run402-mcp/sdk/dist/retry.d.ts

MCP server for Run402 — full-stack backend infrastructure for AI agents: Postgres, auth, storage, serverless functions and atomic deploys. Paid with x402/MPP. Includes $0.03 image generation.

/**
 * `withRetry(fn, opts?)` — exponential-backoff retry helper for SDK
 * operations. Defaults to the canonical {@link isRetryableRun402Error}
 * policy. `safeToRetry` alone does not trigger a retry; it only says a
 * repeated mutation should not duplicate/corrupt state.
 *
 * Caller-controlled by design: the helper is policy-only. It does NOT own
 * idempotency keys. Pair with the SDK method's own `idempotencyKey`
 * option by baking it into the closure:
 *
 * @example
 *   const release = await withRetry(
 *     () => r.project(id).apply(spec, { idempotencyKey: "my-deploy-2026-05-01" }),
 *     { attempts: 3, onRetry: (e, n, ms) => console.warn(`retry ${n} in ${ms}ms`) },
 *   );
 *
 * The closure carries the same `idempotencyKey` on every retry, so the
 * gateway dedups duplicates server-side and a retried mutation does not
 * double-apply.
 */
export interface RetryOptions {
    /** Total attempts (1 = no retry, 2 = 1 retry, etc.). Default: 3. */
    attempts?: number;
    /** Initial delay before the first retry. Default: 250 ms. */
    baseDelayMs?: number;
    /** Cap on the per-attempt delay. Default: 5_000 ms. */
    maxDelayMs?: number;
    /**
     * Override the retry decision. Default: {@link isRetryableRun402Error}.
     * Receives the thrown error and the 1-based index of the attempt that just
     * failed; returning `true` schedules another attempt (subject to the
     * `attempts` cap), `false` re-throws immediately.
     */
    retryIf?: (error: unknown, attempt: number) => boolean;
    /**
     * Called synchronously after each retryable failure, before the delay
     * starts. Useful for logging / telemetry. `attempt` is the 1-based index
     * of the attempt that just failed; `delayMs` is the wait until the next
     * attempt. Throws inside this callback are swallowed — a buggy logger
     * cannot abort the retry chain.
     */
    onRetry?: (error: unknown, attempt: number, delayMs: number) => void;
}
/**
 * Run `fn` with exponential backoff, retrying retryable {@link Run402Error}
 * subclasses up to `attempts` times. After exhausting attempts (or when
 * `retryIf` returns false), throws the LAST observed error — preserving the
 * structured envelope so the caller's catch handler can branch on `kind`.
 */
export declare function withRetry<T>(fn: () => Promise<T>, opts?: RetryOptions): Promise<T>;
//# sourceMappingURL=retry.d.ts.map