mirror of
https://github.com/langchain-ai/langgraph.git
synced 2026-10-14 20:25:23 +02:00
Igor SoarezandGitHub
87f1c8eb9a
### Problem `Pregel.astream_events` declared `interrupt_before`, `interrupt_after` and `control` as keyword-only parameters, but forwarded them only on the `version="v3"` branch. On async `version="v1"`/`"v2"` the values were bound to named parameters and never entered `**kwargs`, so they were silently dropped on the way to `astream` — no error, no effect: - `request_drain()` on a caller-supplied `RunControl` did nothing; the graph created its own control and ran to completion. - Static interrupts passed as `astream_events(..., interrupt_before=[...])` never fired. This is a regression introduced by #7677 (first released in `1.2.0a3`). Before that, `Pregel` did not override `astream_events`, and these keywords reached `astream` through `**kwargs` — langchain-core's v1 and v2 event implementations both forward caller kwargs to `astream` (`tracers/log_stream.py`, `tracers/event_stream.py`). Downstream impact: langgraph-api passes run-level `interrupt_before` / `interrupt_after` into `graph.astream_events(..., version="v2", **kwargs)` for runs with `stream_mode="events"` (`models/run.py`, `stream.py`), so server runs using that stream mode have silently dropped static interrupts since langgraph 1.2.0a3. Sync `stream_events(version="v1"/"v2")` is unaffected: langchain-core raises `NotImplementedError` there, so there was no working sync path to break. ### Fix Remove the three parameters from the public runtime implementations of `stream_events` / `astream_events` (and from the explicit passing in their v3 dispatch calls), so the values travel through `**kwargs` again: - **v1/v2** recover the pre-#7677 passthrough. `astream` receives exactly what the caller passed — a supplied value, an explicit `None`, or nothing for an omitted argument. - **v3** is unchanged. The values bind on `_pregel_stream_v3` / `_apregel_stream_v3`, which keep their named parameters and since-inception (#7519, `1.2.0a1`) explicit-`None` defaults: an omitted argument arrives at `astream` as `None`, exactly as before. What `astream` receives after this change: | caller supplies | async v1/v2 | v3 (sync and async) | |---|---|---| | nothing | argument absent | explicit `None` | | explicit `None` | `None` | `None` | | a value | the value | the value | The typed `version="v3"` `@overload` stubs are untouched, so v3 call sites keep their static types (invalid values still fail `ty` there, same as on `main`). `RemoteGraph` and the v3-only `transformers` parameter are untouched. Complexity: net −24 lines in `main.py` with no branches, sentinels or wrappers added. The public capture-and-drop pattern that caused the bug is gone; the two private helpers each have a single call path and forward unconditionally, so they cannot drop anything. Docstrings now describe the three parameters as honored on every async version (type-checked only on the v3 overloads), and no longer promise synchronous v1/v2 event streaming, which langchain-core does not implement. ### Compatibility notes - The three parameters were keyword-only, so no caller can break; keyword callers bind through `**kwargs` identically. - The names leave the runtime signatures (`inspect.signature`), while the v3 overloads keep the typing. - v3 is unchanged: `_pregel_stream_v3`/`_apregel_stream_v3` still default all three to `None` and pass them explicitly, so an override's non-`None` default never applied on v3 and still doesn't. The only behavior change beyond restoring the pre-#7677 passthrough is on v1/v2 with an explicit `None`: since 1.2.0a3 the named-parameter capture dropped it, so an override's non-`None` default was silently applied instead; now the explicit `None` reaches `astream`, as it did before 1.2.0 and as it does on v3. ### Testing New tests in `tests/test_stream_events_v3_kwarg_forwarding.py`: - `interrupt_before` / `interrupt_after` / pre-drained `control` parametrized over v1/v2/v3 (12 of them fail on unpatched `main`). - Sync v3 static interrupts. - Mid-run drain: a node calls `request_drain()`; asserts `GraphDrained` propagates, the caller's own `RunControl` was used, and the checkpoint keeps the pending step. - Recording tests pin, for all three names and on both the async and sync v3 paths, exactly what `(a)stream` receives: supplied values (with `control` identity), explicit `None`, and absence for omitted arguments. Verified by mutation: replacing the implementation with each rejected alternative (unpatched parent; injecting `None` for omitted kwargs on v1/v2; dropping explicit `None` on v1/v2; kwargs-only private helpers) makes the corresponding tests fail. Suites: 246 passed on the four adjacent streaming test files; 906 passed / 1 skipped across every test file calling `(a)stream_events` plus `test_runtime.py`. `ruff check` / `format --check` clean; `ty check langgraph` clean. Not verifiable locally (CI please): Python 3.10 (the new tests carry no version skip), minimum supported langchain-core, and a live langgraph-api server run. ### Release note > Fixed a regression since 1.2.0a3: `Pregel.astream_events` silently dropped > `interrupt_before`, `interrupt_after` and `control` for > `version="v1"`/`"v2"`. Static interrupts and `RunControl` drains now work on > every async version. Server runs using `stream_mode="events"` were affected > and get their static interrupts back.
