Don't Be Afraid of the Dark
summary
Argues that risk drops when teams run real production-feeling tests early: use feature flags and tiny HTTP timeouts, enqueue async jobs (Resque) or send messages, write to throwaway stores like Redis, and run long-running tests — one reservation probe runs for over a week and shows low error rates, a signup-recommendation test survives a quadrupled traffic spike but reveals a fallback bug, Kafka broker rotations expose config fragility, and GitHub’s scientist pattern helps validate a migration.
AI-written summary · about
chapters (17)
- 0:00Introduction and setup
- 1:20Speaker introduction
- 3:06Fear in complex systems
- 5:02Risk and the quadrant of fear
- 7:06Mapping project icebergs
- 9:10Testing at production scale
- 11:31Prerequisites for safe testing
- 13:19Asking hypothetical questions
- 15:51Isolating changes and data
- 18:34Ship test learn iterate
- 19:09Real-world example inventory
- 20:37Real-world example recommendations
- 22:57Learning infrastructure quirks
- 24:00Technical migrations as exoskeletons
- 26:25Migrating legacy systems
- 28:47Game days and incident prep
- 30:07Choosing how to illuminate darkness
related talks
22mWhat Could Go Wrong? The Subtle Science Of Coding for Failure
20mCalcifying Crisis Readiness
40mCatching More Bugs with Fewer False Alarms
47mSecurity Pre-Cognition
23mSecure by Design: Rethinking Test Credentials for Synthetic Monitoring
39mAutomating myself out of a job...
23mHow I Learned To Be Paranoid and Loved It- 40mCatching More Bugs with Fewer False Alarms