Against On-Call: A Polemic
summary
Argument: humans perform poorly under on-call stress—baseline error rate ~0.5% and errors double under time pressure (Microsoft chess study), 26% vs 19% report work–family conflict, sleep loss lowers life expectancy—so on-call is harmful. Root causes are preventable: 70% of outages trace to config or binary pushes and ~62% involve dev-process failures. Solutions stress automation and design: canary deployments, durable resilient composable components (e.g., Cassandra/Kafka patterns), safe-failure partitions and industry-wide cloud primitives that remove emergency on-call.
AI-written summary · about
chapters (7)
- 0:00Start
- 5:45Emergency Medicine
- 20:17Classification of Risks
- 22:43Known Unknowns and Unknown Knowns
- 29:59Counter-Arguments
- 32:07Machine Learning May Be an Effective Substitute for Human on Call
- 42:55How Does the Unreliability Arise
related talks
28mBuilding a healthy on-call culture
06mIgnite: Key Factors for Successful 24×7 On-Call Duty
05mOn-Call in a World Full of Toil
31mHow to Build a Healthy On-call Culture
19mIts Time We Rethink On-Call Culture
30mOncall: An Equal Opportunity Waste of Time
44mYou build it, you run it
57mCognitive Bias and On-Call