Escalating complexity: DevOps learnings from Air France 447
summary
Air France 447 crashes with 228 lost after a chaotic 15 minutes; investigators spend three years recovering the flight data. The cockpit sees conflicting pitot (pedo) tube readings, an autopilot disconnect and 70 stall warnings while the flight computer shifts into alternate law reconfiguration and ACARS spits small textual alerts rather than actionable cues. Key takeaways: cockpit feedback is sensory design, paired responders and damped alerts reduce alert fatigue, and sometimes not touching controls is the counterintuitive correct action.
AI-written summary · about
chapters (32)
- 0:00Intro
- 1:013 hours and 45 minutes later, it crashed into the Atlantic Ocean, killing 216 passengers, and 12 aircrew.
- 7:54human error
- 10:29hindsight != foresight
- 10:42We have all the facts
- 11:46Fog of War
- 11:59Local rationality
- 12:29Modes of operation
- 15:06Realisation
- 15:59Sensory feedback
- 17:18Optimise for zam you
- 17:32Input control
- 18:06Averaged input
- 20:40Do you practice this?
- 21:06Vocalise
- 21:24Record
- 21:45isolated sensors
- 21:55different values
- 22:10discrepancies
- 23:09dashboards
- 23:32linkable
- 23:56correlation
- 24:3770 stall warnings
- 24:53autopilot disconnect
- 25:12alert priority level?
- 25:23Alert Fatigue
- 25:39dampening
- 26:32Flapjack
- 27:35Failure is pervasive
- 27:58Anthropocentrism
- 28:08Technocentrism
- 28:30No amoral actors
related talks
01h 00mLearning from Disaster
25mMayday! Software Lessons From an Aviation Disaster
01h 01mFly the Enterprise: Applying Aviation Lessons to DevOps Teams
37mAir Crashes
39mWhat a full production crash taught us
36mIncident insights from NASA, NTSB, and the CDC
29mLearning from incidents: how things went right
23mNavigating in the Dark