What causes "Could not reach YouTube." in unslothai/unsloth?

Verify egress: curl https://www.youtube.com from the same host.

unslothai/unsloth · error · HTTPException

Could not reach YouTube.

Error message

Could not reach YouTube.

What it means

502 from POST /youtube/transcript: fetch_transcript failed with an httpx.HTTPError — the outbound HTTP layer could not complete requests to YouTube (DNS failure, connection refused/reset, TLS error, or timeout). The backend logs the specifics ('YouTube transcript fetch failed for <id>') and maps it to a gateway-style 502 because the failure is upstream, not in the request itself.

Source

Thrown at studio/backend/routes/youtube.py:65

    text: str
    truncated: bool


@router.post("/transcript", response_model = TranscriptResponse)
async def get_transcript(
    request: TranscriptRequest, current_subject: str = Depends(get_current_subject)
) -> TranscriptResponse:
    video_id = extract_video_id(request.url)
    if video_id is None:
        raise HTTPException(status_code = 400, detail = "That is not a YouTube video link.")

    try:
        transcript = await fetch_transcript(video_id, request.languages)
    except TranscriptUnavailable as error:
        raise HTTPException(status_code = 422, detail = str(error)) from error
    except httpx.HTTPError as error:
        logger.warning(f"YouTube transcript fetch failed for {video_id}: {error}")
        raise HTTPException(
            status_code = 502,
            detail = "Could not reach YouTube.",
        ) from error

    return TranscriptResponse(
        videoId = transcript.video_id,
        url = watch_url(transcript.video_id),
        title = transcript.title,
        author = transcript.author,
        lengthSeconds = transcript.length_seconds,
        language = transcript.language,
        languageCode = transcript.language_code,
        isGenerated = transcript.is_generated,
        text = transcript.text,
        truncated = transcript.truncated,
    )

View on GitHub (pinned to 203007d190)

Solutions

Verify egress: curl https://www.youtube.com from the same host.
Configure proxy env vars (HTTPS_PROXY) or the proxy's CA in httpx if behind a corporate proxy.
Retry after a short backoff — many occurrences are transient throttles.
Check the server log line for the precise httpx error class (ConnectError vs ReadTimeout guides the fix).

Example fix

// client: bounded retry with backoff
for (let i = 0; i < 3; i++) {
  try { return await api.post('/youtube/transcript', { url }); }
  catch (e) { if (e.status !== 502 || i === 2) throw e; await sleep(2 ** i * 1000); }
}

Defensive patterns

Strategy: retry

Validate before calling

// client-side reachability probe before the call (optional)
const online = typeof navigator === 'undefined' ? true : navigator.onLine;
if (!online) throw new Error('You appear to be offline.');

Try / catch

async function withRetry(fn, attempts = 3) {
  for (let i = 0; ; i++) {
    try { return await fn(); }
    catch (e) {
      if (e?.status !== 502 || i >= attempts - 1) throw e;
      await new Promise(r => setTimeout(r, 2 ** i * 1000));
    }
  }
}

Prevention

Apply exponential-backoff retry only to 502 (upstream/network), not 4xx
Configure HTTPS_PROXY/CA for corporate networks on the backend host
Check the backend log for the underlying httpx error class before assuming code fault

When it happens

Trigger: Studio host has no internet or a blocking firewall/proxy; YouTube returns RSTs / rate-limits the IP; corporate MITM proxy with an untrusted CA breaks TLS; transient network blip during fetch.

Common situations: Air-gapped or proxied dev environments; heavy polling from one IP triggering throttling; DNS misconfiguration in a container.

Related errors

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/075a07811e577bf7. Report an issue: GitHub.