DB Resilience Architecture
This board maps the recommended mitigations for the PostgreSQL + pgvector single point of failure identified in the blast-radius analysis.
Two mitigations
- Read Replica — async streaming replica; re-points read-heavy services (SSO, Portal, NextAuth) so a primary stall does not block them
- Connection Pool Circuit Breaker — wraps all Drizzle ORM connections; opens on timeout/error threshold and fast-fails with 503 instead of hanging Lambda contexts
Zones
- Write Path (Primary) — services that must write to the primary: Onboarding, Org Members, User Profile, Billing, Deployment, Container Images
- Read Path (Replica) — services re-pointed to the replica: SSO & Platform Auth, Portal, Next.js Web (NextAuth sessions)
- Circuit Breaker Layer — sits in front of every DB connection; protects Lambda concurrency budget