Introduction

Smart routing only helps when the origin pool behind it reflects reality. If Cloudflare Argo or related traffic steering keeps sending requests to an unhealthy backend, the problem is usually not the routing feature by itself but stale origin health, incomplete failover logic, or hostname rules that still target the wrong server. The fix is to verify what Cloudflare considers healthy and how that maps to the service actually receiving traffic.

Symptoms

  • Traffic continues reaching a degraded or failing origin instead of a healthy alternative
  • One backend keeps receiving requests despite obvious application failures
  • Failover behavior does not match the health state you expect
  • The issue started after origin pool changes, maintenance, or backend scaling
  • Some hostnames recover while others still route to the bad origin

Common Causes

  • Cloudflare health checks do not test the path that truly reflects application health
  • Origin pool membership or weighting still includes an unhealthy backend
  • Hostname-specific routing rules override the failover design you intended
  • Maintenance changes left one origin technically reachable but functionally broken
  • Health evaluation and traffic steering rely on stale assumptions about backend readiness

Step-by-Step Fix

  1. Identify the affected hostname and confirm which origin is currently receiving the traffic.
  2. Review the relevant Cloudflare routing, load balancing, or Argo-related configuration tied to that hostname.
  3. Check the health probe path and verify it reflects real application readiness rather than only a shallow TCP or generic HTTP success.
  4. Confirm the unhealthy origin is not still marked eligible because of stale weights, pool membership, or permissive health criteria.
  5. Inspect whether hostname-specific overrides or origin rules bypass the intended failover pool.
  6. Remove or isolate the unhealthy backend from active routing once you confirm it should not receive production traffic.
  7. Re-test requests through Cloudflare and watch whether traffic now lands on a healthy origin consistently.
  8. Validate recovery for adjacent hostnames and paths that may share the same origin pool.
  9. Keep health checks tied to meaningful app behavior so smart routing decisions reflect real service quality, not superficial reachability.