图表定义

Graph

源代码在 libs/langgraph/langgraph/graph/graph.py 中

class Graph:
    def __init__(self) -> None:
        self.nodes: dict[str, NodeSpec] = {}
        self.edges = set[tuple[str, str]]()
        self.branches: defaultdict[str, dict[str, Branch]] = defaultdict(dict)
        self.support_multiple_edges = False
        self.compiled = False

    @property
    def _all_edges(self) -> set[tuple[str, str]]:
        return self.edges

    @overload
    def add_node(
        self,
        node: RunnableLike,
        *,
        metadata: Optional[dict[str, Any]] = None,
    ) -> Self: ...

    @overload
    def add_node(
        self,
        node: str,
        action: RunnableLike,
        *,
        metadata: Optional[dict[str, Any]] = None,
    ) -> Self: ...

    def add_node(
        self,
        node: Union[str, RunnableLike],
        action: Optional[RunnableLike] = None,
        *,
        metadata: Optional[dict[str, Any]] = None,
    ) -> Self:
        if isinstance(node, str):
            for character in (NS_SEP, NS_END):
                if character in node:
                    raise ValueError(
                        f"'{character}' is a reserved character and is not allowed in the node names."
                    )

        if self.compiled:
            logger.warning(
                "Adding a node to a graph that has already been compiled. This will "
                "not be reflected in the compiled graph."
            )
        if not isinstance(node, str):
            action = node
            node = getattr(action, "name", getattr(action, "__name__"))
            if node is None:
                raise ValueError(
                    "Node name must be provided if action is not a function"
                )
        if action is None:
            raise RuntimeError(
                "Expected a function or Runnable action in add_node. Received None."
            )
        if node in self.nodes:
            raise ValueError(f"Node `{node}` already present.")
        if node == END or node == START:
            raise ValueError(f"Node `{node}` is reserved.")

        self.nodes[cast(str, node)] = NodeSpec(
            coerce_to_runnable(action, name=cast(str, node), trace=False), metadata
        )
        return self

    def add_edge(self, start_key: str, end_key: str) -> Self:
        if self.compiled:
            logger.warning(
                "Adding an edge to a graph that has already been compiled. This will "
                "not be reflected in the compiled graph."
            )
        if start_key == END:
            raise ValueError("END cannot be a start node")
        if end_key == START:
            raise ValueError("START cannot be an end node")

        # run this validation only for non-StateGraph graphs
        if not hasattr(self, "channels") and start_key in set(
            start for start, _ in self.edges
        ):
            raise ValueError(
                f"Already found path for node '{start_key}'.\n"
                "For multiple edges, use StateGraph with an Annotated state key."
            )

        self.edges.add((start_key, end_key))
        return self

    def add_conditional_edges(
        self,
        source: str,
        path: Union[
            Callable[..., Union[Hashable, list[Hashable]]],
            Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
            Runnable[Any, Union[Hashable, list[Hashable]]],
        ],
        path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
        then: Optional[str] = None,
    ) -> Self:
        """Add a conditional edge from the starting node to any number of destination nodes.

        Args:
            source (str): The starting node. This conditional edge will run when
                exiting this node.
            path (Union[Callable, Runnable]): The callable that determines the next
                node or nodes. If not specifying `path_map` it should return one or
                more nodes. If it returns END, the graph will stop execution.
            path_map (Optional[dict[Hashable, str]]): Optional mapping of paths to node
                names. If omitted the paths returned by `path` should be node names.
            then (Optional[str]): The name of a node to execute after the nodes
                selected by `path`.

        Returns:
            None

        Note: Without typehints on the `path` function's return value (e.g., `-> Literal["foo", "__end__"]:`)
            or a path_map, the graph visualization assumes the edge could transition to any node in the graph.

        """  # noqa: E501
        if self.compiled:
            logger.warning(
                "Adding an edge to a graph that has already been compiled. This will "
                "not be reflected in the compiled graph."
            )
        # coerce path_map to a dictionary
        try:
            if isinstance(path_map, dict):
                path_map_ = path_map.copy()
            elif isinstance(path_map, list):
                path_map_ = {name: name for name in path_map}
            elif isinstance(path, Runnable):
                path_map_ = None
            elif rtn_type := get_type_hints(path.__call__).get(  # type: ignore[operator]
                "return"
            ) or get_type_hints(path).get("return"):
                if get_origin(rtn_type) is Literal:
                    path_map_ = {name: name for name in get_args(rtn_type)}
                else:
                    path_map_ = None
            else:
                path_map_ = None
        except Exception:
            path_map_ = None
        # find a name for the condition
        path = coerce_to_runnable(path, name=None, trace=True)
        name = path.name or "condition"
        # validate the condition
        if name in self.branches[source]:
            raise ValueError(
                f"Branch with name `{path.name}` already exists for node " f"`{source}`"
            )
        # save it
        self.branches[source][name] = Branch(path, path_map_, then)
        return self

    def set_entry_point(self, key: str) -> Self:
        """Specifies the first node to be called in the graph.

        Equivalent to calling `add_edge(START, key)`.

        Parameters:
            key (str): The key of the node to set as the entry point.

        Returns:
            None
        """
        return self.add_edge(START, key)

    def set_conditional_entry_point(
        self,
        path: Union[
            Callable[..., Union[Hashable, list[Hashable]]],
            Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
            Runnable[Any, Union[Hashable, list[Hashable]]],
        ],
        path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
        then: Optional[str] = None,
    ) -> Self:
        """Sets a conditional entry point in the graph.

        Args:
            path (Union[Callable, Runnable]): The callable that determines the next
                node or nodes. If not specifying `path_map` it should return one or
                more nodes. If it returns END, the graph will stop execution.
            path_map (Optional[dict[str, str]]): Optional mapping of paths to node
                names. If omitted the paths returned by `path` should be node names.
            then (Optional[str]): The name of a node to execute after the nodes
                selected by `path`.

        Returns:
            None
        """
        return self.add_conditional_edges(START, path, path_map, then)

    def set_finish_point(self, key: str) -> Self:
        """Marks a node as a finish point of the graph.

        If the graph reaches this node, it will cease execution.

        Parameters:
            key (str): The key of the node to set as the finish point.

        Returns:
            None
        """
        return self.add_edge(key, END)

    def validate(self, interrupt: Optional[Sequence[str]] = None) -> Self:
        # assemble sources
        all_sources = {src for src, _ in self._all_edges}
        for start, branches in self.branches.items():
            all_sources.add(start)
            for cond, branch in branches.items():
                if branch.then is not None:
                    if branch.ends is not None:
                        for end in branch.ends.values():
                            if end != END:
                                all_sources.add(end)
                    else:
                        for node in self.nodes:
                            if node != start and node != branch.then:
                                all_sources.add(node)
        # validate sources
        for source in all_sources:
            if source not in self.nodes and source != START:
                raise ValueError(f"Found edge starting at unknown node '{source}'")

        # assemble targets
        all_targets = {end for _, end in self._all_edges}
        for start, branches in self.branches.items():
            for cond, branch in branches.items():
                if branch.then is not None:
                    all_targets.add(branch.then)
                if branch.ends is not None:
                    for end in branch.ends.values():
                        if end not in self.nodes and end != END:
                            raise ValueError(
                                f"At '{start}' node, '{cond}' branch found unknown target '{end}'"
                            )
                        all_targets.add(end)
                else:
                    all_targets.add(END)
                    for node in self.nodes:
                        if node != start and node != branch.then:
                            all_targets.add(node)
        # validate targets
        for node in self.nodes:
            if node not in all_targets:
                raise ValueError(f"Node `{node}` is not reachable")
        for target in all_targets:
            if target not in self.nodes and target != END:
                raise ValueError(f"Found edge ending at unknown node `{target}`")
        # validate interrupts
        if interrupt:
            for node in interrupt:
                if node not in self.nodes:
                    raise ValueError(f"Interrupt node `{node}` not found")

        self.compiled = True
        return self

    def compile(
        self,
        checkpointer: Checkpointer = None,
        interrupt_before: Optional[Union[All, list[str]]] = None,
        interrupt_after: Optional[Union[All, list[str]]] = None,
        debug: bool = False,
    ) -> "CompiledGraph":
        # assign default values
        interrupt_before = interrupt_before or []
        interrupt_after = interrupt_after or []

        # validate the graph
        self.validate(
            interrupt=(
                (interrupt_before if interrupt_before != "*" else []) + interrupt_after
                if interrupt_after != "*"
                else []
            )
        )

        # create empty compiled graph
        compiled = CompiledGraph(
            builder=self,
            nodes={},
            channels={START: EphemeralValue(Any), END: EphemeralValue(Any)},
            input_channels=START,
            output_channels=END,
            stream_mode="values",
            stream_channels=[],
            checkpointer=checkpointer,
            interrupt_before_nodes=interrupt_before,
            interrupt_after_nodes=interrupt_after,
            auto_validate=False,
            debug=debug,
        )

        # attach nodes, edges, and branches
        for key, node in self.nodes.items():
            compiled.attach_node(key, node)

        for start, end in self.edges:
            compiled.attach_edge(start, end)

        for start, branches in self.branches.items():
            for name, branch in branches.items():
                compiled.attach_branch(start, name, branch)

        # validate the compiled graph
        return compiled.validate()

add_conditional_edges(source: str, path: Union[Callable[..., Union[Hashable, list[Hashable]]], Callable[..., Awaitable[Union[Hashable, list[Hashable]]]], Runnable[Any, Union[Hashable, list[Hashable]]]], path_map: Optional[Union[dict[Hashable, str], list[str]]] = None, then: Optional[str] = None) -> Self

从起始节点添加一个条件边到任意数量的目标节点。

参数

• source (str) –

  起始节点。当退出该节点时，将运行此条件边。

• path (Union[Callable, Runnable]) –

  确定下一个节点或节点的函数。如果未指定 path_map，则应返回一个或多个节点。如果返回 END，则图形将停止执行。

• path_map (Optional[dict[Hashable, str]], 默认： None ) –

  路径到节点名称的可选映射。如果省略，则 path 返回的路径应为节点名称。

• then (Optional[str], 默认： None ) –

  由 path 选择的节点执行完毕后，将要执行的节点的名称。

返回值

• Self –

  无

在 path 函数的返回值上没有类型提示（例如， -> Literal["foo", "__end__"]:）

或者没有 path_map，图形可视化假定边可以转换到图形中的任何节点。

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def add_conditional_edges(
    self,
    source: str,
    path: Union[
        Callable[..., Union[Hashable, list[Hashable]]],
        Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
        Runnable[Any, Union[Hashable, list[Hashable]]],
    ],
    path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
    then: Optional[str] = None,
) -> Self:
    """Add a conditional edge from the starting node to any number of destination nodes.

    Args:
        source (str): The starting node. This conditional edge will run when
            exiting this node.
        path (Union[Callable, Runnable]): The callable that determines the next
            node or nodes. If not specifying `path_map` it should return one or
            more nodes. If it returns END, the graph will stop execution.
        path_map (Optional[dict[Hashable, str]]): Optional mapping of paths to node
            names. If omitted the paths returned by `path` should be node names.
        then (Optional[str]): The name of a node to execute after the nodes
            selected by `path`.

    Returns:
        None

    Note: Without typehints on the `path` function's return value (e.g., `-> Literal["foo", "__end__"]:`)
        or a path_map, the graph visualization assumes the edge could transition to any node in the graph.

    """  # noqa: E501
    if self.compiled:
        logger.warning(
            "Adding an edge to a graph that has already been compiled. This will "
            "not be reflected in the compiled graph."
        )
    # coerce path_map to a dictionary
    try:
        if isinstance(path_map, dict):
            path_map_ = path_map.copy()
        elif isinstance(path_map, list):
            path_map_ = {name: name for name in path_map}
        elif isinstance(path, Runnable):
            path_map_ = None
        elif rtn_type := get_type_hints(path.__call__).get(  # type: ignore[operator]
            "return"
        ) or get_type_hints(path).get("return"):
            if get_origin(rtn_type) is Literal:
                path_map_ = {name: name for name in get_args(rtn_type)}
            else:
                path_map_ = None
        else:
            path_map_ = None
    except Exception:
        path_map_ = None
    # find a name for the condition
    path = coerce_to_runnable(path, name=None, trace=True)
    name = path.name or "condition"
    # validate the condition
    if name in self.branches[source]:
        raise ValueError(
            f"Branch with name `{path.name}` already exists for node " f"`{source}`"
        )
    # save it
    self.branches[source][name] = Branch(path, path_map_, then)
    return self

set_entry_point(key: str) -> Self

指定要调用的第一个节点。

相当于调用 add_edge(START, key)。

参数

• key (str) –

  要设置为入口点的节点的键。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_entry_point(self, key: str) -> Self:
    """Specifies the first node to be called in the graph.

    Equivalent to calling `add_edge(START, key)`.

    Parameters:
        key (str): The key of the node to set as the entry point.

