langflow-ai/langflow · error · HTTPException

Error creating graph: {e}

Error message

Error creating graph: {e}

What it means

Raised when the flow module's get_graph() function itself throws during execution (sync or async). The loader logs the full traceback via logger.exception and wraps the original exception in HTTPException 500 with detail 'Error creating graph: {e}'. The finally block still cleans up the trace context and removes the module from sys.modules, so retrying re-imports fresh.

Source

Thrown at src/backend/base/langflow/agentic/services/helpers/flow_loader.py:202

    # to every tool. Any later deepcopy/pickle of those tools triggers
    # LangfuseResourceManager.__new__() with no kwargs and raises. Tracing belongs
    # to flow execution, not graph assembly, so we clear it for the duration.
    trace_token = None
    try:
        from langflow.services.tracing.service import trace_context_var

        trace_token = trace_context_var.set(None)
    except ImportError:
        trace_token = None

    try:
        if inspect.iscoroutinefunction(get_graph_func):
            graph = await get_graph_func(**kwargs)
        else:
            graph = get_graph_func(**kwargs)
    except Exception as e:
        logger.exception(f"Error executing get_graph(): {e}")
        raise HTTPException(status_code=500, detail=f"Error creating graph: {e}") from e
    finally:
        if trace_token is not None:
            from langflow.services.tracing.service import trace_context_var

            trace_context_var.reset(trace_token)
        if module_name in sys.modules:
            del sys.modules[module_name]

    validate_flow_for_current_settings(graph)
    return graph


async def load_graph_for_execution(
    flow_path: Path,
    flow_type: str,
    provider: str | None = None,
    model_name: str | None = None,
    api_key_var: str | None = None,

View on GitHub (pinned to 976ec789d2)

Solutions

1. Read the server logs — logger.exception records the full traceback of the underlying error; fix the root cause shown there.
2. Run the module standalone (python -c "import my_flow; my_flow.get_graph()") to reproduce the failure outside the loader.
3. If provider/model_name/api_key_var are being injected, make sure the graph factory accepts and uses them correctly.
4. Check that all dependencies imported inside get_graph are installed in the backend environment.

Example fix

# before
def get_graph():
    return Graph(openai_api_key=os.environ["OPENAI_API_KEY"])  # KeyError -> 500

# after
def get_graph(api_key_var: str | None = None):
    key = os.environ.get(api_key_var or "OPENAI_API_KEY", "")
    return Graph(openai_api_key=key)

try:
    graph = await load_flow_graph(flow_path, provider=provider, model_name=model_name)
except HTTPException as e:
    log.exception("get_graph failed for %s", flow_path)  # server log has the real traceback
    raise

When it happens

Trigger: flow_path resolves to a valid module with get_graph, but calling it raises: missing API key env var, invalid model name passed via kwargs (provider/model_name/api_key_var are forwarded when the signature accepts them), a component constructor failing, or an import error triggered lazily inside the factory.

Common situations: Provider/model_name arguments mismatched with what the flow expects; expired or unset credentials the graph builder reads at build time; code that works in a notebook but depends on cwd-relative paths; component version changes after a Langflow upgrade.

Related errors

• Could not load flow module: {flow_path}
• Error loading flow module: {e}
• Flow module must define 'get_graph()' function: {flow_path}
• parse_exception(exc)
• parse_exception(exc)

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/e8277cd39a369ebe. Report an issue: GitHub.