Anomaly Detection in Time Series from Scratch Using...
summary
Claims that simple statistics (not ML) catch slow burns and spikes using Z‑score and MAD; implements it against Graphite, visualizes and alerts in Grafana, and builds Grafana dashboards as code. A small service called gomal service reads time‑shifted weekly windows, trims outliers with a percentile + “cherry‑picking” median-of-Z normalization, handles daylight‑savings alignment, and applies an origin adjustment multiplier. It scales to many business metrics, mostly works, and reduced false baselines caused by past outages.
AI-written summary · about
chapters (10)
- 0:00Introduction and context
- 1:09Limitations of simple monitoring
- 3:52Statistical anomaly detection
- 7:25Graphite and deviation challenges
- 11:11Building the Granomaly service
- 14:34Granomaly algorithm design
- 18:21Refining outlier exclusion
- 22:59Addressing operational hurdles
- 28:34Managing dashboards as code
- 31:11Scaling and final takeaways
related talks
40mObservability’s Sixth Sense: Detecting Anomalies in Metrics
35mObservability’s Sixth Sense: Detecting Anomalies in Metrics
38mWatch Everything, Watch Anything: Anomaly Detection
16mDetecting Anomalies in Logs with Time Series Analysis
06mEasy anomaly Detection with PromQL
35mObservability’s Sixth Sense: Detecting Anomalies in Metrics
30mPractical Anomaly Detection at Scale With PromQL
04mLightning Talk: Anomaly Detection - Robust Strategy