Bug Description

Version: lfx 0.3.2

Hi! I think I found a bug that surfaces when using arun_flow_from_json in standalone mode (without a LangFlow server) on a flow that has frozen vertices. The execution crashes with a TypeError: 'NoneType' object is not subscriptable at lfx/graph/graph/base.py:1565.

I believe the issue isn't really about whether frozen nodes should work in standalone mode — the problem is that when chat_service is None, the fallback get_cache_func returns None instead of CacheMiss(), which causes an unhandled TypeError deep inside the graph with no hint that frozen is the root cause.

Here's the relevant code path. In two places within base.py (lines ~1446 and ~1671), the fallback is defined as:

# base.py ~L1446  (duplicated at ~L1671)
else:
    # Fallback no-op cache functions for tests or when service unavailable
    async def get_cache_func(*args, **kwargs):  # noqa: ARG001
        return None                              # returns None instead of CacheMiss()

Then, in build_vertex (~L1553), when a frozen vertex is encountered:

if not vertex.frozen or is_loop_component:
    should_build = True
else:
    if get_cache is not None:
        cached_result = await get_cache(key=vertex.id)  # returns None, not CacheMiss
    else:
        cached_result = CacheMiss()

    if isinstance(cached_result, CacheMiss):   # False — None is not CacheMiss
        should_build = True
    else:
        cached_vertex_dict = cached_result["result"]  # TypeError: None["result"]

The guard if get_cache is not None is True because get_cache_func is a valid callable. So isinstance(None, CacheMiss) is False, the code falls into the else branch, and None["result"] raises TypeError. This also isn't caught by the adjacent except KeyError, so it propagates all the way up and crashes the flow.

The same pattern appears in both execute_vertex (~L1440) and process (~L1665).

Reproduction

1. Create a flow in the LangFlow UI and freeze at least one component (toggle the "Freeze" option on any node).
2. Export the flow as JSON.
3. Execute it standalone:

import asyncio
import json
from lfx.load import arun_flow_from_json

with open("frozen_flow.json") as f:
    flow = json.load(f)

asyncio.run(arun_flow_from_json(flow=flow, input_value="hello", fallback_to_env_vars=True))

Actual result:

ValueError: Error running graph: 'NoneType' object is not subscriptable
  File ".../lfx/graph/graph/base.py", line 1565, in build_vertex
    cached_vertex_dict = cached_result["result"]
                         ~~~~~~~~~~~~~^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable

Workaround: unfreeze all nodes in the flow before execution, or strip "frozen": true from the flow JSON before passing it to arun_flow_from_json.

Expected behavior

When chat_service is unavailable, it would be great if the code handled this case more gracefully. Two approaches that would both work well:

• Option A (silent fallback): have the fallback get_cache_func return CacheMiss() instead of None, so frozen vertices are simply rebuilt when no cache is available. This is a one-line change in two places:

  async def get_cache_func(*args, **kwargs):  # noqa: ARG001
      return CacheMiss()  # was: return None

• Option B (explicit error): raise a descriptive error early that identifies the frozen vertex and explains what's needed, so the user knows exactly what to fix:

  if vertex.frozen and chat_service is None:
      raise ValueError(
          f"Vertex '{vertex_id}' is frozen but no cache service is available. "
          "Frozen nodes require a running LangFlow server. "
          "Unfreeze the component or run the flow through the LangFlow API."
      )

Option A feels like the lower-risk change, but Option B gives a much better developer experience. Either way, the current TypeError surfacing from deep inside the graph is quite hard to debug without knowing where to look.

Thanks for taking a look!

Who can help?

@ogabrielluiz @italojohnny

Operating System

Ubuntu Linux 22.04

Langflow Version

1.8.2

Python Version

None

Screenshot

No response

Flow File

No response