Database Parameter Auto-Tuning Through CI/CD Pipelines
summary
A CI/CD pipeline called cicd-tune integrates Screwdriver CI with two ML models (a feature-extraction + TPS estimator pair) to infer MySQL parameter settings from dev-environment metrics, run sysbench-like workloads during build, and show pass/fail against SLA so developers decide deployment. Demo shows autotuned settings can both cut latency (marked “good to deploy”) and badly regress it — an observed 10x latency increase — so the system gates risky changes and flags where dev-to-prod translation needs further modeling.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 3:50Ly Corporation’s data infrastructure
- 7:03DB parameter tuning challenges
- 12:56ML tuning with CI/CD pipelines
- 15:28Live demo of CICD Tune
- 18:32Future auto-tuning research
- 21:27Acknowledgments & Q&A
related talks
17mWorkload-Based MySQL Tuning
18mWhy Postgres Is Better Than MySQL for Automated Knob Tuning
20mDebate: Autotuning
16mWhether tuning MySQL configuration important?
27mBuilding safety tooling for risk-free AI tuning of Postgres
53mMySQL DevOps
26mIntelligent Deployment Pipelines
41mMySQL Performance for DevOps