Skip to content

Interface: RetryConfig

Resolved policy for absorbing upstream 429 Too Many Requests responses.

Databricks Foundation Model endpoints are pay-per-token with an account-level throughput ceiling that is not exposed as a readable number, so the proxy cannot pace against it proactively - it can only react. When enabled, a 429 is retried in-proxy with exponential backoff (honoring any Retry-After) instead of being relayed to the client, so an agentic caller like Codex sees a slow success rather than “exceeded retry limit”.

baseDelayMs: number

First backoff, doubled each attempt (before jitter and the Retry-After override).


enabled: boolean

Retry 429s in-proxy rather than relaying them.


maxDelayMs: number

Ceiling for any single backoff, including a server-sent Retry-After.


maxRetries: number

Maximum retry attempts after the initial try (so N+1 total requests).