Tune Your Microservices by Learning from Traces
summary
Architecture ingests Istio traces (≈700,000/day) from Elasticsearch, prepares fixed windows and trains LSTM/seq2seq time-series models via Kubeflow on Kubernetes, stores models in object storage and runs detection at service granularity. Short-window models flag abrupt duration spikes; long-window models suppress learned seasonality to avoid false alarms during autoscaling. Test sets use ~1,990 points per experiment. System deploys per-service models, not a single global model, and notifies ops on identified anomalies.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:06Kubeflow
- 1:34Architecture Overview
- 5:20Training & Modeling
- 15:09Model Serving
- 17:01Tuning microservices based on result
related talks
29mTune Your Microservices by Learning from Traces
14mPerformance Optimization in High-Volume Transaction Systems
30mDebugging at Scale Using Elastic and Machine Learning
14mLeveraging Distributed Tracing in Kubernetes & Containerized Environments
49mMonitor Your Microservices with Logs, Metrics, Pings and Traces
42mDistributed Monitoring: How to understand the Chaos
32mInvestigating Performance Issues In Microservices Arch.
21mNavigating the Complexities of Service to Service Invocations: Deep and Brief Dive Into Causality