Standby route behavior
Understand implemented standby checks and the infrastructure they still need.
The backend implements primary and standby listener behavior for a Cloudflare outage. Planned standby hosts use oauthcat.com, including api.oauthcat.com, auth.oauthcat.com and scim.oauthcat.com. This describes code behavior and target routing; it does not assert that DNS, certificates, an AWS listener or automatic outage activation are provisioned or live. It is not multi-region availability or a database backup.
Register before starting standby sign-ins
Section titled “Register before starting standby sign-ins”SAML metadata publishes primary ACS index 0 and standby index 1 with the same entity ID and certificates. OIDC needs its standby redirect URI registered at the IdP. The IT admin confirms that registration in the hosted setup portal; it is never inferred from the URL appearing in metadata.
Unregistered connections cannot start or finish a standby flow and return connection_standby_unavailable. Registration does not relax signature, destination, audience, domain or activation checks.
For SAML, standby_acs_selection: url is the default. If the configured IdP requires indexed requestable URLs, set standby_acs_selection: index through the connection update API: a standby AuthnRequest then sends AssertionConsumerServiceIndex="1" instead of a URL. This is an explicit configuration choice; core does not infer it from an IdP vendor. Primary requests still send the primary URL. In both modes, the response’s Destination must equal the URL where it arrives.
Client and browser boundaries
Section titled “Client and browser boundaries”Standby CORS accepts only the configured standby frontend origins; primary frontend origins are refused on that listener. Primary and standby are separate sites, so dashboard/portal cookies do not carry over. Existing users must authenticate again. New sign-ups are paused on standby with sign_up_paused; primary sign-ups also pause when the challenge service is unavailable.
The Node SDK includes standby handling. Changing a broker host also changes browser URLs and callback registration requirements; retain exact URI validation.
IdP SCIM clients normally retry the primary URL and catch up after recovery. Customer custom domains pointing at Cloudflare have no Stage-1 standby coverage. Outbound webhooks originate from the backend, but that does not guarantee delivery while a destination is unavailable. Broader failover orchestration and deployed infrastructure require separate verification.
Incident signals
Section titled “Incident signals”Core emits edge.failover.activated and edge.failover.resolved once per environment for each incident transition. Their edge_failover data carries the shared incident_id, transition status and occurred_at; these events are not organization scoped. Consume them through events and webhooks and use the Events API for catch-up after interrupted delivery. This does not assert that outage activation or standby infrastructure is deployed.
Customer failover notifications use events and status guidance, not customer emails. See status for the status-page boundary.
Prepare for a Cloudflare outage
Section titled “Prepare for a Cloudflare outage”Do these once, at setup time, so a standby activation needs nothing from you:
- Ask each customer’s IT admin to register the standby ACS URL (SAML) or standby redirect URI (OIDC). The standby URL is in your SAML metadata at index 1 and is
acs_urls[1]on the connection. Each IdP guide has a standby section: Okta, Microsoft Entra ID, AD FS, OneLogin and JumpCloud; Google Workspace accepts only one ACS URL, so it has no standby path. For other SAML IdPs and for OIDC, see the generic standby steps. - They confirm the registration on the connection’s page in the setup portal, which marks the standby URL as registered.
- Use an SDK build with standby handling, and keep its default hosts unless you run your own deployment.
- Subscribe to
edge.failover.activatedandedge.failover.resolvedif you want to tell your own users.
During an outage, users of registered connections sign in through the standby hosts. Users of unregistered connections get connection_standby_unavailable until the primary route recovers. Dashboard and portal users sign in again on the standby site.
The standby AWS listener is provisioned only once there is traffic for it. Until then, the standby API hosts have no DNS records: an SDK failover attempt fails fast and the SDK reports the primary error. Registering a standby URL before then is harmless. For recovery targets beyond a Cloudflare outage, see the SLA FAQ.