    Returns:
        None
    """
    return self.add_edge(START, key)

set_conditional_entry_point(path: Union[Callable[..., Union[Hashable, list[Hashable]]], Callable[..., Awaitable[Union[Hashable, list[Hashable]]]], Runnable[Any, Union[Hashable, list[Hashable]]]], path_map: Optional[Union[dict[Hashable, str], list[str]]] = None, then: Optional[str] = None) -> Self

在图形中设置条件入口点。

参数

• path (Union[Callable, Runnable]) –

  确定下一个节点或节点的函数。如果未指定 path_map，则应返回一个或多个节点。如果返回 END，则图形将停止执行。

• path_map (Optional[dict[str, str]], 默认： None ) –

  路径到节点名称的可选映射。如果省略，则 path 返回的路径应为节点名称。

• then (Optional[str], 默认： None ) –

  由 path 选择的节点执行完毕后，将要执行的节点的名称。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_conditional_entry_point(
    self,
    path: Union[
        Callable[..., Union[Hashable, list[Hashable]]],
        Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
        Runnable[Any, Union[Hashable, list[Hashable]]],
    ],
    path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
    then: Optional[str] = None,
) -> Self:
    """Sets a conditional entry point in the graph.

    Args:
        path (Union[Callable, Runnable]): The callable that determines the next
            node or nodes. If not specifying `path_map` it should return one or
            more nodes. If it returns END, the graph will stop execution.
        path_map (Optional[dict[str, str]]): Optional mapping of paths to node
            names. If omitted the paths returned by `path` should be node names.
        then (Optional[str]): The name of a node to execute after the nodes
            selected by `path`.

    Returns:
        None
    """
    return self.add_conditional_edges(START, path, path_map, then)

set_finish_point(key: str) -> Self

将节点标记为图形的结束点。

如果图形到达此节点，则将停止执行。

参数

• key (str) –

  要设置为结束点的节点的键。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_finish_point(self, key: str) -> Self:
    """Marks a node as a finish point of the graph.

    If the graph reaches this node, it will cease execution.

    Parameters:
        key (str): The key of the node to set as the finish point.

    Returns:
        None
    """
    return self.add_edge(key, END)

CompiledGraph

基类： Pregel

源代码在 libs/langgraph/langgraph/graph/graph.py 中

class CompiledGraph(Pregel):
    builder: Graph

    def __init__(self, *, builder: Graph, **kwargs: Any) -> None:
        super().__init__(**kwargs)
        self.builder = builder

    def attach_node(self, key: str, node: NodeSpec) -> None:
        self.channels[key] = EphemeralValue(Any)
        self.nodes[key] = (
            PregelNode(channels=[], triggers=[], metadata=node.metadata)
            | node.runnable
            | ChannelWrite([ChannelWriteEntry(key)], tags=[TAG_HIDDEN])
        )
        cast(list[str], self.stream_channels).append(key)

    def attach_edge(self, start: str, end: str) -> None:
        if end == END:
            # publish to end channel
            self.nodes[start].writers.append(
                ChannelWrite([ChannelWriteEntry(END)], tags=[TAG_HIDDEN])
            )
        else:
            # subscribe to start channel
            self.nodes[end].triggers.append(start)
            cast(list[str], self.nodes[end].channels).append(start)

    def attach_branch(self, start: str, name: str, branch: Branch) -> None:
        def branch_writer(
            packets: Sequence[Union[str, Send]], config: RunnableConfig
        ) -> Optional[ChannelWrite]:
            writes = [
                (
                    ChannelWriteEntry(f"branch:{start}:{name}:{p}" if p != END else END)
                    if not isinstance(p, Send)
                    else p
                )
                for p in packets
            ]
            return ChannelWrite(
                cast(Sequence[Union[ChannelWriteEntry, Send]], writes),
                tags=[TAG_HIDDEN],
            )

        # add hidden start node
        if start == START and start not in self.nodes:
            self.nodes[start] = Channel.subscribe_to(START, tags=[TAG_HIDDEN])

        # attach branch writer
        self.nodes[start] |= branch.run(branch_writer)

        # attach branch readers
        ends = branch.ends.values() if branch.ends else [node for node in self.nodes]
        for end in ends:
            if end != END:
                channel_name = f"branch:{start}:{name}:{end}"
                self.channels[channel_name] = EphemeralValue(Any)
                self.nodes[end].triggers.append(channel_name)
                cast(list[str], self.nodes[end].channels).append(channel_name)

    def get_graph(
        self,
        config: Optional[RunnableConfig] = None,
        *,
        xray: Union[int, bool] = False,
    ) -> DrawableGraph:
        """Returns a drawable representation of the computation graph."""
        graph = DrawableGraph()
        start_nodes: dict[str, DrawableNode] = {
            START: graph.add_node(self.get_input_schema(config), START)
        }
        end_nodes: dict[str, DrawableNode] = {}
        if xray:
            subgraphs = {
                k: v for k, v in self.get_subgraphs() if isinstance(v, CompiledGraph)
            }
        else:
            subgraphs = {}

        def add_edge(
            start: str,
            end: str,
            label: Optional[Hashable] = None,
            conditional: bool = False,
        ) -> None:
            if end == END and END not in end_nodes:
                end_nodes[END] = graph.add_node(self.get_output_schema(config), END)
            return graph.add_edge(
                start_nodes[start],
                end_nodes[end],
                str(label) if label is not None else None,
                conditional,
            )

        for key, n in self.builder.nodes.items():
            node = n.runnable
            metadata = n.metadata or {}
            if key in self.interrupt_before_nodes and key in self.interrupt_after_nodes:
                metadata["__interrupt"] = "before,after"
            elif key in self.interrupt_before_nodes:
                metadata["__interrupt"] = "before"
            elif key in self.interrupt_after_nodes:
                metadata["__interrupt"] = "after"
            if xray:
                subgraph = (
                    subgraphs[key].get_graph(
                        config=config,
                        xray=xray - 1
                        if isinstance(xray, int)
                        and not isinstance(xray, bool)
                        and xray > 0
                        else xray,
                    )
                    if key in subgraphs
                    else node.get_graph(config=config)
                )
                subgraph.trim_first_node()
                subgraph.trim_last_node()
                if len(subgraph.nodes) > 1:
                    e, s = graph.extend(subgraph, prefix=key)
                    if e is None:
                        raise ValueError(
                            f"Could not extend subgraph '{key}' due to missing entrypoint"
                        )
                    if s is not None:
                        start_nodes[key] = s
                    end_nodes[key] = e
                else:
                    nn = graph.add_node(node, key, metadata=metadata or None)
                    start_nodes[key] = nn
                    end_nodes[key] = nn
            else:
                nn = graph.add_node(node, key, metadata=metadata or None)
                start_nodes[key] = nn
                end_nodes[key] = nn
        for start, end in sorted(self.builder._all_edges):
            add_edge(start, end)
        for start, branches in self.builder.branches.items():
            default_ends = {
                **{k: k for k in self.builder.nodes if k != start},
                END: END,
            }
            for _, branch in branches.items():
                if branch.ends is not None:
                    ends = branch.ends
                elif branch.then is not None:
                    ends = {k: k for k in default_ends if k not in (END, branch.then)}
                else:
                    ends = cast(dict[Hashable, str], default_ends)
                for label, end in ends.items():
                    add_edge(
                        start,
                        end,
                        label if label != end else None,
                        conditional=True,
                    )
                    if branch.then is not None:
                        add_edge(end, branch.then)

        return graph

stream_mode: StreamMode = stream_mode class-attribute instance-attribute

流式传输输出的模式，默认为“values”。

stream_channels: Optional[Union[str, Sequence[str]]] = stream_channels class-attribute instance-attribute

要流式传输的通道，默认为所有不在保留通道中的通道。

step_timeout: Optional[float] = step_timeout class-attribute instance-attribute

等待步骤完成的最长时间，以秒为单位。默认为 None。

debug: bool = debug if debug is not None else get_debug() instance-attribute

是否在执行期间打印调试信息。默认为 False。

checkpointer: Checkpointer = checkpointer class-attribute instance-attribute

用于保存和加载图表状态的检查点。默认为 None。

store: Optional[BaseStore] = store class-attribute instance-attribute

用于 SharedValues 的内存存储。默认为 None。

retry_policy: Optional[RetryPolicy] = retry_policy class-attribute instance-attribute

运行任务时使用的重试策略。设置为 None 以禁用。

get_state(config: RunnableConfig, *, subgraphs: bool = False) -> StateSnapshot

获取图的当前状态。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def get_state(
    self, config: RunnableConfig, *, subgraphs: bool = False
) -> StateSnapshot:
    """Get the current state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return pregel.get_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                subgraphs=subgraphs,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(self.config, config) if self.config else config
    saved = checkpointer.get_tuple(config)
    return self._prepare_state_snapshot(
        config, saved, recurse=checkpointer if subgraphs else None
    )

aget_state(config: RunnableConfig, *, subgraphs: bool = False) -> StateSnapshot async

获取图的当前状态。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def aget_state(
    self, config: RunnableConfig, *, subgraphs: bool = False
) -> StateSnapshot:
    """Get the current state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        async for _, pregel in self.aget_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return await pregel.aget_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                subgraphs=subgraphs,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(self.config, config) if self.config else config
    saved = await checkpointer.aget_tuple(config)
    return await self._aprepare_state_snapshot(
        config, saved, recurse=checkpointer if subgraphs else None
    )

get_state_history(config: RunnableConfig, *, filter: Optional[Dict[str, Any]] = None, before: Optional[RunnableConfig] = None, limit: Optional[int] = None) -> Iterator[StateSnapshot]

获取图的状态历史记录。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def get_state_history(
    self,
    config: RunnableConfig,
    *,
    filter: Optional[Dict[str, Any]] = None,
    before: Optional[RunnableConfig] = None,
    limit: Optional[int] = None,
) -> Iterator[StateSnapshot]:
    """Get the history of the state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            yield from pregel.get_state_history(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                filter=filter,
                before=before,
                limit=limit,
            )
            return
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(
        self.config,
        config,
        {CONF: {CONFIG_KEY_CHECKPOINT_NS: checkpoint_ns}},
    )
    # eagerly consume list() to avoid holding up the db cursor
    for checkpoint_tuple in list(
        checkpointer.list(config, before=before, limit=limit, filter=filter)
    ):
        yield self._prepare_state_snapshot(
            checkpoint_tuple.config, checkpoint_tuple
        )

aget_state_history(config: RunnableConfig, *, filter: Optional[Dict[str, Any]] = None, before: Optional[RunnableConfig] = None, limit: Optional[int] = None) -> AsyncIterator[StateSnapshot] async

获取图的状态历史记录。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def aget_state_history(
    self,
    config: RunnableConfig,
    *,
    filter: Optional[Dict[str, Any]] = None,
    before: Optional[RunnableConfig] = None,
    limit: Optional[int] = None,
) -> AsyncIterator[StateSnapshot]:
    """Get the history of the state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        async for _, pregel in self.aget_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            async for state in pregel.aget_state_history(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                filter=filter,
                before=before,
                limit=limit,
            ):
                yield state
            return
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(
        self.config,
        config,
        {CONF: {CONFIG_KEY_CHECKPOINT_NS: checkpoint_ns}},
    )
    # eagerly consume list() to avoid holding up the db cursor
    for checkpoint_tuple in [
        c
        async for c in checkpointer.alist(
            config, before=before, limit=limit, filter=filter
        )
    ]:
        yield await self._aprepare_state_snapshot(
            checkpoint_tuple.config, checkpoint_tuple
        )

update_state(config: RunnableConfig, values: Optional[Union[dict[str, Any], Any]], as_node: Optional[str] = None) -> RunnableConfig

