Survivorship Fairy Tales or When 1% Matters
summary
Concrete failure cases and fixes across systems: an Ethereum smart‑contract hybrid exchange hit a 15 tx/s limit and order‑book desynchronization that caused bot trades to waste gas until an added auto‑recovery and block‑validation layer synced blocks; a PHP website lost database indexes causing latency to jump from ~50ms to ~200ms and page loads to spike to ~10s until indexes and disk IO were restored; heavy graph queries (20–30 joins) slowed from ~150ms to 8–10s without precomputed caches; and 1% corruption in analytics data ruined historical trust, demonstrating the need for backups and validation.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:11Cloud loan startup mistakes
- 5:18Highload blockchain exchange
- 12:10News project index optimization
- 14:41Mobile games metrics
- 17:52Data analytics sacred cow
- 20:02AI graph query performance
- 23:16Business elasticity
- 25:00Q&A
related talks
49mBeware of Survivorship Bias!
48mFault Tolerance on the Cheap Making Systems That Probably Won't Fall Over
20mHow Misjudging the Importance of an Unloved Project can Turn Into a Dis
29mApps, Stacks, and Frameworks: Avoiding “Shiny Object” Syndrome
30mLearning from Learnings: Anatomy of Three Incidents
28mThe Failing CTO - What we can learn from others' mistakes
25mDealing with data in dynamic systems
36mHigh Availability at Braintree