apache/seatunnel · error · RuntimeError

Bedrock Mantle stream ended abnormally

Error message

Bedrock Mantle stream ended abnormally: {message}

What it means

Raised by chat_stream() when the Mantle Responses stream terminates with a failure-class event (response.failed, response.incomplete, response.error, or a generic error) that carries no response object for _raise_on_terminal_status to interpret. The event's message (or the event type itself) is embedded in the RuntimeError.

Solutions

  1. Retry the request, ideally with exponential backoff, since many stream failures are transient
  2. Log the embedded message/etype and check AWS Bedrock service health and quotas
  3. Catch the RuntimeError around stream iteration and restart the stream from the last received chunk

Example fix

// before
for chunk in provider.chat_stream(messages):
    process(chunk)
// after
for attempt in range(3):
    try:
        for chunk in provider.chat_stream(messages):
            process(chunk)
        break
    except RuntimeError as e:
        if "ended abnormally" in str(e) and attempt < 2:
            time.sleep(2 ** attempt)
            continue
        raise
Defensive patterns

Strategy: retry

Validate before calling

if not provider:
    raise ValueError("provider must be initialized before streaming")

Type guard

def is_abnormal_stream_end(e: BaseException) -> bool:
    return isinstance(e, RuntimeError) and "ended abnormally" in str(e)

Try / catch

for attempt in range(3):
    try:
        for chunk in provider.chat_stream(messages):
            handle(chunk)
        break
    except RuntimeError as e:
        if "ended abnormally" in str(e) and attempt < 2:
            time.sleep(2 ** attempt)
        else:
            raise

Prevention

When it happens

Trigger: Streaming via chat_stream() and receiving etype in {"response.failed","response.incomplete","response.error","error"} where event.response is None, so the fallback message = getattr(event, "message", "") or etype path raises.

Common situations: Transient Bedrock service errors during generation; malformed requests rejected mid-stream; throttling or internal model errors surfaced only as stream events.

Understand the failure class

Background: 'Something went wrong' / 'Request failed (500)' / 'HTTP error! status: 404' — what failed HTTP requests actually mean and how to find the real cause — this error's family across 28 libraries.

Related errors

AI-assisted analysis of apache/seatunnel@cf67b549a7 (2026-09-10). Data as JSON: /api/errors/80bf8b14834c08ca. Report an issue: GitHub.

Appendix: source

Thrown at seatunnel-cli/seatunnel_cli/llm_provider.py:1244

                if item_type == "function_call" and current_tool_id:
                    yield {"type": "tool_stop", "tool_use_id": current_tool_id}
                    current_tool_id = None
                elif item_type == "reasoning":
                    yield {
                        "type": "mantle_responses_item",
                        "item": self._dump_item(item),
                    }
            elif etype == "response.refusal.done":
                refusal = getattr(event, "refusal", "") or ""
                raise RuntimeError(
                    f"Bedrock Mantle model refused the request: {refusal}")
            elif etype in ("response.failed", "response.incomplete",
                           "response.error", "error"):
                resp = getattr(event, "response", None)
                self._raise_on_terminal_status(resp) if resp is not None \
                    else None
                message = getattr(event, "message", "") or etype
                raise RuntimeError(
                    f"Bedrock Mantle stream ended abnormally: {message}")
            elif etype == "response.completed":
                self._raise_on_terminal_status(getattr(event, "response", None))
                yield {
                    "type": "message_stop",
                    "stop_reason": "tool_use" if saw_tool_use else "end_turn",
                }


# ─── OrcaRouter Provider ───

class OrcaRouterProvider(OpenAIProvider):
    """OrcaRouter AI gateway provider (OpenAI Chat Completions API).

    OrcaRouter is an OpenAI-compatible gateway that exposes many models (and
    provider/model routing namespaces) behind one endpoint. Models are
    addressed as ``provider/model``, e.g. ``orcarouter/auto`` routes and
    grades automatically. Defaults to the ``orcarouter/auto`` model, which

View on GitHub (pinned to cf67b549a7)