使用给定的值更新图的状态，就好像它们来自节点 as_node。如果未提供 as_node，它将设置为更新状态的最后一个节点，如果不明确。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def update_state(
    self,
    config: RunnableConfig,
    values: Optional[Union[dict[str, Any], Any]],
    as_node: Optional[str] = None,
) -> RunnableConfig:
    """Update the state of the graph with the given values, as if they came from
    node `as_node`. If `as_node` is not provided, it will be set to the last node
    that updated the state, if not ambiguous.
    """
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    # delegate to subgraph
    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return pregel.update_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                values,
                as_node,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    # get last checkpoint
    config = merge_configs(self.config, config) if self.config else config
    saved = checkpointer.get_tuple(config)
    checkpoint = copy_checkpoint(saved.checkpoint) if saved else empty_checkpoint()
    checkpoint_previous_versions = (
        saved.checkpoint["channel_versions"].copy() if saved else {}
    )
    step = saved.metadata.get("step", -1) if saved else -1
    # merge configurable fields with previous checkpoint config
    checkpoint_config = patch_configurable(
        config,
        {CONFIG_KEY_CHECKPOINT_NS: config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")},
    )
    if saved:
        checkpoint_config = patch_configurable(config, saved.config[CONF])
    # find last node that updated the state, if not provided
    if values is None and as_node is None:
        next_config = checkpointer.put(
            checkpoint_config,
            create_checkpoint(checkpoint, None, step),
            {
                "source": "update",
                "step": step + 1,
                "writes": {},
                "parents": saved.metadata.get("parents", {}) if saved else {},
            },
            {},
        )
        return patch_checkpoint_map(next_config, saved.metadata if saved else None)
    elif as_node is None and not any(
        v for vv in checkpoint["versions_seen"].values() for v in vv.values()
    ):
        if (
            isinstance(self.input_channels, str)
            and self.input_channels in self.nodes
        ):
            as_node = self.input_channels
    elif as_node is None:
        last_seen_by_node = sorted(
            (v, n)
            for n, seen in checkpoint["versions_seen"].items()
            if n in self.nodes
            for v in seen.values()
        )
        # if two nodes updated the state at the same time, it's ambiguous
        if last_seen_by_node:
            if len(last_seen_by_node) == 1:
                as_node = last_seen_by_node[0][1]
            elif last_seen_by_node[-1][0] != last_seen_by_node[-2][0]:
                as_node = last_seen_by_node[-1][1]
    if as_node is None:
        raise InvalidUpdateError("Ambiguous update, specify as_node")
    if as_node not in self.nodes:
        raise InvalidUpdateError(f"Node {as_node} does not exist")
    # update channels
    with ChannelsManager(self.channels, checkpoint, config) as (
        channels,
        managed,
    ):
        # create task to run all writers of the chosen node
        writers = self.nodes[as_node].flat_writers
        if not writers:
            raise InvalidUpdateError(f"Node {as_node} has no writers")
        writes: deque[tuple[str, Any]] = deque()
        task = PregelTaskWrites(as_node, writes, [INTERRUPT])
        task_id = str(uuid5(UUID(checkpoint["id"]), INTERRUPT))
        run = RunnableSequence(*writers) if len(writers) > 1 else writers[0]
        # execute task
        run.invoke(
            values,
            patch_config(
                config,
                run_name=self.name + "UpdateState",
                configurable={
                    # deque.extend is thread-safe
                    CONFIG_KEY_SEND: partial(
                        local_write,
                        writes.extend,
                        self.nodes.keys(),
                    ),
                    CONFIG_KEY_READ: partial(
                        local_read,
                        step + 1,
                        checkpoint,
                        channels,
                        managed,
                        task,
                        config,
                    ),
                },
            ),
        )
        # save task writes
        if saved:
            checkpointer.put_writes(checkpoint_config, task.writes, task_id)
        # apply to checkpoint and save
        assert not apply_writes(
            checkpoint, channels, [task], checkpointer.get_next_version
        ), "Can't write to SharedValues from update_state"
        checkpoint = create_checkpoint(checkpoint, channels, step + 1)
        next_config = checkpointer.put(
            checkpoint_config,
            checkpoint,
            {
                "source": "update",
                "step": step + 1,
                "writes": {as_node: values},
                "parents": saved.metadata.get("parents", {}) if saved else {},
            },
            get_new_channel_versions(
                checkpoint_previous_versions, checkpoint["channel_versions"]
            ),
        )
        return patch_checkpoint_map(next_config, saved.metadata if saved else None)

stream(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None, output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, subgraphs: bool = False) -> Iterator[Union[dict[str, Any], Any]]

为单个输入流式传输图步骤。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入。

• config (Optional[RunnableConfig], default: None ) –

  用于运行的配置。

• stream_mode (Optional[Union[StreamMode, list[StreamMode]]], default: None ) –

  流式传输输出的模式，默认为 self.stream_mode。选项包括“values”、“updates”和“debug”。values：为每一步发出状态的当前值。updates：仅为每一步发出状态的更新。输出是一个字典，节点名称作为键，更新后的值作为值。debug：为每一步发出调试事件。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  要流式传输的键，默认为所有非上下文通道。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之前中断的节点，默认为图中的所有节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之后中断的节点，默认为图中的所有节点。

• debug (Optional[bool], default: None ) –

  是否在执行期间打印调试信息，默认为 False。

• subgraphs (bool, default: False ) –

  是否流式传输子图，默认为 False。

Yields

• Union[dict[str, Any], Any] –

  图中每一步的输出。输出形状取决于 stream_mode。

Examples

使用不同的 stream 模式和一个图

>>> import operator
>>> from typing_extensions import Annotated, TypedDict
>>> from langgraph.graph import StateGraph
>>> from langgraph.constants import START
...
>>> class State(TypedDict):
...     alist: Annotated[list, operator.add]
...     another_list: Annotated[list, operator.add]
...
>>> builder = StateGraph(State)
>>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
>>> builder.add_node("b", lambda _state: {"alist": ["there"]})
>>> builder.add_edge("a", "b")
>>> builder.add_edge(START, "a")
>>> graph = builder.compile()

使用 stream_mode="values"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
...     print(event)
{'alist': ['Ex for stream_mode="values"'], 'another_list': []}
{'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
{'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}

使用 stream_mode="updates"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
...     print(event)
{'a': {'another_list': ['hi']}}
{'b': {'alist': ['there']}}

使用 stream_mode="debug"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
...     print(event)
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def stream(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None,
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    subgraphs: bool = False,
) -> Iterator[Union[dict[str, Any], Any]]:
    """Stream graph steps for a single input.

    Args:
        input: The input to the graph.
        config: The configuration to use for the run.
        stream_mode: The mode to stream output, defaults to self.stream_mode.
            Options are 'values', 'updates', and 'debug'.
            values: Emit the current values of the state for each step.
            updates: Emit only the updates to the state for each step.
                Output is a dict with the node name as key and the updated values as value.
            debug: Emit debug events for each step.
        output_keys: The keys to stream, defaults to all non-context channels.
        interrupt_before: Nodes to interrupt before, defaults to all nodes in the graph.
        interrupt_after: Nodes to interrupt after, defaults to all nodes in the graph.
        debug: Whether to print debug information during execution, defaults to False.
        subgraphs: Whether to stream subgraphs, defaults to False.

    Yields:
        The output of each step in the graph. The output shape depends on the stream_mode.

    Examples:
        Using different stream modes with a graph:
        ```pycon
        >>> import operator
        >>> from typing_extensions import Annotated, TypedDict
        >>> from langgraph.graph import StateGraph
        >>> from langgraph.constants import START
        ...
        >>> class State(TypedDict):
        ...     alist: Annotated[list, operator.add]
        ...     another_list: Annotated[list, operator.add]
        ...
        >>> builder = StateGraph(State)
        >>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
        >>> builder.add_node("b", lambda _state: {"alist": ["there"]})
        >>> builder.add_edge("a", "b")
        >>> builder.add_edge(START, "a")
        >>> graph = builder.compile()
        ```
        With stream_mode="values":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
        ...     print(event)
        {'alist': ['Ex for stream_mode="values"'], 'another_list': []}
        {'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
        {'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}
        ```
        With stream_mode="updates":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
        ...     print(event)
        {'a': {'another_list': ['hi']}}
        {'b': {'alist': ['there']}}
        ```
        With stream_mode="debug":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
        ...     print(event)
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}
        ```
    """

    stream = SyncQueue()

    def output() -> Iterator:
        while True:
            try:
                ns, mode, payload = stream.get(block=False)
            except queue.Empty:
                break
            if subgraphs and isinstance(stream_mode, list):
                yield (ns, mode, payload)
            elif isinstance(stream_mode, list):
                yield (mode, payload)
            elif subgraphs:
                yield (ns, payload)
            else:
                yield payload

    config = ensure_config(self.config, config)
    callback_manager = get_callback_manager_for_config(config)
    run_manager = callback_manager.on_chain_start(
        None,
        input,
        name=config.get("run_name", self.get_name()),
        run_id=config.get("run_id"),
    )
    try:
        # assign defaults
        (
            debug,
            stream_modes,
            output_keys,
            interrupt_before_,
            interrupt_after_,
            checkpointer,
            store,
        ) = self._defaults(
            config,
            stream_mode=stream_mode,
            output_keys=output_keys,
            interrupt_before=interrupt_before,
            interrupt_after=interrupt_after,
            debug=debug,
        )
        # set up messages stream mode
        if "messages" in stream_modes:
            run_manager.inheritable_handlers.append(
                StreamMessagesHandler(stream.put)
            )
        # set up custom stream mode
        if "custom" in stream_modes:
            config[CONF][CONFIG_KEY_STREAM_WRITER] = lambda c: stream.put(
                ((), "custom", c)
            )
        with SyncPregelLoop(
            input,
            stream=StreamProtocol(stream.put, stream_modes),
            config=config,
            store=store,
            checkpointer=checkpointer,
            nodes=self.nodes,
            specs=self.channels,
            output_keys=output_keys,
            stream_keys=self.stream_channels_asis,
            debug=debug,
        ) as loop:
            # create runner
            runner = PregelRunner(
                submit=loop.submit,
                put_writes=loop.put_writes,
            )
            # enable subgraph streaming
            if subgraphs:
                loop.config[CONF][CONFIG_KEY_STREAM] = loop.stream
            # enable concurrent streaming
            if subgraphs or "messages" in stream_modes or "custom" in stream_modes:
                # we are careful to have a single waiter live at any one time
                # because on exit we increment semaphore count by exactly 1
                waiter: Optional[concurrent.futures.Future] = None
                # because sync futures cannot be cancelled, we instead
                # release the stream semaphore on exit, which will cause
                # a pending waiter to return immediately
                loop.stack.callback(stream._count.release)

                def get_waiter() -> concurrent.futures.Future[None]:
                    nonlocal waiter
                    if waiter is None or waiter.done():
                        waiter = loop.submit(stream.wait)
                        return waiter
                    else:
                        return waiter
            else:
                get_waiter = None  # type: ignore[assignment]
            # Similarly to Bulk Synchronous Parallel / Pregel model
            # computation proceeds in steps, while there are channel updates
            # channel updates from step N are only visible in step N+1
            # channels are guaranteed to be immutable for the duration of the step,
            # with channel updates applied only at the transition between steps
            while loop.tick(
                input_keys=self.input_channels,
                interrupt_before=interrupt_before_,
                interrupt_after=interrupt_after_,
                manager=run_manager,
            ):
                for _ in runner.tick(
                    loop.tasks.values(),
                    timeout=self.step_timeout,
                    retry_policy=self.retry_policy,
                    get_waiter=get_waiter,
                ):
                    # emit output
                    yield from output()
        # emit output
        yield from output()
        # handle exit
        if loop.status == "out_of_steps":
            raise GraphRecursionError(
                f"Recursion limit of {config['recursion_limit']} reached "
                "without hitting a stop condition. You can increase the "
                "limit by setting the `recursion_limit` config key."
            )
        # set final channel values as run output
        run_manager.on_chain_end(loop.output)
    except BaseException as e:
        run_manager.on_chain_error(e)
        raise

astream(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None, output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, subgraphs: bool = False) -> AsyncIterator[Union[dict[str, Any], Any]] async

为单个输入流式传输图步骤。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入。

• config (Optional[RunnableConfig], default: None ) –

  用于运行的配置。

• stream_mode (Optional[Union[StreamMode, list[StreamMode]]], default: None ) –

  流式传输输出的模式，默认为 self.stream_mode。选项包括“values”、“updates”和“debug”。values：为每一步发出状态的当前值。updates：仅为每一步发出状态的更新。输出是一个字典，节点名称作为键，更新后的值作为值。debug：为每一步发出调试事件。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  要流式传输的键，默认为所有非上下文通道。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之前中断的节点，默认为图中的所有节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之后中断的节点，默认为图中的所有节点。

• debug (Optional[bool], default: None ) –

  是否在执行期间打印调试信息，默认为 False。

• subgraphs (bool, default: False ) –

  是否流式传输子图，默认为 False。

Yields

• AsyncIterator[Union[dict[str, Any], Any]] –

  图中每一步的输出。输出形状取决于 stream_mode。

Examples

使用不同的 stream 模式和一个图

>>> import operator
>>> from typing_extensions import Annotated, TypedDict
>>> from langgraph.graph import StateGraph
>>> from langgraph.constants import START
...
>>> class State(TypedDict):
...     alist: Annotated[list, operator.add]
...     another_list: Annotated[list, operator.add]
...
>>> builder = StateGraph(State)
>>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
>>> builder.add_node("b", lambda _state: {"alist": ["there"]})
>>> builder.add_edge("a", "b")
>>> builder.add_edge(START, "a")
>>> graph = builder.compile()

使用 stream_mode="values"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
...     print(event)
{'alist': ['Ex for stream_mode="values"'], 'another_list': []}
{'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
{'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}

使用 stream_mode="updates"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
...     print(event)
{'a': {'another_list': ['hi']}}
{'b': {'alist': ['there']}}

使用 stream_mode="debug"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
...     print(event)
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def astream(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None,
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    subgraphs: bool = False,
) -> AsyncIterator[Union[dict[str, Any], Any]]:
    """Stream graph steps for a single input.

