[bug] OpenInference tracer fails to handle RemoveMessage from langchain_core

[AvinashMudireddy31]

Describe the bug
The _extract_message_role function in the OpenInference LangChain instrumentation tracer fails to handle RemoveMessage from langchain_core.messages.modifier. When processing a RemoveMessage class, the function raises a ValueError.

To Reproduce

1. Use LangChain with OpenInference instrumentation enabled.
2. Create a workflow or chain that uses RemoveMessage from langchain_core.messages.modifier (ex: Remove a message from complied Langgraph state).
3. The tracer will throw a ValueError with message: "Cannot parse message of type: RemoveMessage"

Expected behavior
The _extract_message_role function should properly handle RemoveMessage without throwing an error.

Screenshots
N/A

Desktop (please complete the following information):

• OS: macOS
• Python Version: 3.12
• langchain-core>=0.3.72
• openinference-instrumentation-langchain>=0.1.52
• arize-phoenix>=10.13.1

Additional context
Add any other context about the problem here.

[dosubot]

Hi @AvinashMudireddy31! I'm Dosu and I’m helping the openinference team.

This is a known issue: the OpenInference LangChain tracer's _extract_message_role (aka _parse_message_data) function does not recognize RemoveMessage from langchain_core.messages.modifier, so it raises a ValueError when it encounters one. This happens because the function only handles a fixed set of message types and falls through to an error for anything else, including RemoveMessage see code reference.

There is an open issue tracking this bug, and several users have confirmed the problem and proposed a workaround. The suggested patch is to add a branch for RemoveMessage in the message parsing logic, for example:

elif message_class_name.startswith("RemoveMessage"):
    role = "remove"

This will prevent the error and allow the tracer to process workflows that include message deletion see discussion and patch.

RemoveMessage is a modifier used to delete messages from the conversation state in LangChain workflows. It is not a conversational message, so you can either assign it a dummy role (like "remove") or skip it entirely, depending on your tracing needs see usage example.

Until an official fix is released, you can patch your local _extract_message_role (or _parse_message_data) function as above to avoid the error.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[caroger]

thanks @AvinashMudireddy31 for filing the issue.

we will take a look at this.