π OpenHands: AI-Driven Development
import { AxiosError } from "axios";
const BASE_DELAY_MS = 1000;
const MAX_DELAY_MS = 30000;
/**
* Extra backoff multiplier applied on top of the normal exponential curve
* once a request has actually been rate-limited (429). A plain network
* blip should back off quickly; a server telling us to slow down should
* back off harder so a burst of tabs/queries don't all retry in lockstep
* and immediately re-trip the limiter.
*/
const RATE_LIMIT_BACKOFF_MULTIPLIER = 2;
/**
* True for a 429 from either transport this app talks to a backend with:
* Axios (agent-server / legacy cloud calls) or the duck-typed `{ status }`
* shape `@openhands/typescript-client`'s `HttpError` uses for the cloud
* proxy. `HttpError` isn't imported here to keep this usable from any
* query regardless of which client raised the error.
*/
export function isRateLimitError(error: unknown): boolean {
if (error instanceof AxiosError) {
return error.response?.status === 429 || error.status === 429;
}
if (error && typeof error === "object" && "status" in error) {
return (error as { status?: unknown }).status === 429;
}
return false;
}
/**
* Default `retryDelay` for the shared query client: exponential backoff
* capped at 30s, with full jitter so concurrent queries (e.g. every open
* browser tab refetching on window focus at once) don't retry in sync and
* re-create the burst that got them rate-limited in the first place.
* Doubles the base delay on a confirmed 429 before applying jitter.
*/
export function getQueryRetryDelay(
attemptIndex: number,
error: unknown,
): number {
const multiplier = isRateLimitError(error)
? RATE_LIMIT_BACKOFF_MULTIPLIER
: 1;
const capped = Math.min(
BASE_DELAY_MS * multiplier * 2 ** attemptIndex,
MAX_DELAY_MS,
);
return Math.round(capped * (0.5 + Math.random() * 0.5));
}