    Args:
        input: The input to the graph.
        config: The configuration to use for the run.
        stream_mode: The mode to stream output, defaults to self.stream_mode.
            Options are 'values', 'updates', and 'debug'.
            values: Emit the current values of the state for each step.
            updates: Emit only the updates to the state for each step.
                Output is a dict with the node name as key and the updated values as value.
            debug: Emit debug events for each step.
        output_keys: The keys to stream, defaults to all non-context channels.
        interrupt_before: Nodes to interrupt before, defaults to all nodes in the graph.
        interrupt_after: Nodes to interrupt after, defaults to all nodes in the graph.
        debug: Whether to print debug information during execution, defaults to False.
        subgraphs: Whether to stream subgraphs, defaults to False.

    Yields:
        The output of each step in the graph. The output shape depends on the stream_mode.

    Examples:
        Using different stream modes with a graph:
        ```pycon
        >>> import operator
        >>> from typing_extensions import Annotated, TypedDict
        >>> from langgraph.graph import StateGraph
        >>> from langgraph.constants import START
        ...
        >>> class State(TypedDict):
        ...     alist: Annotated[list, operator.add]
        ...     another_list: Annotated[list, operator.add]
        ...
        >>> builder = StateGraph(State)
        >>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
        >>> builder.add_node("b", lambda _state: {"alist": ["there"]})
        >>> builder.add_edge("a", "b")
        >>> builder.add_edge(START, "a")
        >>> graph = builder.compile()
        ```
        With stream_mode="values":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
        ...     print(event)
        {'alist': ['Ex for stream_mode="values"'], 'another_list': []}
        {'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
        {'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}
        ```
        With stream_mode="updates":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
        ...     print(event)
        {'a': {'another_list': ['hi']}}
        {'b': {'alist': ['there']}}
        ```
        With stream_mode="debug":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
        ...     print(event)
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}
        ```
    """

    stream = AsyncQueue()
    aioloop = asyncio.get_running_loop()

    def output() -> Iterator:
        while True:
            try:
                ns, mode, payload = stream.get_nowait()
            except asyncio.QueueEmpty:
                break
            if subgraphs and isinstance(stream_mode, list):
                yield (ns, mode, payload)
            elif isinstance(stream_mode, list):
                yield (mode, payload)
            elif subgraphs:
                yield (ns, payload)
            else:
                yield payload

    config = ensure_config(self.config, config)
    callback_manager = get_async_callback_manager_for_config(config)
    run_manager = await callback_manager.on_chain_start(
        None,
        input,
        name=config.get("run_name", self.get_name()),
        run_id=config.get("run_id"),
    )
    # if running from astream_log() run each proc with streaming
    do_stream = next(
        (
            cast(_StreamingCallbackHandler, h)
            for h in run_manager.handlers
            if isinstance(h, _StreamingCallbackHandler)
        ),
        None,
    )
    try:
        # assign defaults
        (
            debug,
            stream_modes,
            output_keys,
            interrupt_before_,
            interrupt_after_,
            checkpointer,
            store,
        ) = self._defaults(
            config,
            stream_mode=stream_mode,
            output_keys=output_keys,
            interrupt_before=interrupt_before,
            interrupt_after=interrupt_after,
            debug=debug,
        )
        # set up messages stream mode
        if "messages" in stream_modes:
            run_manager.inheritable_handlers.append(
                StreamMessagesHandler(stream.put_nowait)
            )
        # set up custom stream mode
        if "custom" in stream_modes:
            config[CONF][CONFIG_KEY_STREAM_WRITER] = lambda c: stream.put_nowait(
                ((), "custom", c)
            )
        async with AsyncPregelLoop(
            input,
            stream=StreamProtocol(stream.put_nowait, stream_modes),
            config=config,
            store=store,
            checkpointer=checkpointer,
            nodes=self.nodes,
            specs=self.channels,
            output_keys=output_keys,
            stream_keys=self.stream_channels_asis,
        ) as loop:
            # create runner
            runner = PregelRunner(
                submit=loop.submit,
                put_writes=loop.put_writes,
                use_astream=do_stream is not None,
            )
            # enable subgraph streaming
            if subgraphs:
                loop.config[CONF][CONFIG_KEY_STREAM] = loop.stream
            # enable concurrent streaming
            if subgraphs or "messages" in stream_modes or "custom" in stream_modes:

                def get_waiter() -> asyncio.Task[None]:
                    return aioloop.create_task(stream.wait())
            else:
                get_waiter = None  # type: ignore[assignment]
            # Similarly to Bulk Synchronous Parallel / Pregel model
            # computation proceeds in steps, while there are channel updates
            # channel updates from step N are only visible in step N+1
            # channels are guaranteed to be immutable for the duration of the step,
            # with channel updates applied only at the transition between steps
            while loop.tick(
                input_keys=self.input_channels,
                interrupt_before=interrupt_before_,
                interrupt_after=interrupt_after_,
                manager=run_manager,
            ):
                async for _ in runner.atick(
                    loop.tasks.values(),
                    timeout=self.step_timeout,
                    retry_policy=self.retry_policy,
                    get_waiter=get_waiter,
                ):
                    # emit output
                    for o in output():
                        yield o
        # emit output
        for o in output():
            yield o
        # handle exit
        if loop.status == "out_of_steps":
            raise GraphRecursionError(
                f"Recursion limit of {config['recursion_limit']} reached "
                "without hitting a stop condition. You can increase the "
                "limit by setting the `recursion_limit` config key."
            )
        # set final channel values as run output
        await run_manager.on_chain_end(loop.output)
    except BaseException as e:
        await asyncio.shield(run_manager.on_chain_error(e))
        raise

invoke(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: StreamMode = 'values', output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, **kwargs: Any) -> Union[dict[str, Any], Any]

运行图，使用单个输入和配置。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入数据。它可以是字典或任何其他类型。

• config (Optional[RunnableConfig], default: None ) –

  可选。图运行的配置。

• stream_mode (StreamMode, default: 'values' ) –

  Optional[str]。图运行的流模式。默认值为 "values"。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  可选。要从图运行中检索的输出键。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。在中断图运行之前的节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。在中断图运行之后的节点。

• debug (Optional[bool], default: None ) –

  可选。启用图运行的调试模式。

• **kwargs (Any, default: {} ) –

  传递给图运行的附加关键字参数。

返回值

• Union[dict[str, Any], Any] –

  图运行的输出。如果 stream_mode 为 "values"，则返回最新的输出。

• Union[dict[str, Any], Any] –

  如果 stream_mode 不是 "values"，则返回输出块的列表。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def invoke(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: StreamMode = "values",
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    **kwargs: Any,
) -> Union[dict[str, Any], Any]:
    """Run the graph with a single input and config.

    Args:
        input: The input data for the graph. It can be a dictionary or any other type.
        config: Optional. The configuration for the graph run.
        stream_mode: Optional[str]. The stream mode for the graph run. Default is "values".
        output_keys: Optional. The output keys to retrieve from the graph run.
        interrupt_before: Optional. The nodes to interrupt the graph run before.
        interrupt_after: Optional. The nodes to interrupt the graph run after.
        debug: Optional. Enable debug mode for the graph run.
        **kwargs: Additional keyword arguments to pass to the graph run.

    Returns:
        The output of the graph run. If stream_mode is "values", it returns the latest output.
        If stream_mode is not "values", it returns a list of output chunks.
    """
    output_keys = output_keys if output_keys is not None else self.output_channels
    if stream_mode == "values":
        latest: Union[dict[str, Any], Any] = None
    else:
        chunks = []
    for chunk in self.stream(
        input,
        config,
        stream_mode=stream_mode,
        output_keys=output_keys,
        interrupt_before=interrupt_before,
        interrupt_after=interrupt_after,
        debug=debug,
        **kwargs,
    ):
        if stream_mode == "values":
            latest = chunk
        else:
            chunks.append(chunk)
    if stream_mode == "values":
        return latest
    else:
        return chunks

ainvoke(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: StreamMode = 'values', output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, **kwargs: Any) -> Union[dict[str, Any], Any] async

异步在单个输入上调用图。

参数

• input (Union[dict[str, Any], Any]) –

  计算的输入数据。它可以是字典或任何其他类型。

• config (Optional[RunnableConfig], default: None ) –

  可选。计算的配置。

• stream_mode (StreamMode, default: 'values' ) –

  可选。计算的流模式。默认值为 "values"。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  可选。要包含在结果中的输出键。默认值为 None。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。要在其之前中断的节点。默认值为 None。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。要在其之后中断的节点。默认值为 None。

• debug (Optional[bool], default: None ) –

  可选。是否启用调试模式。默认值为 None。

• **kwargs (Any, default: {} ) –

  附加关键字参数。

返回值

• Union[dict[str, Any], Any] –

  计算的结果。如果 stream_mode 为 "values"，则返回最新的值。

• Union[dict[str, Any], Any] –

  如果 stream_mode 为 "chunks"，则返回块列表。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def ainvoke(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: StreamMode = "values",
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    **kwargs: Any,
) -> Union[dict[str, Any], Any]:
    """Asynchronously invoke the graph on a single input.

    Args:
        input: The input data for the computation. It can be a dictionary or any other type.
        config: Optional. The configuration for the computation.
        stream_mode: Optional. The stream mode for the computation. Default is "values".
        output_keys: Optional. The output keys to include in the result. Default is None.
        interrupt_before: Optional. The nodes to interrupt before. Default is None.
        interrupt_after: Optional. The nodes to interrupt after. Default is None.
        debug: Optional. Whether to enable debug mode. Default is None.
        **kwargs: Additional keyword arguments.

    Returns:
        The result of the computation. If stream_mode is "values", it returns the latest value.
        If stream_mode is "chunks", it returns a list of chunks.
    """

    output_keys = output_keys if output_keys is not None else self.output_channels
    if stream_mode == "values":
        latest: Union[dict[str, Any], Any] = None
    else:
        chunks = []
    async for chunk in self.astream(
        input,
        config,
        stream_mode=stream_mode,
        output_keys=output_keys,
        interrupt_before=interrupt_before,
        interrupt_after=interrupt_after,
        debug=debug,
        **kwargs,
    ):
        if stream_mode == "values":
            latest = chunk
        else:
            chunks.append(chunk)
    if stream_mode == "values":
        return latest
    else:
        return chunks

get_graph(config: Optional[RunnableConfig] = None, *, xray: Union[int, bool] = False) -> DrawableGraph

返回计算图的可绘制表示形式。

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def get_graph(
    self,
    config: Optional[RunnableConfig] = None,
    *,
    xray: Union[int, bool] = False,
) -> DrawableGraph:
    """Returns a drawable representation of the computation graph."""
    graph = DrawableGraph()
    start_nodes: dict[str, DrawableNode] = {
        START: graph.add_node(self.get_input_schema(config), START)
    }
    end_nodes: dict[str, DrawableNode] = {}
    if xray:
        subgraphs = {
            k: v for k, v in self.get_subgraphs() if isinstance(v, CompiledGraph)
        }
    else:
        subgraphs = {}

    def add_edge(
        start: str,
        end: str,
        label: Optional[Hashable] = None,
        conditional: bool = False,
    ) -> None:
        if end == END and END not in end_nodes:
            end_nodes[END] = graph.add_node(self.get_output_schema(config), END)
        return graph.add_edge(
            start_nodes[start],
            end_nodes[end],
            str(label) if label is not None else None,
            conditional,
        )

    for key, n in self.builder.nodes.items():
        node = n.runnable
        metadata = n.metadata or {}
        if key in self.interrupt_before_nodes and key in self.interrupt_after_nodes:
            metadata["__interrupt"] = "before,after"
        elif key in self.interrupt_before_nodes:
            metadata["__interrupt"] = "before"
        elif key in self.interrupt_after_nodes:
            metadata["__interrupt"] = "after"
        if xray:
            subgraph = (
                subgraphs[key].get_graph(
                    config=config,
                    xray=xray - 1
                    if isinstance(xray, int)
                    and not isinstance(xray, bool)
                    and xray > 0
                    else xray,
                )
                if key in subgraphs
                else node.get_graph(config=config)
            )
            subgraph.trim_first_node()
            subgraph.trim_last_node()
            if len(subgraph.nodes) > 1:
                e, s = graph.extend(subgraph, prefix=key)
                if e is None:
                    raise ValueError(
                        f"Could not extend subgraph '{key}' due to missing entrypoint"
                    )
                if s is not None:
                    start_nodes[key] = s
                end_nodes[key] = e
            else:
                nn = graph.add_node(node, key, metadata=metadata or None)
                start_nodes[key] = nn
                end_nodes[key] = nn
        else:
            nn = graph.add_node(node, key, metadata=metadata or None)
            start_nodes[key] = nn
            end_nodes[key] = nn
    for start, end in sorted(self.builder._all_edges):
        add_edge(start, end)
    for start, branches in self.builder.branches.items():
        default_ends = {
            **{k: k for k in self.builder.nodes if k != start},
            END: END,
        }
        for _, branch in branches.items():
            if branch.ends is not None:
                ends = branch.ends
            elif branch.then is not None:
                ends = {k: k for k in default_ends if k not in (END, branch.then)}
            else:
                ends = cast(dict[Hashable, str], default_ends)
            for label, end in ends.items():
                add_edge(
                    start,
                    end,
                    label if label != end else None,
                    conditional=True,
                )
                if branch.then is not None:
                    add_edge(end, branch.then)

    return graph

StateGraph

Bases: Graph

一个图，其节点通过读写共享状态进行通信。每个节点的签名为 State -> Partial.

