feat: add configurable rate limit retry delays [vibe by AI]

src/plugin.ts

@@ -352,7 +352,7 @@ async function persistAccountPool(
   });
 }
 
-function retryAfterMsFromResponse(response: Response): number {
+function retryAfterMsFromResponse(response: Response, defaultRetryMs: number = 60_000): number {
   const retryAfterMsHeader = response.headers.get("retry-after-ms");
   if (retryAfterMsHeader) {
     const parsed = Number.parseInt(retryAfterMsHeader, 10);
@@ -369,7 +369,7 @@ function retryAfterMsFromResponse(response: Response): number {
     }
   }
 
-  return 60_000;
+  return defaultRetryMs;
 }
 
 function parseDurationToMs(duration: string): number | null {
@@ -602,12 +602,14 @@ function recordAndGetCapacityBackoff(family: ModelFamily, model?: string | null)
  * @param accountIndex - The account index
  * @param quotaKey - The quota key (e.g., "gemini-cli", "gemini-antigravity", "claude")
  * @param serverRetryAfterMs - Server-provided retry delay (if any)
+ * @param maxBackoffMs - Maximum backoff delay in milliseconds (default 60000)
  * @returns { attempt, delayMs, isDuplicate } - isDuplicate=true if within dedup window
  */
 function getRateLimitBackoff(
   accountIndex: number, 
   quotaKey: string,
-  serverRetryAfterMs: number | null
+  serverRetryAfterMs: number | null,
+  maxBackoffMs: number = 60_000
 ): { attempt: number; delayMs: number; isDuplicate: boolean } {
   const now = Date.now();
   const stateKey = `${accountIndex}:${quotaKey}`;
@@ -617,7 +619,7 @@ function getRateLimitBackoff(
   if (previous && (now - previous.lastAt < RATE_LIMIT_DEDUP_WINDOW_MS)) {
     // Same rate limit event from concurrent request - don't increment
     const baseDelay = serverRetryAfterMs ?? 1000;
-    const backoffDelay = Math.min(baseDelay * Math.pow(2, previous.consecutive429 - 1), 60_000);
+    const backoffDelay = Math.min(baseDelay * Math.pow(2, previous.consecutive429 - 1), maxBackoffMs);
     return { 
       attempt: previous.consecutive429, 
       delayMs: Math.max(baseDelay, backoffDelay),
@@ -637,7 +639,7 @@ function getRateLimitBackoff(
   });
 
   const baseDelay = serverRetryAfterMs ?? 1000;
-  const backoffDelay = Math.min(baseDelay * Math.pow(2, attempt - 1), 60_000);
+  const backoffDelay = Math.min(baseDelay * Math.pow(2, attempt - 1), maxBackoffMs);
   return { attempt, delayMs: Math.max(baseDelay, backoffDelay), isDuplicate: false };
 }
 
@@ -1325,11 +1327,13 @@ export const createAntigravityPlugin = (providerId: string) => async (
                     tokenConsumed = false;
                   }
 
-                  const headerRetryMs = retryAfterMsFromResponse(response);
+                  const defaultRetryMs = (config.default_retry_after_seconds ?? 60) * 1000;
+                  const maxBackoffMs = (config.max_backoff_seconds ?? 60) * 1000;
+                  const headerRetryMs = retryAfterMsFromResponse(response, defaultRetryMs);
                   const bodyInfo = await extractRetryInfoFromBody(response);
                   const serverRetryMs = bodyInfo.retryDelayMs ?? headerRetryMs;
                   const quotaKey = headerStyleToQuotaKey(headerStyle, family);
-                  const { attempt, delayMs, isDuplicate } = getRateLimitBackoff(account.index, quotaKey, serverRetryMs);
+                  const { attempt, delayMs, isDuplicate } = getRateLimitBackoff(account.index, quotaKey, serverRetryMs, maxBackoffMs);
 
                   const rateLimitReason = parseRateLimitReason(bodyInfo.reason, bodyInfo.message, bodyInfo.quotaResetTime);
                   const isServiceCapacityExhausted = rateLimitReason === "SERVICE_CAPACITY_EXHAUSTED";

src/plugin/config/schema.ts

@@ -256,12 +256,28 @@ export const AntigravityConfigSchema = z.object({
   pid_offset_enabled: z.boolean().default(false),
 
    /**
-    * Switch to another account immediately on first rate limit (after 1s delay).
-    * When disabled, retries same account first, then switches on second rate limit.
+     * Switch to another account immediately on first rate limit (after 1s delay).
+     * When disabled, retries same account first, then switches on second rate limit.
+     * 
+     * @default true
+     */
+    switch_on_first_rate_limit: z.boolean().default(true),
+
+   /**
+    * Default retry delay in seconds when API doesn't return a retry-after header.
+    * Lower values allow faster retries but may trigger more 429 errors.
+    * 
+    * @default 60
+    */
+   default_retry_after_seconds: z.number().min(1).max(300).default(60),
+
+   /**
+    * Maximum backoff delay in seconds for exponential retry.
+    * This caps how long the exponential backoff can grow.
     * 
-    * @default true
+    * @default 60
     */
-   switch_on_first_rate_limit: z.boolean().default(true),
+   max_backoff_seconds: z.number().min(5).max(300).default(60),
 
    // =========================================================================
    // Health Score (used by hybrid and priority-queue strategies)
@@ -345,6 +361,8 @@ export const DEFAULT_CONFIG: AntigravityConfig = {
   account_selection_strategy: 'hybrid',
   pid_offset_enabled: false,
   switch_on_first_rate_limit: true,
+  default_retry_after_seconds: 60,
+  max_backoff_seconds: 60,
   auto_update: true,
   signature_cache: {
     enabled: true,