Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

fix(alerts): apply the default when a detector config field is null (#125011)

posthog[bot]1f3b52e72c94
diff --git a/posthog/tasks/alerts/detectors/base.py b/posthog/tasks/alerts/detectors/base.py
index 26cd5c44801d..7358b37d9635 100644
--- a/posthog/tasks/alerts/detectors/base.py
+++ b/posthog/tasks/alerts/detectors/base.py
@@ -54,6 +54,9 @@ class BaseDetector(ABC):
# Default anomaly probability threshold. Higher = fewer alerts.
DEFAULT_THRESHOLD = 0.95
+ # Default rolling window size for the statistical detectors.
+ DEFAULT_WINDOW = 30
+
# Default number of recent points to exclude from training data.
# Prevents the model from fitting on the points it's about to score.
# Higher values make the model slower to adapt to recent distribution shifts.
@@ -63,6 +66,11 @@ class BaseDetector(ABC):
self.config = config
self.preprocessing_config = config.get("preprocessing") or {}
self.training_offset: int = config.get("training_offset_n", self.DEFAULT_TRAINING_OFFSET)
+ # A schema-normalized config keeps an unset field as an explicit None, which `dict.get` does not replace.
+ threshold = config.get("threshold")
+ self.threshold: float = self.DEFAULT_THRESHOLD if threshold is None else threshold
+ window = config.get("window")
+ self.window: int = self.DEFAULT_WINDOW if window is None else window
@abstractmethod
def detect(self, data: np.ndarray) -> DetectionResult:
@@ -156,8 +164,7 @@ class BaseDetector(ABC):
return rows
def _scored_result(self, scores: list[float | None], metadata: list[dict[str, Any]]) -> DetectionResult:
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- triggered = [i for i, score in enumerate(scores) if score is not None and score > threshold]
+ triggered = [i for i, score in enumerate(scores) if score is not None and score > self.threshold]
return DetectionResult(
is_anomaly=bool(triggered) and triggered[-1] == len(scores) - 1,
score=scores[-1] if scores else None,
diff --git a/posthog/tasks/alerts/detectors/pyod_detectors/base_pyod.py b/posthog/tasks/alerts/detectors/pyod_detectors/base_pyod.py
index 2a276473dbde..bf87cce35f26 100644
--- a/posthog/tasks/alerts/detectors/pyod_detectors/base_pyod.py
+++ b/posthog/tasks/alerts/detectors/pyod_detectors/base_pyod.py
@@ -36,7 +36,7 @@ class BasePyODDetector(BaseDetector):
return DetectionResult(is_anomaly=False)
data = self.preprocess(data)
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
+ threshold = self.threshold
if data.ndim == 1:
data = data.reshape(-1, 1)
@@ -114,7 +114,7 @@ class BasePyODDetector(BaseDetector):
return DetectionResult(is_anomaly=False)
data = self.preprocess(data)
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
+ threshold = self.threshold
if data.ndim == 1:
data = data.reshape(-1, 1)
diff --git a/posthog/tasks/alerts/detectors/statistical/iqr.py b/posthog/tasks/alerts/detectors/statistical/iqr.py
index a1fab3993915..5ec1a4442c21 100644
--- a/posthog/tasks/alerts/detectors/statistical/iqr.py
+++ b/posthog/tasks/alerts/detectors/statistical/iqr.py
@@ -55,9 +55,9 @@ class IQRDetector(BaseDetector):
"""
def detect(self, data: np.ndarray) -> DetectionResult:
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- multiplier = self.config.get("multiplier", 1.5)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ multiplier = self.config.get("multiplier") or 1.5
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -119,9 +119,9 @@ class IQRDetector(BaseDetector):
)
def detect_batch(self, data: np.ndarray) -> DetectionResult:
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- multiplier = self.config.get("multiplier", 1.5)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ multiplier = self.config.get("multiplier") or 1.5
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -179,8 +179,8 @@ class IQRDetector(BaseDetector):
)
def fit(self, history: np.ndarray) -> DetectorArtifact:
- multiplier = self.config.get("multiplier", 1.5)
- window = self.config.get("window", 30)
+ multiplier = self.config.get("multiplier") or 1.5
+ window = self.window
diffs_n = 1 if self.preprocessing_config.get("diffs_n") else 0
rows = self._fit_rows(history, min_rows=window + diffs_n)
values = rows if rows.ndim == 1 else rows[:, 0]
diff --git a/posthog/tasks/alerts/detectors/statistical/mad.py b/posthog/tasks/alerts/detectors/statistical/mad.py
index b7e77300cf61..b21a07c21431 100644
--- a/posthog/tasks/alerts/detectors/statistical/mad.py
+++ b/posthog/tasks/alerts/detectors/statistical/mad.py
@@ -29,8 +29,8 @@ class MADDetector(BaseDetector):
"""
def detect(self, data: np.ndarray) -> DetectionResult:
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -75,8 +75,8 @@ class MADDetector(BaseDetector):
)
def detect_batch(self, data: np.ndarray) -> DetectionResult:
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -118,7 +118,7 @@ class MADDetector(BaseDetector):
)
def fit(self, history: np.ndarray) -> DetectorArtifact:
- window = self.config.get("window", 30)
+ window = self.window
diffs_n = 1 if self.preprocessing_config.get("diffs_n") else 0
rows = self._fit_rows(history, min_rows=window + diffs_n)
values = rows if rows.ndim == 1 else rows[:, 0]
diff --git a/posthog/tasks/alerts/detectors/statistical/zscore.py b/posthog/tasks/alerts/detectors/statistical/zscore.py
index 31dd1014cec9..4b6738e84efc 100644
--- a/posthog/tasks/alerts/detectors/statistical/zscore.py
+++ b/posthog/tasks/alerts/detectors/statistical/zscore.py
@@ -45,8 +45,8 @@ class ZScoreDetector(BaseDetector):
def detect(self, data: np.ndarray) -> DetectionResult:
"""Check if the most recent point is an anomaly based on z-score."""