每个状态键都可以选择性地使用 reducer 函数进行注释，该函数将用于聚合从多个节点接收到的该键的值。reducer 函数的签名为 (Value, Value) -> Value。

参数

• state_schema (Type[Any], default: None ) –

  定义状态的模式类。

• config_schema (Optional[Type[Any]], default: None ) –

  定义配置的模式类。使用它在您的 API 中公开可配置的参数。

Examples

>>> from langchain_core.runnables import RunnableConfig
>>> from typing_extensions import Annotated, TypedDict
>>> from langgraph.checkpoint.memory import MemorySaver
>>> from langgraph.graph import StateGraph
>>>
>>> def reducer(a: list, b: int | None) -> list:
...     if b is not None:
...         return a + [b]
...     return a
>>>
>>> class State(TypedDict):
...     x: Annotated[list, reducer]
>>>
>>> class ConfigSchema(TypedDict):
...     r: float
>>>
>>> graph = StateGraph(State, config_schema=ConfigSchema)
>>>
>>> def node(state: State, config: RunnableConfig) -> dict:
...     r = config["configurable"].get("r", 1.0)
...     x = state["x"][-1]
...     next_value = x * r * (1 - x)
...     return {"x": next_value}
>>>
>>> graph.add_node("A", node)
>>> graph.set_entry_point("A")
>>> graph.set_finish_point("A")
>>> compiled = graph.compile()
>>>
>>> print(compiled.config_specs)
[ConfigurableFieldSpec(id='r', annotation=<class 'float'>, name=None, description=None, default=None, is_shared=False, dependencies=None)]
>>>
>>> step1 = compiled.invoke({"x": 0.5}, {"configurable": {"r": 3.0}})
>>> print(step1)
{'x': [0.5, 0.75]}

源代码位于 libs/langgraph/langgraph/graph/state.py 中

class StateGraph(Graph):
    """A graph whose nodes communicate by reading and writing to a shared state.
    The signature of each node is State -> Partial<State>.

    Each state key can optionally be annotated with a reducer function that
    will be used to aggregate the values of that key received from multiple nodes.
    The signature of a reducer function is (Value, Value) -> Value.

    Args:
        state_schema (Type[Any]): The schema class that defines the state.
        config_schema (Optional[Type[Any]]): The schema class that defines the configuration.
            Use this to expose configurable parameters in your API.


    Examples:
        >>> from langchain_core.runnables import RunnableConfig
        >>> from typing_extensions import Annotated, TypedDict
        >>> from langgraph.checkpoint.memory import MemorySaver
        >>> from langgraph.graph import StateGraph
        >>>
        >>> def reducer(a: list, b: int | None) -> list:
        ...     if b is not None:
        ...         return a + [b]
        ...     return a
        >>>
        >>> class State(TypedDict):
        ...     x: Annotated[list, reducer]
        >>>
        >>> class ConfigSchema(TypedDict):
        ...     r: float
        >>>
        >>> graph = StateGraph(State, config_schema=ConfigSchema)
        >>>
        >>> def node(state: State, config: RunnableConfig) -> dict:
        ...     r = config["configurable"].get("r", 1.0)
        ...     x = state["x"][-1]
        ...     next_value = x * r * (1 - x)
        ...     return {"x": next_value}
        >>>
        >>> graph.add_node("A", node)
        >>> graph.set_entry_point("A")
        >>> graph.set_finish_point("A")
        >>> compiled = graph.compile()
        >>>
        >>> print(compiled.config_specs)
        [ConfigurableFieldSpec(id='r', annotation=<class 'float'>, name=None, description=None, default=None, is_shared=False, dependencies=None)]
        >>>
        >>> step1 = compiled.invoke({"x": 0.5}, {"configurable": {"r": 3.0}})
        >>> print(step1)
        {'x': [0.5, 0.75]}"""

    nodes: dict[str, StateNodeSpec]  # type: ignore[assignment]
    channels: dict[str, BaseChannel]
    managed: dict[str, ManagedValueSpec]
    schemas: dict[Type[Any], dict[str, Union[BaseChannel, ManagedValueSpec]]]

    def __init__(
        self,
        state_schema: Optional[Type[Any]] = None,
        config_schema: Optional[Type[Any]] = None,
        *,
        input: Optional[Type[Any]] = None,
        output: Optional[Type[Any]] = None,
    ) -> None:
        super().__init__()
        if state_schema is None:
            if input is None or output is None:
                raise ValueError("Must provide state_schema or input and output")
            state_schema = input
            warnings.warn(
                "Initializing StateGraph without state_schema is deprecated. "
                "Please pass in an explicit state_schema instead of just an input and output schema.",
                LangGraphDeprecationWarning,
                stacklevel=2,
            )
        else:
            if input is None:
                input = state_schema
            if output is None:
                output = state_schema
        self.schemas = {}
        self.channels = {}
        self.managed = {}
        self.schema = state_schema
        self.input = input
        self.output = output
        self._add_schema(state_schema)
        self._add_schema(input, allow_managed=False)
        self._add_schema(output, allow_managed=False)
        self.config_schema = config_schema
        self.waiting_edges: set[tuple[tuple[str, ...], str]] = set()

    @property
    def _all_edges(self) -> set[tuple[str, str]]:
        return self.edges | {
            (start, end) for starts, end in self.waiting_edges for start in starts
        }

    def _add_schema(self, schema: Type[Any], /, allow_managed: bool = True) -> None:
        if schema not in self.schemas:
            _warn_invalid_state_schema(schema)
            channels, managed = _get_channels(schema)
            if managed and not allow_managed:
                names = ", ".join(managed)
                schema_name = getattr(schema, "__name__", "")
                raise ValueError(
                    f"Invalid managed channels detected in {schema_name}: {names}."
                    " Managed channels are not permitted in Input/Output schema."
                )
            self.schemas[schema] = {**channels, **managed}
            for key, channel in channels.items():
                if key in self.channels:
                    if self.channels[key] != channel:
                        if isinstance(channel, LastValue):
                            pass
                        else:
                            raise ValueError(
                                f"Channel '{key}' already exists with a different type"
                            )
                else:
                    self.channels[key] = channel
            for key, managed in managed.items():
                if key in self.managed:
                    if self.managed[key] != managed:
                        raise ValueError(
                            f"Managed value '{key}' already exists with a different type"
                        )
                else:
                    self.managed[key] = managed

    @overload
    def add_node(
        self,
        node: RunnableLike,
        *,
        metadata: Optional[dict[str, Any]] = None,
        input: Optional[Type[Any]] = None,
        retry: Optional[RetryPolicy] = None,
    ) -> Self:
        """Adds a new node to the state graph.
        Will take the name of the function/runnable as the node name.

        Args:
            node (RunnableLike): The function or runnable this node will run.

        Raises:
            ValueError: If the key is already being used as a state key.

        Returns:
            None
        """
        ...

    @overload
    def add_node(
        self,
        node: str,
        action: RunnableLike,
        *,
        metadata: Optional[dict[str, Any]] = None,
        input: Optional[Type[Any]] = None,
        retry: Optional[RetryPolicy] = None,
    ) -> Self:
        """Adds a new node to the state graph.

        Args:
            node (str): The key of the node.
            action (RunnableLike): The action associated with the node.

        Raises:
            ValueError: If the key is already being used as a state key.

        Returns:
            None
        """
        ...

    def add_node(
        self,
        node: Union[str, RunnableLike],
        action: Optional[RunnableLike] = None,
        *,
        metadata: Optional[dict[str, Any]] = None,
        input: Optional[Type[Any]] = None,
        retry: Optional[RetryPolicy] = None,
    ) -> Self:
        """Adds a new node to the state graph.

        Will take the name of the function/runnable as the node name.

        Args:
            node (Union[str, RunnableLike)]: The function or runnable this node will run.
            action (Optional[RunnableLike]): The action associated with the node. (default: None)
            metadata (Optional[dict[str, Any]]): The metadata associated with the node. (default: None)
            input (Optional[Type[Any]]): The input schema for the node. (default: the graph's input schema)
            retry (Optional[RetryPolicy]): The policy for retrying the node. (default: None)
        Raises:
            ValueError: If the key is already being used as a state key.


        Examples:
            ```pycon
            >>> from langgraph.graph import START, StateGraph
            ...
            >>> def my_node(state, config):
            ...    return {"x": state["x"] + 1}
            ...
            >>> builder = StateGraph(dict)
            >>> builder.add_node(my_node)  # node name will be 'my_node'
            >>> builder.add_edge(START, "my_node")
            >>> graph = builder.compile()
            >>> graph.invoke({"x": 1})
            {'x': 2}
            ```
            Customize the name:

            ```pycon
            >>> builder = StateGraph(dict)
            >>> builder.add_node("my_fair_node", my_node)
            >>> builder.add_edge(START, "my_fair_node")
            >>> graph = builder.compile()
            >>> graph.invoke({"x": 1})
            {'x': 2}
            ```

        Returns:
            None
        """
        if not isinstance(node, str):
            action = node
            if isinstance(action, Runnable):
                node = action.get_name()
            else:
                node = getattr(action, "__name__", action.__class__.__name__)
            if node is None:
                raise ValueError(
                    "Node name must be provided if action is not a function"
                )
        if node in self.channels:
            raise ValueError(f"'{node}' is already being used as a state key")
        if self.compiled:
            logger.warning(
                "Adding a node to a graph that has already been compiled. This will "
                "not be reflected in the compiled graph."
            )
        if not isinstance(node, str):
            action = node
            node = cast(str, getattr(action, "name", getattr(action, "__name__", None)))
            if node is None:
                raise ValueError(
                    "Node name must be provided if action is not a function"
                )
        if action is None:
            raise RuntimeError
        if node in self.nodes:
            raise ValueError(f"Node `{node}` already present.")
        if node == END or node == START:
            raise ValueError(f"Node `{node}` is reserved.")

        for character in (NS_SEP, NS_END):
            if character in cast(str, node):
                raise ValueError(
                    f"'{character}' is a reserved character and is not allowed in the node names."
                )

        try:
            if isfunction(action) and (
                hints := get_type_hints(action.__call__) or get_type_hints(action)
            ):
                if input is None:
                    first_parameter_name = next(
                        iter(inspect.signature(action).parameters.keys())
                    )
                    if input_hint := hints.get(first_parameter_name):
                        if isinstance(input_hint, type) and get_type_hints(input_hint):
                            input = input_hint
        except (TypeError, StopIteration):
            pass
        if input is not None:
            self._add_schema(input)
        self.nodes[cast(str, node)] = StateNodeSpec(
            coerce_to_runnable(action, name=cast(str, node), trace=False),
            metadata,
            input=input or self.schema,
            retry_policy=retry,
        )
        return self

    def add_edge(self, start_key: Union[str, list[str]], end_key: str) -> Self:
        """Adds a directed edge from the start node to the end node.

        If the graph transitions to the start_key node, it will always transition to the end_key node next.

        Args:
            start_key (Union[str, list[str]]): The key(s) of the start node(s) of the edge.
            end_key (str): The key of the end node of the edge.

        Raises:
            ValueError: If the start key is 'END' or if the start key or end key is not present in the graph.

        Returns:
            None
        """
        if isinstance(start_key, str):
            return super().add_edge(start_key, end_key)

        if self.compiled:
            logger.warning(
                "Adding an edge to a graph that has already been compiled. This will "
                "not be reflected in the compiled graph."
            )
        for start in start_key:
            if start == END:
                raise ValueError("END cannot be a start node")
            if start not in self.nodes:
                raise ValueError(f"Need to add_node `{start}` first")
        if end_key == START:
            raise ValueError("START cannot be an end node")
        if end_key != END and end_key not in self.nodes:
            raise ValueError(f"Need to add_node `{end_key}` first")

        self.waiting_edges.add((tuple(start_key), end_key))
        return self

    def compile(
        self,
        checkpointer: Checkpointer = None,
        *,
        store: Optional[BaseStore] = None,
        interrupt_before: Optional[Union[All, list[str]]] = None,
        interrupt_after: Optional[Union[All, list[str]]] = None,
        debug: bool = False,
    ) -> "CompiledStateGraph":
        """Compiles the state graph into a `CompiledGraph` object.

