ReBAC at Scale: Let’s Save the p99
summary
A ReBAC (relationship-based access control) deployment at Agicap hit latency and cost issues after two years of centralized use as product scale increased. The talk explains which parts break at scale (relation explosion, hot paths, cache degradation, zero-downtime migrations), which metrics to track, model refactors that yield the biggest improvements, and technical levers like caching and read replicas to stabilize p99 latency. Attendees get modelling rules, anti-patterns, and a checklist to keep a centralized authorization platform fast, resilient, and evolvable.
AI-written summary · about
chapters (10)
- 0:00ReBAC introduction
- 6:38P99 incident detection
- 13:36Diagnosing slowdown causes
- 20:20Flattening complex relations
- 24:13Write adonents implementation
- 28:27Load testing validation
- 34:03Solution tradeoffs
- 37:14Preventing future regressions
- 38:17Lessons and best practices
- 41:06Q&A
related talks
46mRebuilding Access Management with ReBAC
47mFine Grained Authorisation with Relationship Based Access Control
41mFine Grained Authorisation with Relationship-Based Access Control
46mInfuse Business Logic into Permissions with ReBAC
39mSecuring RAG Pipelines with Fine Grained Authorization
24mReBAC with Kyverno: Automating Multi-Tenant RBAC at Scale
25mModernizing Authorization for Cloud Native Applications Using OpenFGA
33mSecuring RAG Pipelines with Fine Grained Authorization