What causes "WebSocket connection closed" in iflytek/astron-agent?

Retry async_call_spark_fc once or twice with backoff — a dropped connection is often transient

iflytek/astron-agent · critical · CustomException

SPARK_REQUEST_ERROR

SPARK_REQUEST_ERROR

Error message

WebSocket connection closed

What it means

SPARK_REQUEST_ERROR raised inside _recv_messages when websockets.ConnectionClosed escapes the recv loop — the Spark WebSocket dropped before a final (status==2) frame with the function_call was received. The client never got a complete answer, so no function name/arguments can be returned.

Solutions

Retry async_call_spark_fc once or twice with backoff — a dropped connection is often transient
Inspect the span's error events and Spark response frames logged before the close for a server-side close code/reason
Verify network path (proxy, firewall, LB idle timeout) allows long-lived outbound WSS connections
Check Spark service status / your appid quota and concurrency limits that may cause the server to drop connections

Example fix

// before
name, usage, args = await spark_fc.async_call_spark_fc(user_input, span)
// after
for attempt in range(3):
    try:
        name, usage, args = await spark_fc.async_call_spark_fc(user_input, span)
        break
    except CustomException as e:
        if 'WebSocket connection closed' in str(e.err_msg) and attempt < 2:
            await asyncio.sleep(2 ** attempt)
            continue
        raise

Defensive patterns

Strategy: retry

Validate before calling

# preflight: check endpoint reachability
import socket
host = urlparse(model_url).hostname
socket.getaddrinfo(host, 443)  # raises early on DNS/network issues

Try / catch

for attempt in range(3):
    try:
        return await spark_fc.async_call_spark_fc(user_input, span)
    except CustomException as e:
        if 'WebSocket connection closed' in str(e.err_msg) and attempt < 2:
            await asyncio.sleep(2 ** attempt)
            continue
        raise

Prevention

Implement automatic retry with exponential backoff on transient WS drops
Ensure proxies/LB idle timeouts exceed the expected request duration
Monitor Spark endpoint availability and alert on elevated drop rates
Keep requests short; chunk very long conversations

When it happens

Trigger: During await ws_handle.recv() in the receive loop, the WebSocket protocol transitions to CLOSED (server closed the connection, network drop, or idle-connection teardown) before the function-call result frame arrives.

Common situations: Spark gateway closing long-idle connections; corporate proxy/firewall killing WebSockets; server restart or load-balancer timeout mid-request; unstable network between service and Spark endpoint.

Understand the failure class

Background: ECONNREFUSED and "connection refused" / "could not connect to server" errors: what they mean and how to fix them — this error's family across 44 libraries.

Connection failures: ECONNREFUSED, ECONNRESET, and friends — why connections get refused, reset, or dropped.

Related errors

AI-assisted analysis of iflytek/astron-agent@5e758547a8 (2026-09-12). Data as JSON: /api/errors/9d0c567172f8867a. Report an issue: GitHub.

Appendix: source

Thrown at core/workflow/infra/providers/llm/iflytek_spark/spark_fc_llm.py:137

                        cause_error="Sensitive content detected, LLM did not find function_call field",
                    )
                if status != 2:
                    continue
                token_usage = msg["payload"]["usage"]["text"]
                if "function_call" not in msg["payload"]["choices"]["text"][0]:
                    raise CustomException(
                        err_code=CodeEnum.SPARK_FUNCTION_NOT_CHOICE_ERROR,
                        err_msg="Cannot find function_call field in LLM response",
                        cause_error="Cannot find function_call field in LLM response",
                    )

                name = msg["payload"]["choices"]["text"][0]["function_call"]["name"]
                arguments = msg["payload"]["choices"]["text"][0]["function_call"][
                    "arguments"
                ]
                return name, token_usage, arguments
            except websockets.ConnectionClosed:
                raise CustomException(
                    err_code=CodeEnum.SPARK_REQUEST_ERROR,
                    err_msg="WebSocket connection closed",
                    cause_error="WebSocket connection closed",
                )
            except Exception as e:
                raise CustomException(
                    err_code=CodeEnum.SPARK_REQUEST_ERROR,
                    err_msg=f"{e}",
                    cause_error=f"{e}",
                )

    async def _process_message(
        self, msg: dict, span: Span
    ) -> tuple[str | None, dict | None, str | None]:
        """
        Process a single function call message from the API response.

        :param msg: Message dictionary from Spark API

View on GitHub (pinned to 5e758547a8)