How to Serve and Protect (with Client Isolation)
summary
Maps SREs implement client isolation by adding layered defenses: coarse global protection at the edge, per-client quotas (one component was oversubscribed ~7x), front-end tagging and prioritization, early caching, selective load shedding, and splitting hot paths into microservices. They find exempting big clients breaks isolation, priority inversion and fan-out failures bite, and backend drops waste work. The result is predictable graceful degradation: high-priority UX requests stay live while low-value and batch traffic gets shed.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:47Google Maps
- 1:49What happened
- 3:20Client Isolation
- 4:04Load shedding
- 5:28A simplified system
- 6:37Strategies
- 7:02Caching
- 8:52QPS
- 10:31Prioritisation
- 12:45Priority inversion
- 13:34Microservices
- 15:29Exemptions
- 16:32Client Isolation maxim
- 17:02Protection maxim
- 19:04Backend optimization
- 22:44Benefits
related talks
14mWeb Apps, through the Lens of SRE (Site Reliability Engineering)
18mScaling APIs to Billions: Extreme-Scale Reliability
41mSeven patterns to increase application availability
27mLayered Security
53mThe server that talked back: a deep dive into SSRFs
29mSingle Page Apps: What an API Developer Should Know
32mBeyond Optimization: Resilient Cloud Microservices at Scale
30mAchieving the Web Isolation Nirvana - How far along are we?