It passed auth, then production caught fire.
summary
A concrete failure mode: distributed services implement authorization differently so six services return divergent decisions for the same user and action, one tiny permission change triggers retries and queue depth hits 30,000, and rolling back that change restores latency. Fix is a central policy decision point (Serbos or OPA), policies stored in Git, CI tests and replayable audit logs, sidecar deployment for local speed, and deterministic allow/deny answers instead of scattered hard‑coded checks.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:022 a.m. production fire drill
- 3:31Auth vs authorization
- 5:16Distributed auth failures
- 8:05AI agents auth risks
- 10:13Authorization as reliability
- 13:02Externalize auth with standards
- 15:495 benefits of centralized auth
- 17:35Debugging auth failures
- 18:383 critical system questions
- 19:20Implementing externalized auth
related talks
32mAuthz as a Dev Workflow: Architecting Better Cloud Native Apps
29mAuthz as a dev workflow
29mWhy Authorization is Hard
41mBuilding Modern Authorization: A Practical Guide To Centralized, Distributed, and...
36mAccess control in message-driven systems
45mThe Future of Authorization in Applications
51mPermissions: Centralized or Decentralized? Both!
27mSecurity-by-construction: How to weave authorization into modern app stacks using Open Policy Agent