Repository navigation
Bug: LangchainIntegration crashes on Python 3.14 — ModuleNotFoundError: No module named 'langchain_classic' #6265
Description
How do you use Sentry?
Sentry Saas (sentry.io)
Version
2.60.0
Steps to Reproduce
Bug: LangchainIntegration crashes on Python 3.14 — ModuleNotFoundError: No module named 'langchain_classic'
Environment
- sentry-sdk: 2.57.0 (reproduced through 2.60.0)
- Python: 3.14.3 (
python:3.14-slim-bookworm) - langchain: 0.3.25 / langchain-core: 0.3.75
- Introduced by: feat(langchain): Support v1 #4874 (feat(langchain): Support v1, merged Oct 23 2025)
Steps to Reproduce
- Use Python 3.14
- Install
sentry-sdk==2.60.0+langchain==0.3.25 - Call
sentry_sdk.init()with any integration that auto-enablesLangchainIntegration - Start gunicorn workers
Expected Result
LangchainIntegration degrades gracefully — AgentExecutor falls back to None and the application starts normally.
Actual Result
Worker process crashes at startup with:
ModuleNotFoundError: No module named 'langchain_classic'
from langchain_classic.agents import AgentExecutor
File ".../sentry_sdk/integrations/langchain.py", line 60, in
File ".../gunicorn/arbiter.py", line 608, in spawn_worker
error [ERROR] Exception in worker process
Root cause — two problems in langchain.py lines 58–66:
try:
from langchain_classic.agents import AgentExecutor # >=v1
except ImportError:
try:
from langchain.agents import AgentExecutor # <v1
except ImportError:
AgentExecutor = None
1. langchain-classic is uninstallable with langchain 0.3.x — it forces langchain-core>=1.4.0, breaking langchain 0.3.x, langgraph 0.3.x, and
langgraph-prebuilt.
2. Python 3.14 breaks the fallback — langchain.agents uses __getattr__-based lazy loading. On Python 3.14 the fallback raises an exception not
caught by except ImportError, propagating the crash.
Workaround:
sentry_sdk.init(
dsn="...",
integrations=[DjangoIntegration())],
auto_enabling_integrations=False,
)
Activity
linear-code commentedon May 13, 2026
PY-2416
ericapisani commentedon May 14, 2026
@SebastianDT1 Thanks for opening an issue with us, I've added this to the team's backlog and someone will take a look.
alexander-alderman-webb commentedon May 15, 2026
Problem occurs when execute the
class Chainstatement of thelangchain.chains.base.Chainclass.The
langchain.agentsmodule we load also imports the class.A related bug report is pydantic/pydantic#13036
Here's the Dockerfile to repro (assumes you have a
repro.py, ideally with some LangChain imports):