RedisCache only accepts values that can be pickled

[codenprogressive]

Bug Description

The flow takes an input and an Agent components. See attached.

I'm using Redis for cache. Executing the flow will result into this error:

RedisCache only accepts values that can be pickled. 

Traceback (most recent call last):
  File "/langflow_install_workspace/langflow/src/backend/base/langflow/services/cache/service.py", line 245, in set
    if pickled := dill.dumps(value, recurse=True):
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/langflow_install_workspace/langflow/.venv/lib/python3.12/site-packages/dill/_dill.py", line 289, in dumps
    dump(obj, file, protocol, byref, fmode, recurse, **kwds)#, strictio)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

I did another test using chat input and Large Model components. I got a slight different error but all are pointing to /src/backend/base/langflow/services/cache/service.py , line if pickled := dill.dumps(value, recurse=True)

cannot pickle 'SSLContext' object

Traceback (most recent call last):
  File "/langflow_install_workspace/langflow/src/backend/base/langflow/api/build.py", line 349, in _build_vertex
    vertex_build_result = await graph.build_vertex(
                          ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/langflow_install_workspace/langflow/src/lfx/src/lfx/graph/graph/base.py", line 1624, in build_vertex
    await set_cache(key=vertex.id, data=vertex_dict)
  File "/langflow_install_workspace/langflow/src/backend/base/langflow/services/chat/service.py", line 37, in set_cache
    await self.cache_service.upsert(str(key), result_dict, lock=lock or self.async_cache_locks[key])
  File "/langflow_install_workspace/langflow/src/backend/base/langflow/services/cache/service.py", line 272, in upsert
    await self.set(key, value)
  File "/langflow_install_workspace/langflow/src/backend/base/langflow/services/cache/service.py", line 245, in set
    if pickled := dill.dumps(value, recurse=True):
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/langflow_install_workspace/langflow/.venv/lib/python3.12/site-packages/dill/_dill.py", line 289, in dumps
    dump(obj, file, protocol, byref, fmode, recurse, **kwds)#, strictio)

Reproduction

1. enable Redis Cache LANGFLOW_CACHE_TYPE=redis
2. add two components: chat input and Large Model components
3. ask a question
4. error will popup

Expected behavior

If Redis does not support certain object then bypass that pickle action.

Who can help?

No response

Operating System

Linux (debian 12)

Langflow Version

1.9.3

Python Version

3.12

Screenshot

No response

Flow File

No response

[codenprogressive]

#13170 #13169 #13167 #13166 #13165

[andifilhohub]

LE-1242

[andifilhohub]

the fix has been merged

[codenprogressive]

Hi @andifilhohub, I upgraded to release 1.10.0. The issue is still occurring. can you please reopen the issue? (I don't have permissions)

The flow builds successfully, then fails when Langflow writes the vertex state to Redis cache.

Here is the chain from my logs:

1. Graph.build_vertex() builds LanguageModelComponent-H38jo
2. After build, it caches the vertex dict:

base.py Lines 1698-1708

                if set_cache is not None:
                    vertex_dict = {
                        "built": vertex.built,
                        "results": vertex.results,
                        "artifacts": vertex.artifacts,
                        "built_object": vertex.built_object,
                        "built_result": vertex.built_result,
                        "full_data": vertex.full_data,
                    }
                    await set_cache(key=vertex.id, data=vertex_dict)

3. ChatService.set_cache → RedisCache.set → dill.dumps(value, recurse=True)

service.py Lines 280-289

    async def set(self, key, value, lock=None) -> None:
        try:
            if pickled := dill.dumps(value, recurse=True):
                result = await self._client.setex(self._key(key), self.expiration_time, pickled)
                ...
        except pickle.PicklingError as exc:
            msg = "RedisCache only accepts values that can be pickled. "
            raise TypeError(msg) from exc

4. dill walks into the built LLM object: ChatGoogleGenerativeAI (langchain-google-genai 4.1.3) → google.genai API client → _async_httpx_client_args containing {'verify': <ssl.SSLContext ...>}
SSLContext (and also httpx clients, thread locks, retry objects) cannot be pickled. That is why I see the error even for a simple "hello" input — the model client is created during build, and caching that live object is what breaks.

my config uses Redis:
.env.runtime Lines 114-114

LANGFLOW_CACHE_TYPE=redis

Here is a snipet code to reproduce the issue:

import dill, ssl
from langchain_google_genai import ChatGoogleGenerativeAI
from pydantic import SecretStr

model = ChatGoogleGenerativeAI(model='gemini-2.5-flash', google_api_key=SecretStr('test-key'))
try:
    dill.dumps(model, recurse=True)
    print('pickle OK')
except Exception as e:
    print(type(e).__name__, e)

TypeError cannot pickle 'SSLContext' object