        The compiled graph implements the `Runnable` interface and can be invoked,
        streamed, batched, and run asynchronously.

        Args:
            checkpointer (Checkpointer): An optional checkpoint saver object.
                This serves as a fully versioned "memory" for the graph, allowing
                the graph to be paused and resumed, and replayed from any point.
            interrupt_before (Optional[Sequence[str]]): An optional list of node names to interrupt before.
            interrupt_after (Optional[Sequence[str]]): An optional list of node names to interrupt after.
            debug (bool): A flag indicating whether to enable debug mode.

        Returns:
            CompiledStateGraph: The compiled state graph.
        """
        # assign default values
        interrupt_before = interrupt_before or []
        interrupt_after = interrupt_after or []

        # validate the graph
        self.validate(
            interrupt=(
                (interrupt_before if interrupt_before != "*" else []) + interrupt_after
                if interrupt_after != "*"
                else []
            )
        )

        # prepare output channels
        output_channels = (
            "__root__"
            if len(self.schemas[self.output]) == 1
            and "__root__" in self.schemas[self.output]
            else [
                key
                for key, val in self.schemas[self.output].items()
                if not is_managed_value(val)
            ]
        )
        stream_channels = (
            "__root__"
            if len(self.channels) == 1 and "__root__" in self.channels
            else [
                key for key, val in self.channels.items() if not is_managed_value(val)
            ]
        )

        compiled = CompiledStateGraph(
            builder=self,
            config_type=self.config_schema,
            nodes={},
            channels={
                **self.channels,
                **self.managed,
                START: EphemeralValue(self.input),
            },
            input_channels=START,
            stream_mode="updates",
            output_channels=output_channels,
            stream_channels=stream_channels,
            checkpointer=checkpointer,
            interrupt_before_nodes=interrupt_before,
            interrupt_after_nodes=interrupt_after,
            auto_validate=False,
            debug=debug,
            store=store,
        )

        compiled.attach_node(START, None)
        for key, node in self.nodes.items():
            compiled.attach_node(key, node)

        for start, end in self.edges:
            compiled.attach_edge(start, end)

        for starts, end in self.waiting_edges:
            compiled.attach_edge(starts, end)

        for start, branches in self.branches.items():
            for name, branch in branches.items():
                compiled.attach_branch(start, name, branch)

        return compiled.validate()

add_conditional_edges(source: str, path: Union[Callable[..., Union[Hashable, list[Hashable]]], Callable[..., Awaitable[Union[Hashable, list[Hashable]]]], Runnable[Any, Union[Hashable, list[Hashable]]]], path_map: Optional[Union[dict[Hashable, str], list[str]]] = None, then: Optional[str] = None) -> Self

从起始节点添加一个条件边到任意数量的目标节点。

参数

• source (str) –

  起始节点。当退出该节点时，将运行此条件边。

• path (Union[Callable, Runnable]) –

  确定下一个节点或节点的函数。如果未指定 path_map，则应返回一个或多个节点。如果返回 END，则图形将停止执行。

• path_map (Optional[dict[Hashable, str]], 默认： None ) –

  路径到节点名称的可选映射。如果省略，则 path 返回的路径应为节点名称。

• then (Optional[str], 默认： None ) –

  由 path 选择的节点执行完毕后，将要执行的节点的名称。

返回值

• Self –

  无

在 path 函数的返回值上没有类型提示（例如， -> Literal["foo", "__end__"]:）

或者没有 path_map，图形可视化假定边可以转换到图形中的任何节点。

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def add_conditional_edges(
    self,
    source: str,
    path: Union[
        Callable[..., Union[Hashable, list[Hashable]]],
        Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
        Runnable[Any, Union[Hashable, list[Hashable]]],
    ],
    path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
    then: Optional[str] = None,
) -> Self:
    """Add a conditional edge from the starting node to any number of destination nodes.

    Args:
        source (str): The starting node. This conditional edge will run when
            exiting this node.
        path (Union[Callable, Runnable]): The callable that determines the next
            node or nodes. If not specifying `path_map` it should return one or
            more nodes. If it returns END, the graph will stop execution.
        path_map (Optional[dict[Hashable, str]]): Optional mapping of paths to node
            names. If omitted the paths returned by `path` should be node names.
        then (Optional[str]): The name of a node to execute after the nodes
            selected by `path`.

    Returns:
        None

    Note: Without typehints on the `path` function's return value (e.g., `-> Literal["foo", "__end__"]:`)
        or a path_map, the graph visualization assumes the edge could transition to any node in the graph.

    """  # noqa: E501
    if self.compiled:
        logger.warning(
            "Adding an edge to a graph that has already been compiled. This will "
            "not be reflected in the compiled graph."
        )
    # coerce path_map to a dictionary
    try:
        if isinstance(path_map, dict):
            path_map_ = path_map.copy()
        elif isinstance(path_map, list):
            path_map_ = {name: name for name in path_map}
        elif isinstance(path, Runnable):
            path_map_ = None
        elif rtn_type := get_type_hints(path.__call__).get(  # type: ignore[operator]
            "return"
        ) or get_type_hints(path).get("return"):
            if get_origin(rtn_type) is Literal:
                path_map_ = {name: name for name in get_args(rtn_type)}
            else:
                path_map_ = None
        else:
            path_map_ = None
    except Exception:
        path_map_ = None
    # find a name for the condition
    path = coerce_to_runnable(path, name=None, trace=True)
    name = path.name or "condition"
    # validate the condition
    if name in self.branches[source]:
        raise ValueError(
            f"Branch with name `{path.name}` already exists for node " f"`{source}`"
        )
    # save it
    self.branches[source][name] = Branch(path, path_map_, then)
    return self

set_entry_point(key: str) -> Self

指定要调用的第一个节点。

相当于调用 add_edge(START, key)。

参数

• key (str) –

  要设置为入口点的节点的键。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_entry_point(self, key: str) -> Self:
    """Specifies the first node to be called in the graph.

    Equivalent to calling `add_edge(START, key)`.

    Parameters:
        key (str): The key of the node to set as the entry point.

    Returns:
        None
    """
    return self.add_edge(START, key)

set_conditional_entry_point(path: Union[Callable[..., Union[Hashable, list[Hashable]]], Callable[..., Awaitable[Union[Hashable, list[Hashable]]]], Runnable[Any, Union[Hashable, list[Hashable]]]], path_map: Optional[Union[dict[Hashable, str], list[str]]] = None, then: Optional[str] = None) -> Self

在图形中设置条件入口点。

参数

• path (Union[Callable, Runnable]) –

  确定下一个节点或节点的函数。如果未指定 path_map，则应返回一个或多个节点。如果返回 END，则图形将停止执行。

• path_map (Optional[dict[str, str]], 默认： None ) –

  路径到节点名称的可选映射。如果省略，则 path 返回的路径应为节点名称。

• then (Optional[str], 默认： None ) –

  由 path 选择的节点执行完毕后，将要执行的节点的名称。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_conditional_entry_point(
    self,
    path: Union[
        Callable[..., Union[Hashable, list[Hashable]]],
        Callable[..., Awaitable[Union[Hashable, list[Hashable]]]],
        Runnable[Any, Union[Hashable, list[Hashable]]],
    ],
    path_map: Optional[Union[dict[Hashable, str], list[str]]] = None,
    then: Optional[str] = None,
) -> Self:
    """Sets a conditional entry point in the graph.

    Args:
        path (Union[Callable, Runnable]): The callable that determines the next
            node or nodes. If not specifying `path_map` it should return one or
            more nodes. If it returns END, the graph will stop execution.
        path_map (Optional[dict[str, str]]): Optional mapping of paths to node
            names. If omitted the paths returned by `path` should be node names.
        then (Optional[str]): The name of a node to execute after the nodes
            selected by `path`.

    Returns:
        None
    """
    return self.add_conditional_edges(START, path, path_map, then)

set_finish_point(key: str) -> Self

将节点标记为图形的结束点。

如果图形到达此节点，则将停止执行。

参数

• key (str) –

  要设置为结束点的节点的键。

返回值

• Self –

  无

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def set_finish_point(self, key: str) -> Self:
    """Marks a node as a finish point of the graph.

    If the graph reaches this node, it will cease execution.

    Parameters:
        key (str): The key of the node to set as the finish point.

    Returns:
        None
    """
    return self.add_edge(key, END)

add_node(node: Union[str, RunnableLike], action: Optional[RunnableLike] = None, *, metadata: Optional[dict[str, Any]] = None, input: Optional[Type[Any]] = None, retry: Optional[RetryPolicy] = None) -> Self

在状态图中添加一个新节点。

将使用函数/可运行项的名称作为节点名称。

参数

• node (Union[str, RunnableLike)]) –

  此节点将运行的函数或可运行项。

• action (Optional[RunnableLike], default: None ) –

  与节点关联的操作。（默认：None）

• metadata (Optional[dict[str, Any]], default: None ) –

  与节点关联的元数据。（默认：None）

• input (Optional[Type[Any]], default: None ) –

  节点的输入模式。（默认：图的输入模式）

• retry (Optional[RetryPolicy], default: None ) –

  重试节点的策略。（默认：None）

引发：ValueError：如果键已用作状态键。

Examples

>>> from langgraph.graph import START, StateGraph
...
>>> def my_node(state, config):
...    return {"x": state["x"] + 1}
...
>>> builder = StateGraph(dict)
>>> builder.add_node(my_node)  # node name will be 'my_node'
>>> builder.add_edge(START, "my_node")
>>> graph = builder.compile()
>>> graph.invoke({"x": 1})
{'x': 2}

自定义名称

>>> builder = StateGraph(dict)
>>> builder.add_node("my_fair_node", my_node)
>>> builder.add_edge(START, "my_fair_node")
>>> graph = builder.compile()
>>> graph.invoke({"x": 1})
{'x': 2}

返回值

• Self –

  无

源代码位于 libs/langgraph/langgraph/graph/state.py 中

def add_node(
    self,
    node: Union[str, RunnableLike],
    action: Optional[RunnableLike] = None,
    *,
    metadata: Optional[dict[str, Any]] = None,
    input: Optional[Type[Any]] = None,
    retry: Optional[RetryPolicy] = None,
) -> Self:
    """Adds a new node to the state graph.

    Will take the name of the function/runnable as the node name.

    Args:
        node (Union[str, RunnableLike)]: The function or runnable this node will run.
        action (Optional[RunnableLike]): The action associated with the node. (default: None)
        metadata (Optional[dict[str, Any]]): The metadata associated with the node. (default: None)
        input (Optional[Type[Any]]): The input schema for the node. (default: the graph's input schema)
        retry (Optional[RetryPolicy]): The policy for retrying the node. (default: None)
    Raises:
        ValueError: If the key is already being used as a state key.


    Examples:
        ```pycon
        >>> from langgraph.graph import START, StateGraph
        ...
        >>> def my_node(state, config):
        ...    return {"x": state["x"] + 1}
        ...
        >>> builder = StateGraph(dict)
        >>> builder.add_node(my_node)  # node name will be 'my_node'
        >>> builder.add_edge(START, "my_node")
        >>> graph = builder.compile()
        >>> graph.invoke({"x": 1})
        {'x': 2}
        ```
        Customize the name:

        ```pycon
        >>> builder = StateGraph(dict)
        >>> builder.add_node("my_fair_node", my_node)
        >>> builder.add_edge(START, "my_fair_node")
        >>> graph = builder.compile()
        >>> graph.invoke({"x": 1})
        {'x': 2}
        ```

    Returns:
        None
    """
    if not isinstance(node, str):
        action = node
        if isinstance(action, Runnable):
            node = action.get_name()
        else:
            node = getattr(action, "__name__", action.__class__.__name__)
        if node is None:
            raise ValueError(
                "Node name must be provided if action is not a function"
            )
    if node in self.channels:
        raise ValueError(f"'{node}' is already being used as a state key")
    if self.compiled:
        logger.warning(
            "Adding a node to a graph that has already been compiled. This will "
            "not be reflected in the compiled graph."
        )
    if not isinstance(node, str):
        action = node
        node = cast(str, getattr(action, "name", getattr(action, "__name__", None)))
        if node is None:
            raise ValueError(
                "Node name must be provided if action is not a function"
            )
    if action is None:
        raise RuntimeError
    if node in self.nodes:
        raise ValueError(f"Node `{node}` already present.")
    if node == END or node == START:
        raise ValueError(f"Node `{node}` is reserved.")

    for character in (NS_SEP, NS_END):
        if character in cast(str, node):
            raise ValueError(
                f"'{character}' is a reserved character and is not allowed in the node names."
            )

    try:
        if isfunction(action) and (
            hints := get_type_hints(action.__call__) or get_type_hints(action)
        ):
            if input is None:
                first_parameter_name = next(
                    iter(inspect.signature(action).parameters.keys())
                )
                if input_hint := hints.get(first_parameter_name):
                    if isinstance(input_hint, type) and get_type_hints(input_hint):
                        input = input_hint
    except (TypeError, StopIteration):
        pass
    if input is not None:
        self._add_schema(input)
    self.nodes[cast(str, node)] = StateNodeSpec(
        coerce_to_runnable(action, name=cast(str, node), trace=False),
        metadata,
        input=input or self.schema,
        retry_policy=retry,
    )
    return self

add_edge(start_key: Union[str, list[str]], end_key: str) -> Self

从起始节点到结束节点添加一条有向边。

如果图过渡到 start_key 节点，它将始终接下来过渡到 end_key 节点。

参数

• start_key (Union[str, list[str]]) –

  边起始节点的键。

• end_key (str) –

  边的结束节点的键。

引发

• ValueError –

  如果起始键是 'END' 或如果起始键或结束键不存在于图中。

返回值

• Self –

  无

源代码位于 libs/langgraph/langgraph/graph/state.py 中

def add_edge(self, start_key: Union[str, list[str]], end_key: str) -> Self:
    """Adds a directed edge from the start node to the end node.

