Houston, We Have a Problem. Designing Fault-Tolerant Systems
summary
The talk explains designing distributed systems to tolerate constant hardware and software failures by applying AWS-style "design for failure" principles. It covers why failures are normal at scale (statistical effects) and techniques to make known and unknown failures invisible to users and enable fast recovery to normal operation.
AI-written summary · about
chapters (12)
- 0:00System reliability introduction
- 4:38Control plane
- 6:07Mass outage in Virginia
- 9:18Isolation and prioritization
- 11:05Always-on pattern
- 14:37Pre-scaling pattern
- 16:20Cell architecture
- 22:42Multi-tenancy
- 25:33Sharded hashing
- 27:19Small fleet triggers large
- 29:49Load shedding
- 34:45Summary and Q&A
related talks
44mHighly Available, Distributed & Resilient Software... Team
46mWhy We All Build Bad Architectures and How to Stop
35mIt Scales Until It Doesn’t
59mMastering the System Design Methodology
23mDesigning Robust and Scalable Distributed Applications
49mJust in Time Architecture
56mDesigning Event-Driven Applications to Improve Scalability
25mHow to Design a Great System When the (Functional) Path Isn’t Defined Yet