Function computeRateLimitWaitMs

src/util/fetch/index.ts:345–368

(response: Response)

 * Default: 60s.
 */
export function computeRateLimitWaitMs(response: Response): number {
  const rateLimitReset = response.headers.get('X-RateLimit-Reset');
  const retryAfter = response.headers.get('Retry-After');
  const openaiReset =
    response.headers.get('x-ratelimit-reset-requests') ||
    response.headers.get('x-ratelimit-reset-tokens');

  let waitTime = 60_000;

  if (openaiReset) {
    const parsedHeaders = parseRateLimitHeaders(Object.fromEntries(response.headers.entries()));
    if (parsedHeaders.resetAt !== undefined) {
      waitTime = Math.max(parsedHeaders.resetAt - Date.now(), 0);
    }
  } else if (rateLimitReset) {
    const resetTime = new Date(Number.parseInt(rateLimitReset) * 1000);
    const now = new Date();
    waitTime = Math.max(resetTime.getTime() - now.getTime() + 1000, 0);
  } else if (retryAfter) {
    waitTime = parseRetryAfter(retryAfter) ?? waitTime;
  }

  return waitTime;
}

/**
 * Maximum jitter (in ms) added to rate-limit waits. Spreads concurrent