fix: unbounded 1-second retry loop on composer send #459

New issue

Open

opened 2026-07-25 23:35:25 +02:00 by dries · 0 comments

dries commented

2026-07-25 23:35:25 +02:00

Owner

Code-quality audit (P2). frontend/src/components/assistant/Composer.tsx:564-586.

const runSend = async (text, imgs) => {
  sendingRef.current = true;
  setSending(true);
  while (mountedRef.current) {
    try {
      await onSendRef.current?.(text, imgs);
      ...
      break;
    } catch (err) {
      if (!(err instanceof BackendUnavailableError)) break;
      await new Promise(resolve => window.setTimeout(resolve, 1_000));
    }
  }
  ...
};

Why it matters

No attempt cap, no backoff, no user-visible cancel. The only exit is unmounting the component. With the backend down this is one request per second, indefinitely, from every open tab — and the composer stays locked (sending === true) the whole time, so the user cannot edit, cancel, or copy their message out. A tab left open overnight against a stopped backend is thousands of failed requests.

Suggested fix

Cap at ~5 attempts with exponential backoff (1s, 2s, 4s, 8s, 16s), then give up and route the message through the existing failedSends path so the user gets it back and can retry deliberately.

Acceptance criteria

Retries are bounded and backed off.
After exhaustion the composer unlocks and the message surfaces via failedSends.
Test: onSend always throws BackendUnavailableError — assert a bounded call count and that sending returns to false.

Effort: S.

Code-quality audit (P2). `frontend/src/components/assistant/Composer.tsx:564-586`. ```js const runSend = async (text, imgs) => { sendingRef.current = true; setSending(true); while (mountedRef.current) { try { await onSendRef.current?.(text, imgs); ... break; } catch (err) { if (!(err instanceof BackendUnavailableError)) break; await new Promise(resolve => window.setTimeout(resolve, 1_000)); } } ... }; ``` ## Why it matters No attempt cap, no backoff, no user-visible cancel. The only exit is unmounting the component. With the backend down this is one request per second, indefinitely, from every open tab — and the composer stays locked (`sending === true`) the whole time, so the user cannot edit, cancel, or copy their message out. A tab left open overnight against a stopped backend is thousands of failed requests. ## Suggested fix Cap at ~5 attempts with exponential backoff (1s, 2s, 4s, 8s, 16s), then give up and route the message through the existing `failedSends` path so the user gets it back and can retry deliberately. ## Acceptance criteria - [ ] Retries are bounded and backed off. - [ ] After exhaustion the composer unlocks and the message surfaces via `failedSends`. - [ ] Test: `onSend` always throws `BackendUnavailableError` — assert a bounded call count and that `sending` returns to `false`. Effort: S.