Automatic Metric Screening for Service Diagnosis
summary
An automated pipeline detects sudden metric changes via kernel density estimation (including Gaussian, Poisson and Beta kernels), converts per-instance anomaly scores, clusters abnormal instances into digests and ranks them by proportion, number of abnormal metrics and anomaly degree. The system reduces diagnosis time by 80%, runs in under 6 minutes, and on 70 historical incidents yields over 80% top‑1 precision; uploader notes top 3 groups cover most root cause modules.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:20Service Diagnosis
- 1:40Diagnosis Process
- 3:29Response Time
- 5:03Limitations
- 6:18Motivation
- 7:07Procedure
- 14:12Anomaly Score
- 15:30Ranking
- 16:55Results
related talks
35mObservability’s Sixth Sense: Detecting Anomalies in Metrics
35mObservability’s Sixth Sense: Detecting Anomalies in Metrics
47mDid My Deploy Degrade Production?
30mBetter data quality with ML & visualization
26mML Monitoring & Observability
38mWatch Everything, Watch Anything: Anomaly Detection
51mAnomaly Detection as a Service - What is normal?
26mProactive Failure Prediction Through Anomaly Detection