    If the graph transitions to the start_key node, it will always transition to the end_key node next.

    Args:
        start_key (Union[str, list[str]]): The key(s) of the start node(s) of the edge.
        end_key (str): The key of the end node of the edge.

    Raises:
        ValueError: If the start key is 'END' or if the start key or end key is not present in the graph.

    Returns:
        None
    """
    if isinstance(start_key, str):
        return super().add_edge(start_key, end_key)

    if self.compiled:
        logger.warning(
            "Adding an edge to a graph that has already been compiled. This will "
            "not be reflected in the compiled graph."
        )
    for start in start_key:
        if start == END:
            raise ValueError("END cannot be a start node")
        if start not in self.nodes:
            raise ValueError(f"Need to add_node `{start}` first")
    if end_key == START:
        raise ValueError("START cannot be an end node")
    if end_key != END and end_key not in self.nodes:
        raise ValueError(f"Need to add_node `{end_key}` first")

    self.waiting_edges.add((tuple(start_key), end_key))
    return self

compile(checkpointer: Checkpointer = None, *, store: Optional[BaseStore] = None, interrupt_before: Optional[Union[All, list[str]]] = None, interrupt_after: Optional[Union[All, list[str]]] = None, debug: bool = False) -> CompiledStateGraph

将状态图编译成一个 CompiledGraph 对象。

编译后的图实现了 Runnable 接口，可以调用、流式传输、批处理和异步运行。

参数

• checkpointer (Checkpointer, default: None ) –

  一个可选的检查点保存器对象。这充当图的完整版本化的“内存”，允许暂停和恢复图，并从任何点重播。

• interrupt_before (Optional[Sequence[str]], default: None ) –

  一个可选的节点名称列表，在这些节点之前中断。

• interrupt_after (Optional[Sequence[str]], default: None ) –

  一个可选的节点名称列表，在这些节点之后中断。

• debug (bool, default: False ) –

  一个标志，指示是否启用调试模式。

返回值

• CompiledStateGraph ( CompiledStateGraph ) –

  编译后的状态图。

源代码位于 libs/langgraph/langgraph/graph/state.py 中

def compile(
    self,
    checkpointer: Checkpointer = None,
    *,
    store: Optional[BaseStore] = None,
    interrupt_before: Optional[Union[All, list[str]]] = None,
    interrupt_after: Optional[Union[All, list[str]]] = None,
    debug: bool = False,
) -> "CompiledStateGraph":
    """Compiles the state graph into a `CompiledGraph` object.

    The compiled graph implements the `Runnable` interface and can be invoked,
    streamed, batched, and run asynchronously.

    Args:
        checkpointer (Checkpointer): An optional checkpoint saver object.
            This serves as a fully versioned "memory" for the graph, allowing
            the graph to be paused and resumed, and replayed from any point.
        interrupt_before (Optional[Sequence[str]]): An optional list of node names to interrupt before.
        interrupt_after (Optional[Sequence[str]]): An optional list of node names to interrupt after.
        debug (bool): A flag indicating whether to enable debug mode.

    Returns:
        CompiledStateGraph: The compiled state graph.
    """
    # assign default values
    interrupt_before = interrupt_before or []
    interrupt_after = interrupt_after or []

    # validate the graph
    self.validate(
        interrupt=(
            (interrupt_before if interrupt_before != "*" else []) + interrupt_after
            if interrupt_after != "*"
            else []
        )
    )

    # prepare output channels
    output_channels = (
        "__root__"
        if len(self.schemas[self.output]) == 1
        and "__root__" in self.schemas[self.output]
        else [
            key
            for key, val in self.schemas[self.output].items()
            if not is_managed_value(val)
        ]
    )
    stream_channels = (
        "__root__"
        if len(self.channels) == 1 and "__root__" in self.channels
        else [
            key for key, val in self.channels.items() if not is_managed_value(val)
        ]
    )

    compiled = CompiledStateGraph(
        builder=self,
        config_type=self.config_schema,
        nodes={},
        channels={
            **self.channels,
            **self.managed,
            START: EphemeralValue(self.input),
        },
        input_channels=START,
        stream_mode="updates",
        output_channels=output_channels,
        stream_channels=stream_channels,
        checkpointer=checkpointer,
        interrupt_before_nodes=interrupt_before,
        interrupt_after_nodes=interrupt_after,
        auto_validate=False,
        debug=debug,
        store=store,
    )

    compiled.attach_node(START, None)
    for key, node in self.nodes.items():
        compiled.attach_node(key, node)

    for start, end in self.edges:
        compiled.attach_edge(start, end)

    for starts, end in self.waiting_edges:
        compiled.attach_edge(starts, end)

    for start, branches in self.branches.items():
        for name, branch in branches.items():
            compiled.attach_branch(start, name, branch)

    return compiled.validate()

CompiledStateGraph

基础：CompiledGraph

源代码位于 libs/langgraph/langgraph/graph/state.py 中

class CompiledStateGraph(CompiledGraph):
    builder: StateGraph

    def get_input_schema(
        self, config: Optional[RunnableConfig] = None
    ) -> type[BaseModel]:
        return _get_schema(
            typ=self.builder.input,
            schemas=self.builder.schemas,
            channels=self.builder.channels,
            name=self.get_name("Input"),
        )

    def get_output_schema(
        self, config: Optional[RunnableConfig] = None
    ) -> type[BaseModel]:
        return _get_schema(
            typ=self.builder.output,
            schemas=self.builder.schemas,
            channels=self.builder.channels,
            name=self.get_name("Output"),
        )

    def attach_node(self, key: str, node: Optional[StateNodeSpec]) -> None:
        if key == START:
            output_keys = [
                k
                for k, v in self.builder.schemas[self.builder.input].items()
                if not is_managed_value(v)
            ]
        else:
            output_keys = list(self.builder.channels) + [
                k
                for k, v in self.builder.managed.items()
                if is_writable_managed_value(v)
            ]

        def _get_state_key(input: Union[None, dict, Any], *, key: str) -> Any:
            if input is None:
                return SKIP_WRITE
            elif isinstance(input, dict):
                return input.get(key, SKIP_WRITE)
            elif get_type_hints(type(input)):
                value = getattr(input, key, SKIP_WRITE)
                return value if value is not None else SKIP_WRITE
            else:
                raise InvalidUpdateError(f"Expected dict, got {input}")

        # state updaters
        write_entries = (
            [ChannelWriteEntry("__root__", skip_none=True)]
            if output_keys == ["__root__"]
            else [
                ChannelWriteEntry(key, mapper=partial(_get_state_key, key=key))
                for key in output_keys
            ]
        )

        # add node and output channel
        if key == START:
            self.nodes[key] = PregelNode(
                tags=[TAG_HIDDEN],
                triggers=[START],
                channels=[START],
                writers=[
                    ChannelWrite(
                        write_entries,
                        tags=[TAG_HIDDEN],
                        require_at_least_one_of=output_keys,
                    ),
                ],
            )
        elif node is not None:
            input_schema = node.input if node else self.builder.schema
            input_values = {k: k for k in self.builder.schemas[input_schema]}
            is_single_input = len(input_values) == 1 and "__root__" in input_values

            self.channels[key] = EphemeralValue(Any, guard=False)
            self.nodes[key] = PregelNode(
                triggers=[],
                # read state keys and managed values
                channels=(list(input_values) if is_single_input else input_values),
                # coerce state dict to schema class (eg. pydantic model)
                mapper=(
                    None
                    if is_single_input or issubclass(input_schema, dict)
                    else partial(_coerce_state, input_schema)
                ),
                writers=[
                    # publish to this channel and state keys
                    ChannelWrite(
                        [ChannelWriteEntry(key, key)] + write_entries,
                        tags=[TAG_HIDDEN],
                        require_at_least_one_of=output_keys,
                    ),
                ],
                metadata=node.metadata,
                retry_policy=node.retry_policy,
                bound=node.runnable,
            )
        else:
            raise RuntimeError

    def attach_edge(self, starts: Union[str, Sequence[str]], end: str) -> None:
        if isinstance(starts, str):
            if starts == START:
                channel_name = f"start:{end}"
                # register channel
                self.channels[channel_name] = EphemeralValue(Any)
                # subscribe to channel
                self.nodes[end].triggers.append(channel_name)
                # publish to channel
                self.nodes[START] |= ChannelWrite(
                    [ChannelWriteEntry(channel_name, START)], tags=[TAG_HIDDEN]
                )
            elif end != END:
                # subscribe to start channel
                self.nodes[end].triggers.append(starts)
        elif end != END:
            channel_name = f"join:{'+'.join(starts)}:{end}"
            # register channel
            self.channels[channel_name] = NamedBarrierValue(str, set(starts))
            # subscribe to channel
            self.nodes[end].triggers.append(channel_name)
            # publish to channel
            for start in starts:
                self.nodes[start] |= ChannelWrite(
                    [ChannelWriteEntry(channel_name, start)], tags=[TAG_HIDDEN]
                )

    def attach_branch(self, start: str, name: str, branch: Branch) -> None:
        def branch_writer(
            packets: Sequence[Union[str, Send]], config: RunnableConfig
        ) -> None:
            if filtered := [p for p in packets if p != END]:
                writes = [
                    (
                        ChannelWriteEntry(f"branch:{start}:{name}:{p}", start)
                        if not isinstance(p, Send)
                        else p
                    )
                    for p in filtered
                ]
                if branch.then and branch.then != END:
                    writes.append(
                        ChannelWriteEntry(
                            f"branch:{start}:{name}::then",
                            WaitForNames(
                                {p.node if isinstance(p, Send) else p for p in filtered}
                            ),
                        )
                    )
                ChannelWrite.do_write(
                    config, cast(Sequence[Union[Send, ChannelWriteEntry]], writes)
                )

        # attach branch publisher
        schema = (
            self.builder.nodes[start].input
            if start in self.builder.nodes
            else self.builder.schema
        )
        self.nodes[start] |= branch.run(
            branch_writer, _get_state_reader(self.builder, schema)
        )

        # attach branch subscribers
        ends = (
            branch.ends.values()
            if branch.ends
            else [node for node in self.builder.nodes if node != branch.then]
        )
        for end in ends:
            if end != END:
                channel_name = f"branch:{start}:{name}:{end}"
                self.channels[channel_name] = EphemeralValue(Any, guard=False)
                self.nodes[end].triggers.append(channel_name)

        # attach then subscriber
        if branch.then and branch.then != END:
            channel_name = f"branch:{start}:{name}::then"
            self.channels[channel_name] = DynamicBarrierValue(str)
            self.nodes[branch.then].triggers.append(channel_name)
            for end in ends:
                if end != END:
                    self.nodes[end] |= ChannelWrite(
                        [ChannelWriteEntry(channel_name, end)], tags=[TAG_HIDDEN]
                    )

stream_mode: StreamMode = stream_mode class-attribute instance-attribute

流式传输输出的模式，默认为“values”。

stream_channels: Optional[Union[str, Sequence[str]]] = stream_channels class-attribute instance-attribute

要流式传输的通道，默认为所有不在保留通道中的通道。

step_timeout: Optional[float] = step_timeout class-attribute instance-attribute

等待步骤完成的最长时间，以秒为单位。默认为 None。

debug: bool = debug if debug is not None else get_debug() instance-attribute

是否在执行期间打印调试信息。默认为 False。

checkpointer: Checkpointer = checkpointer class-attribute instance-attribute

用于保存和加载图表状态的检查点。默认为 None。

store: Optional[BaseStore] = store class-attribute instance-attribute

用于 SharedValues 的内存存储。默认为 None。

retry_policy: Optional[RetryPolicy] = retry_policy 类属性 实例属性

运行任务时使用的重试策略。设置为 None 以禁用。

get_state(config: RunnableConfig, *, subgraphs: bool = False) -> StateSnapshot

获取图的当前状态。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def get_state(
    self, config: RunnableConfig, *, subgraphs: bool = False
) -> StateSnapshot:
    """Get the current state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return pregel.get_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                subgraphs=subgraphs,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(self.config, config) if self.config else config
    saved = checkpointer.get_tuple(config)
    return self._prepare_state_snapshot(
        config, saved, recurse=checkpointer if subgraphs else None
    )

aget_state(config: RunnableConfig, *, subgraphs: bool = False) -> StateSnapshot 异步

