Sailing the Database Seas: Applying SRE Principles at Scale
summary
Booking.com runs MySQL across bare metal, private and public clouds with ~hundreds of clusters and >10k servers managed by ~20 SRE DBAs, and measures read availability 99.99, write availability 99.97 and replication running 99.95 using an in‑house mySQL availability reporter (Go) in Kubernetes plus a badmin agent (Python) per server, Prometheus and Graphite. Daily production capacity tests route real traffic to replicas with a kill switch, a provisioning manager that talks to AWS/OpenStack/bare metal, and conservative scaling; SLOs mostly hold, user‑induced queries cause the largest breaches and RFOS plus seven‑day followups drive fixes.
AI-written summary · about
chapters (6)
- 0:00Database reliability at scale
- 2:42Database reliability engineering
- 5:32Choosing SLIs for SLOs
- 13:36Automating MySQL capacity planning
- 28:09Embracing failure with postmortems
- 33:09Postmortem culture
related talks
23mRunning MySQL on K8s. Deployment, Scaling, Backups and advanced configuration
41mDemocratizing MySQL: From Cloud Managed to Kubernetes Managed
54mBreakout Session
36mCreate your Own MySQL-as-a-Service that Runs Anywhere Using Kubernetes Operators
50mTransforming Your DBA Team Into a DBRE/SRE One!
26mOperationalizing 19 years of infrastructure via MySQL-as-a-Service
51mMySQL Backups
53mMySQL DevOps