wtf( )unctionsystem design, drawn
← all problemsAuthenticationHard

The identity provider went down and half the platform did not notice

Every service here authenticates against the same OIDC provider, and the architecture diagram draws one arrow from each of them to it. On that diagram they all look equally dependent.

They are not. Some of these services ask the provider a question on every single request, and some were handed everything they need days ago and have not spoken to it since.

The identity provider is unreachable for four minutes. Tap every service that stops being able to authorise requests.
!A provider-side outage, four minutes, no warning. Support gets reports from some customers and silence from others, and the pattern makes no sense until somebody looks at how each service actually checks a token.One arrow per service on the diagram, and three different meanings behind it. Which arrow is a question and which is a memory decides who survives.

Boundaries, outermost first: Your platform: Admin console (own sessions), Sign-in page (auth code flow), Reports API (verifies locally), JWKS cache (keys, 24h), Mobile BFF (refresh grant), Orders API (introspects) Outside every boundary: Identity provider (OIDC, third party; FAILED: unreachable) Connections: Sign-in page calls Identity provider — a question, every sign-in Orders API calls Identity provider — a question, every request Mobile BFF calls Identity provider — a question, every refresh JWKS cache controls Identity provider — asked once a day Reports API calls JWKS cache — a memory, not a question

The identity provider has failed. Tap everything that stops working with it — not the failure itself.