Low-level orchestration framework for building stateful agents.
Trusted by companies shaping the future of agents – including Klarna, Replit, Elastic, and more – LangGraph is a low-level orchestration framework for building, managing, and deploying long-running, stateful agents.
pip install -U langgraph
Tip
If you're looking to quickly build agents, check out Deep Agents — a higher-level package built on LangGraph for agents that can plan, use subagents, and leverage file systems for complex tasks.
For an equivalent JS/TS library, check out LangGraph.js and the JS docs.
Why use LangGraph?
LangGraph provides low-level supporting infrastructure for any long-running, stateful workflow or agent:
- Durable execution — Build agents that persist through failures and can run for extended periods, automatically resuming from exactly where they left off.
- Human-in-the-loop — Seamlessly incorporate human oversight by inspecting and modifying agent state at any point during execution.
- Comprehensive memory — Create truly stateful agents with both short-term working memory for ongoing reasoning and long-term persistent memory across sessions.
- Debugging with LangSmith — Gain deep visibility into complex agent behavior with visualization tools that trace execution paths, capture state transitions, and provide detailed runtime metrics.
- Production-ready deployment — Deploy sophisticated agent systems confidently with scalable infrastructure designed to handle the unique challenges of stateful, long-running workflows.
Tip
For developing, debugging, and deploying AI agents and LLM applications, see LangSmith.
LangGraph ecosystem
While LangGraph can be used standalone, it also integrates seamlessly with any LangChain product, giving developers a full suite of tools for building agents.
To improve your LLM application development, pair LangGraph with:
- Deep Agents – Build agents that can plan, use subagents, and leverage file systems for complex tasks.
- LangChain – Provides integrations and composable components to streamline LLM application development.
- LangSmith – Helpful for agent evals and observability. Debug poor-performing LLM app runs, evaluate agent trajectories, gain visibility in production, and improve performance over time.
- LangSmith Deployment – Deploy and scale agents effortlessly with a purpose-built deployment platform for long-running, stateful workflows. Discover, reuse, configure, and share agents across teams – and iterate quickly with visual prototyping in LangSmith Studio.
Documentation
- docs.langchain.com – Comprehensive documentation, including conceptual overviews and guides
- reference.langchain.com/python/langgraph – API reference docs for LangGraph packages
- LangGraph Quickstart – Get started building with LangGraph
- Chat LangChain – Chat with the LangChain documentation and get answers to your questions
Discussions: Visit the LangChain Forum to connect with the community and share all of your technical questions, ideas, and feedback.
Additional resources
- Guides – Quick, actionable code snippets for topics such as streaming, adding memory & persistence, and design patterns (e.g. branching, subgraphs, etc.).
- LangChain Academy – Learn the basics of LangGraph in our free, structured course.
- Case studies – Hear how industry leaders use LangGraph to ship AI applications at scale.
- Contributing Guide – Learn how to contribute to LangChain projects and find good first issues.
- Code of Conduct – Our community guidelines and standards for participation.
Acknowledgements
LangGraph is inspired by Pregel and Apache Beam. The public interface draws inspiration from NetworkX. LangGraph is built by LangChain Inc, the creators of LangChain, but can be used without LangChain.