_mp_dist_from_mp(
  mp_dist = mp[:, 0]
  if mp_config is not None and mp_config.normalize_mp:
  if normalizer is None:
- raise Exception("Need normalizer to normalize MP")
+ raise ServerError("Need normalizer to normalize MP")
  mp_dist = normalizer.normalize(mp_dist)
 
  if pad_to_len is not None:
  if pad_to_len < len(mp_dist):
- raise Exception(
+ raise ServerError(
  "Requested length should be greater than or equal to current mp_dist"
  )
  nan_value_count = np.empty(pad_to_len - len(mp_dist))

src/seer/anomaly_detection/models/external.py

@@ -72,7 +72,9 @@ class DetectAnomaliesRequest(BaseModel):
 
 
 class DetectAnomaliesResponse(BaseModel):
- timeseries: List[TimeSeriesPoint]
+ success: bool
+ message: Optional[str] = Field(None)
+ timeseries: Optional[List[TimeSeriesPoint]] = Field(None)
 
 
 class StoreDataRequest(BaseModel):

src/seer/app.py

@@ -52,6 +52,7 @@
 from seer.bootup import bootup, module
 from seer.configuration import AppConfig
 from seer.dependency_injection import inject, injected, resolve
+from seer.exceptions import ClientError
 from seer.grouping.grouping import (
  BulkCreateGroupingRecordsResponse,
  CreateGroupingRecordsRequest,
@@ -252,15 +253,23 @@ def summarize_issue_endpoint(data: SummarizeIssueRequest) -> SummarizeIssueRespo
 def detect_anomalies_endpoint(data: DetectAnomaliesRequest) -> DetectAnomaliesResponse:
  sentry_sdk.set_tag("organization_id", data.organization_id)
  sentry_sdk.set_tag("project_id", data.project_id)
- return anomaly_detection().detect_anomalies(data)
+ try:
+ response = anomaly_detection().detect_anomalies(data)
+ except ClientError as e:
+ response = DetectAnomaliesResponse(success=False, message=str(e))
+ return response
 
 
 @json_api(blueprint, "/v1/anomaly-detection/store")
 @sentry_sdk.trace
 def store_data_endpoint(data: StoreDataRequest) -> StoreDataResponse:
  sentry_sdk.set_tag("organization_id", data.organization_id)
  sentry_sdk.set_tag("project_id", data.project_id)
- return anomaly_detection().store_data(data)
+ try:
+ response = anomaly_detection().store_data(data)
+ except ClientError as e:
+ response = StoreDataResponse(success=False, message=str(e))
+ return response
 
 
 @blueprint.route("/health/live", methods=["GET"])