Selective Reliability Engineering: There Is No Single Source...
summary
Builds an image-upload service and compares two concrete architectures: a single-server with local filesystem versus a stateless service + cloud DB/blob storage, sets a 0.5s latency target, and adds a subscription and physical-print flow that forces a payments service decision. Argues there is no single source of truth, shows practical failures (403 vs 404 info leak, delayed syncs, multi-provider payment complexity), cites a real safety cascade from aggregated local optimizations, and recommends adding SLA, observability, runbooks and empathy into design requirements.
AI-written summary · about
chapters (8)
- 0:00Selective reliability intro
- 1:42Trade-offs in system design
- 3:05Role of selective reliability
- 4:50Context and scope
- 7:20Philosophy of truth in systems
- 9:47Truth versus security
- 11:53Human impact of design
- 13:16Selecting for empathy
related talks
28mMaking the Impossible Impossible: Improving Reliability by...
37mHiding The Lead
27m“Good Enough” Architecture Part 2
44mThe Art of Architecting for Scale
01h 29mMonolithic Delivery
46mRoad-movie architectures
47mLet's Learn to Identify Technical Requirements for Better Design
33mFrom Enterprise Architecture to Software Design - Critical Thinking