Capacity and Stability Patterns
summary
Eventbrite handles sudden on‑sales that can quadruple one endpoint’s traffic in a minute and serves 150M tickets last year, so engineering focuses on concrete measures: partitioning with bulkheads and dedicated pools, gradual rollouts behind feature flags (gargoyle fork), a waiting room queue to shed load, bounded Redis queues in a Django‑channels service framework, Varnish + ESI outer/inner caching with 5s TTL and a MySQL‑binlog driven invalidation service, plus iptables/nginx rate limits and strict timeouts.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:56Systems
- 3:40Stability
- 10:24Canary testing
- 12:45Feature flags
- 14:36Graceful degradation
- 18:12Load shedding
- 23:39Timeouts
- 26:17Caching
- 27:01Invalidation strategies
- 33:34Capacity Planning
related talks
21mThe Multi Layered Cake of Resilience
32mHow does your application behaves when everything goes wrong?
34mDistributed Systems - What Can Go Wrong and How to Get It Right
37mReactive Architecture on AWS: from 0 to 1.6M events/minute
47mApplying “web scale” patterns (...)
40mChaos Engineering at Scale
59mBuilding resilient event-driven architectures
20mBetter Resilience Through Self Tuning