Troubleshooting Relay traffic
Diagnose queue waits, endpoint cooldowns, route eligibility, usage estimates, and missing request paths.
Last updated July 10, 2026
A request remains waiting
Inspect the queue item’s eligible time, delay reason, applicable resource limits, and any hosted user defer scope. A request may wait for a request/token/spend window, active concurrency, an endpoint cooldown, or a user-specific policy reset.
No route is selected
Confirm that the incoming model or explicit lane resolves to at least one enabled endpoint with a compatible route kind. Check provider and endpoint enabled state, lane memberships, manual ranks, wait budget, cooldowns, and maximum estimated cost.
An endpoint is cooling down
Review the cooldown reason and status code. A 429 or compatible throttle can honor Retry-After; connection failures, timeouts, and 5xx responses use bounded failure backoff. Do not clear cooldowns without understanding the upstream condition.
Usage is marked estimated
Relay uses provider-reported usage when it is present and valid. Otherwise it falls back to request/response estimation. Ensure the endpoint pricing policy and upstream usage fields are configured and returned as expected.
A local endpoint recurses into Relay
Do not configure a provider base URL that points back at Relay’s own routed /v1 gateway. Configure the actual upstream model service URL. Relay rejects recognized self-referential paths before an upstream request is made.