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -102,8 +102,8 @@ class ZScoreDetector(BaseDetector):
def detect_batch(self, data: np.ndarray) -> DetectionResult:
"""Check all points for z-score anomalies."""
- threshold = self.config.get("threshold", self.DEFAULT_THRESHOLD)
- window = self.config.get("window", 30)
+ threshold = self.threshold
+ window = self.window
# preprocess() only ever runs a single first-difference pass when diffs_n is truthy
# (it's a boolean toggle, not a pass count), so exactly one synthetic leading point
# is introduced regardless of the configured magnitude.
@@ -154,7 +154,7 @@ class ZScoreDetector(BaseDetector):
)
def fit(self, history: np.ndarray) -> DetectorArtifact:
- window = self.config.get("window", 30)
+ window = self.window
diffs_n = 1 if self.preprocessing_config.get("diffs_n") else 0
rows = self._fit_rows(history, min_rows=window + diffs_n)
values = rows if rows.ndim == 1 else rows[:, 0]
diff --git a/posthog/tasks/alerts/detectors/test/test_detectors.py b/posthog/tasks/alerts/detectors/test/test_detectors.py
index 5e090c20abf1..c569a4747211 100644
--- a/posthog/tasks/alerts/detectors/test/test_detectors.py
+++ b/posthog/tasks/alerts/detectors/test/test_detectors.py
@@ -5,6 +5,8 @@ import pytest
import numpy as np
from parameterized import parameterized
+from posthog.schema import DetectorConfig
+
from posthog.tasks.alerts.detector import _compute_min_samples_for_detector
from posthog.tasks.alerts.detectors.base import DetectionResult
from posthog.tasks.alerts.detectors.ensemble import EnsembleDetector
@@ -250,20 +252,31 @@ class TestStatisticalDetectors:
@parameterized.expand(
[
- ("zscore", ZScoreDetector),
- ("iqr", IQRDetector),
- ("mad", MADDetector),
+ (f"{name}_{fields}", detector_type, detector_cls, explicit)
+ for name, detector_type, detector_cls in (
+ ("zscore", "zscore", ZScoreDetector),
+ ("iqr", "iqr", IQRDetector),
+ ("mad", "mad", MADDetector),
+ )
+ for fields, explicit in (
+ ("preprocessing_none", {"window": 5, "preprocessing": None}),
+ ("schema_defaults", {}),
+ )
]
)
- def test_handles_explicit_none_preprocessing(self, _name: str, detector_cls: Any) -> None:
- # Serializers round-trip a detector_config with an explicit "preprocessing": None
- # key (not merely omitted) when no preprocessing was configured. `config.get(key,
- # {})` only falls back to `{}` when the key is absent, so a present-but-None value
- # used to reach detect()/detect_batch() as None and crash on `.get("diffs_n", ...)`.
- detector = detector_cls({"threshold": 0.9, "window": 5, "preprocessing": None})
- data = np.arange(20, dtype=float)
- detector.detect(data)
- detector.detect_batch(data)
+ def test_handles_explicit_none_config_values(
+ self, _name: str, detector_type: str, detector_cls: Any, explicit: dict[str, Any]
+ ) -> None:
+ # Serializers round-trip a detector_config through the schema's model_dump(), which
+ # writes every unset field as an explicit None (not merely omitted). `config.get(key,
+ # default)` only falls back when the key is absent, so a present-but-None value used
+ # to reach detect()/detect_batch()/fit() and crash on arithmetic or comparison.
+ config = DetectorConfig.model_validate({"type": detector_type, **explicit}).model_dump()
+ detector = detector_cls(config)
+ data = np.array([10.0, 12.0, 11.0, 13.0] * 15 + [500.0])
+ assert detector.detect(data).score is not None
+ assert detector.detect_batch(data).all_scores[-1] is not None
+ detector.fit(data[:-1])
@parameterized.expand(
[