sgl-project/sglang · error

Error in stream_executor: {get_exception_traceback()}

Error message

Error in stream_executor: {get_exception_traceback()}

What it means

This is the SGLang restricted-python interpreter's stream executor thread reporting that an inline expression you scheduled (via queue) raised a Python exception. The traceback of the original exception is embedded in the warning text; the worker stores the exception and stops processing further queued expressions.

Source

Thrown at python/sglang/lang/interpreter.py:434

    def end(self):
        if self.use_thread:
            if self.worker.is_alive():
                self.queue.put(None)
        self.backend.end_program(self)

    def _thread_worker_func(self):
        error = None

        while True:
            expr = self.queue.get()
            if expr is None:
                self.queue.task_done()
                break

            try:
                self._execute(expr)
            except Exception as e:
                warnings.warn(f"Error in stream_executor: {get_exception_traceback()}")
                error = e
                break
            self.queue.task_done()
            if self.stream_text_event:
                self.stream_text_event.set()

        # Clean the queue and events
        if error is not None:
            try:
                while True:
                    self.queue.task_done()
                    self.queue.get_nowait()
            except queue.Empty:
                pass
            for name in self.variable_event:
                self.variable_event[name].set()
            if self.stream_var_event:
                for name in self.stream_var_event:

View on GitHub (pinned to 0132848349)

Solutions

1. Read the embedded traceback inside the warning text — it names the real failing expression and original exception
2. Fix the offending SGL expression (undefined name, wrong arguments, backend error) in your program
3. Run the same logic synchronously (non-stream mode) to get a cleaner stack trace while debugging
4. If the error is from a backend call, check endpoint availability/API keys before retrying

Example fix

# before
gen("Write a story", max_tokens=nonexistent_var)  # NameError inside stream executor
# after
max_tokens = 512
gen("Write a story", max_tokens=max_tokens)

try:
    result = program(runner, batch)
except Exception as e:
    # the stream worker re-raises the stored exception after the warning
    logging.exception("SGL stream executor failed: %s", e)

When it happens

Trigger: Using sglang.lang (SGL front-end) with stream/async execution: an expression submitted to the stream_executor (e.g. gen, call, control flow primitive) raises at runtime — NameError, TypeError, model API error, etc. The warning is emitted from _thread_worker_func and the failure surfaces later as the stored exception.

Common situations: Writing SGL programs with the sglang frontend where a function references an undefined variable, passes wrong argument types to primitives, or a backend call (OpenAI endpoint) fails mid-stream; upgrading sglang versions where primitive signatures changed.

Related errors

• Crusoe API key required. Pass api_key= or set CRUSOE_API_KEY
• This use case is not supported if api speculative execution
• Unknown dtype: {sampling_params.dtype}
• This use case is not supported. For OpenAI chat models, sgl.
• select/choices is not supported for chat models. Please try

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ae86645c67e59978. Report an issue: GitHub.