获取图的当前状态。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def aget_state(
    self, config: RunnableConfig, *, subgraphs: bool = False
) -> StateSnapshot:
    """Get the current state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        async for _, pregel in self.aget_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return await pregel.aget_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                subgraphs=subgraphs,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(self.config, config) if self.config else config
    saved = await checkpointer.aget_tuple(config)
    return await self._aprepare_state_snapshot(
        config, saved, recurse=checkpointer if subgraphs else None
    )

get_state_history(config: RunnableConfig, *, filter: Optional[Dict[str, Any]] = None, before: Optional[RunnableConfig] = None, limit: Optional[int] = None) -> Iterator[StateSnapshot]

获取图的状态历史记录。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def get_state_history(
    self,
    config: RunnableConfig,
    *,
    filter: Optional[Dict[str, Any]] = None,
    before: Optional[RunnableConfig] = None,
    limit: Optional[int] = None,
) -> Iterator[StateSnapshot]:
    """Get the history of the state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            yield from pregel.get_state_history(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                filter=filter,
                before=before,
                limit=limit,
            )
            return
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(
        self.config,
        config,
        {CONF: {CONFIG_KEY_CHECKPOINT_NS: checkpoint_ns}},
    )
    # eagerly consume list() to avoid holding up the db cursor
    for checkpoint_tuple in list(
        checkpointer.list(config, before=before, limit=limit, filter=filter)
    ):
        yield self._prepare_state_snapshot(
            checkpoint_tuple.config, checkpoint_tuple
        )

aget_state_history(config: RunnableConfig, *, filter: Optional[Dict[str, Any]] = None, before: Optional[RunnableConfig] = None, limit: Optional[int] = None) -> AsyncIterator[StateSnapshot] 异步

获取图的状态历史记录。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def aget_state_history(
    self,
    config: RunnableConfig,
    *,
    filter: Optional[Dict[str, Any]] = None,
    before: Optional[RunnableConfig] = None,
    limit: Optional[int] = None,
) -> AsyncIterator[StateSnapshot]:
    """Get the history of the state of the graph."""
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        async for _, pregel in self.aget_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            async for state in pregel.aget_state_history(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                filter=filter,
                before=before,
                limit=limit,
            ):
                yield state
            return
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    config = merge_configs(
        self.config,
        config,
        {CONF: {CONFIG_KEY_CHECKPOINT_NS: checkpoint_ns}},
    )
    # eagerly consume list() to avoid holding up the db cursor
    for checkpoint_tuple in [
        c
        async for c in checkpointer.alist(
            config, before=before, limit=limit, filter=filter
        )
    ]:
        yield await self._aprepare_state_snapshot(
            checkpoint_tuple.config, checkpoint_tuple
        )

update_state(config: RunnableConfig, values: Optional[Union[dict[str, Any], Any]], as_node: Optional[str] = None) -> RunnableConfig

使用给定的值更新图的状态，就好像它们来自节点 as_node。如果未提供 as_node，它将设置为更新状态的最后一个节点，如果不明确。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def update_state(
    self,
    config: RunnableConfig,
    values: Optional[Union[dict[str, Any], Any]],
    as_node: Optional[str] = None,
) -> RunnableConfig:
    """Update the state of the graph with the given values, as if they came from
    node `as_node`. If `as_node` is not provided, it will be set to the last node
    that updated the state, if not ambiguous.
    """
    checkpointer: Optional[BaseCheckpointSaver] = config[CONF].get(
        CONFIG_KEY_CHECKPOINTER, self.checkpointer
    )
    if not checkpointer:
        raise ValueError("No checkpointer set")

    # delegate to subgraph
    if (
        checkpoint_ns := config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")
    ) and CONFIG_KEY_CHECKPOINTER not in config[CONF]:
        # remove task_ids from checkpoint_ns
        recast_checkpoint_ns = NS_SEP.join(
            part.split(NS_END)[0] for part in checkpoint_ns.split(NS_SEP)
        )
        # find the subgraph with the matching name
        for _, pregel in self.get_subgraphs(
            namespace=recast_checkpoint_ns, recurse=True
        ):
            return pregel.update_state(
                patch_configurable(config, {CONFIG_KEY_CHECKPOINTER: checkpointer}),
                values,
                as_node,
            )
        else:
            raise ValueError(f"Subgraph {recast_checkpoint_ns} not found")

    # get last checkpoint
    config = merge_configs(self.config, config) if self.config else config
    saved = checkpointer.get_tuple(config)
    checkpoint = copy_checkpoint(saved.checkpoint) if saved else empty_checkpoint()
    checkpoint_previous_versions = (
        saved.checkpoint["channel_versions"].copy() if saved else {}
    )
    step = saved.metadata.get("step", -1) if saved else -1
    # merge configurable fields with previous checkpoint config
    checkpoint_config = patch_configurable(
        config,
        {CONFIG_KEY_CHECKPOINT_NS: config[CONF].get(CONFIG_KEY_CHECKPOINT_NS, "")},
    )
    if saved:
        checkpoint_config = patch_configurable(config, saved.config[CONF])
    # find last node that updated the state, if not provided
    if values is None and as_node is None:
        next_config = checkpointer.put(
            checkpoint_config,
            create_checkpoint(checkpoint, None, step),
            {
                "source": "update",
                "step": step + 1,
                "writes": {},
                "parents": saved.metadata.get("parents", {}) if saved else {},
            },
            {},
        )
        return patch_checkpoint_map(next_config, saved.metadata if saved else None)
    elif as_node is None and not any(
        v for vv in checkpoint["versions_seen"].values() for v in vv.values()
    ):
        if (
            isinstance(self.input_channels, str)
            and self.input_channels in self.nodes
        ):
            as_node = self.input_channels
    elif as_node is None:
        last_seen_by_node = sorted(
            (v, n)
            for n, seen in checkpoint["versions_seen"].items()
            if n in self.nodes
            for v in seen.values()
        )
        # if two nodes updated the state at the same time, it's ambiguous
        if last_seen_by_node:
            if len(last_seen_by_node) == 1:
                as_node = last_seen_by_node[0][1]
            elif last_seen_by_node[-1][0] != last_seen_by_node[-2][0]:
                as_node = last_seen_by_node[-1][1]
    if as_node is None:
        raise InvalidUpdateError("Ambiguous update, specify as_node")
    if as_node not in self.nodes:
        raise InvalidUpdateError(f"Node {as_node} does not exist")
    # update channels
    with ChannelsManager(self.channels, checkpoint, config) as (
        channels,
        managed,
    ):
        # create task to run all writers of the chosen node
        writers = self.nodes[as_node].flat_writers
        if not writers:
            raise InvalidUpdateError(f"Node {as_node} has no writers")
        writes: deque[tuple[str, Any]] = deque()
        task = PregelTaskWrites(as_node, writes, [INTERRUPT])
        task_id = str(uuid5(UUID(checkpoint["id"]), INTERRUPT))
        run = RunnableSequence(*writers) if len(writers) > 1 else writers[0]
        # execute task
        run.invoke(
            values,
            patch_config(
                config,
                run_name=self.name + "UpdateState",
                configurable={
                    # deque.extend is thread-safe
                    CONFIG_KEY_SEND: partial(
                        local_write,
                        writes.extend,
                        self.nodes.keys(),
                    ),
                    CONFIG_KEY_READ: partial(
                        local_read,
                        step + 1,
                        checkpoint,
                        channels,
                        managed,
                        task,
                        config,
                    ),
                },
            ),
        )
        # save task writes
        if saved:
            checkpointer.put_writes(checkpoint_config, task.writes, task_id)
        # apply to checkpoint and save
        assert not apply_writes(
            checkpoint, channels, [task], checkpointer.get_next_version
        ), "Can't write to SharedValues from update_state"
        checkpoint = create_checkpoint(checkpoint, channels, step + 1)
        next_config = checkpointer.put(
            checkpoint_config,
            checkpoint,
            {
                "source": "update",
                "step": step + 1,
                "writes": {as_node: values},
                "parents": saved.metadata.get("parents", {}) if saved else {},
            },
            get_new_channel_versions(
                checkpoint_previous_versions, checkpoint["channel_versions"]
            ),
        )
        return patch_checkpoint_map(next_config, saved.metadata if saved else None)

stream(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None, output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, subgraphs: bool = False) -> Iterator[Union[dict[str, Any], Any]]

为单个输入流式传输图步骤。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入。

• config (Optional[RunnableConfig], default: None ) –

  用于运行的配置。

• stream_mode (Optional[Union[StreamMode, list[StreamMode]]], default: None ) –

  流式传输输出的模式，默认为 self.stream_mode。选项包括“values”、“updates”和“debug”。values：为每一步发出状态的当前值。updates：仅为每一步发出状态的更新。输出是一个字典，节点名称作为键，更新后的值作为值。debug：为每一步发出调试事件。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  要流式传输的键，默认为所有非上下文通道。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之前中断的节点，默认为图中的所有节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之后中断的节点，默认为图中的所有节点。

• debug (Optional[bool], default: None ) –

  是否在执行期间打印调试信息，默认为 False。

• subgraphs (bool, default: False ) –

  是否流式传输子图，默认为 False。

Yields

• Union[dict[str, Any], Any] –

  图中每一步的输出。输出形状取决于 stream_mode。

Examples

使用不同的 stream 模式和一个图

>>> import operator
>>> from typing_extensions import Annotated, TypedDict
>>> from langgraph.graph import StateGraph
>>> from langgraph.constants import START
...
>>> class State(TypedDict):
...     alist: Annotated[list, operator.add]
...     another_list: Annotated[list, operator.add]
...
>>> builder = StateGraph(State)
>>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
>>> builder.add_node("b", lambda _state: {"alist": ["there"]})
>>> builder.add_edge("a", "b")
>>> builder.add_edge(START, "a")
>>> graph = builder.compile()

使用 stream_mode="values"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
...     print(event)
{'alist': ['Ex for stream_mode="values"'], 'another_list': []}
{'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
{'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}

使用 stream_mode="updates"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
...     print(event)
{'a': {'another_list': ['hi']}}
{'b': {'alist': ['there']}}

使用 stream_mode="debug"

>>> for event in graph.stream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
...     print(event)
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def stream(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None,
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    subgraphs: bool = False,
) -> Iterator[Union[dict[str, Any], Any]]:
    """Stream graph steps for a single input.

    Args:
        input: The input to the graph.
        config: The configuration to use for the run.
        stream_mode: The mode to stream output, defaults to self.stream_mode.
            Options are 'values', 'updates', and 'debug'.
            values: Emit the current values of the state for each step.
            updates: Emit only the updates to the state for each step.
                Output is a dict with the node name as key and the updated values as value.
            debug: Emit debug events for each step.
        output_keys: The keys to stream, defaults to all non-context channels.
        interrupt_before: Nodes to interrupt before, defaults to all nodes in the graph.
        interrupt_after: Nodes to interrupt after, defaults to all nodes in the graph.
        debug: Whether to print debug information during execution, defaults to False.
        subgraphs: Whether to stream subgraphs, defaults to False.

    Yields:
        The output of each step in the graph. The output shape depends on the stream_mode.

    Examples:
        Using different stream modes with a graph:
        ```pycon
        >>> import operator
        >>> from typing_extensions import Annotated, TypedDict
        >>> from langgraph.graph import StateGraph
        >>> from langgraph.constants import START
        ...
        >>> class State(TypedDict):
        ...     alist: Annotated[list, operator.add]
        ...     another_list: Annotated[list, operator.add]
        ...
        >>> builder = StateGraph(State)
        >>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
        >>> builder.add_node("b", lambda _state: {"alist": ["there"]})
        >>> builder.add_edge("a", "b")
        >>> builder.add_edge(START, "a")
        >>> graph = builder.compile()
        ```
        With stream_mode="values":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
        ...     print(event)
        {'alist': ['Ex for stream_mode="values"'], 'another_list': []}
        {'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
        {'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}
        ```
        With stream_mode="updates":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
        ...     print(event)
        {'a': {'another_list': ['hi']}}
        {'b': {'alist': ['there']}}
        ```
        With stream_mode="debug":

        ```pycon
        >>> for event in graph.stream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
        ...     print(event)
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}
        ```
    """

    stream = SyncQueue()

    def output() -> Iterator:
        while True:
            try:
                ns, mode, payload = stream.get(block=False)
            except queue.Empty:
                break
            if subgraphs and isinstance(stream_mode, list):
                yield (ns, mode, payload)
            elif isinstance(stream_mode, list):
                yield (mode, payload)
            elif subgraphs:
                yield (ns, payload)
            else:
                yield payload

    config = ensure_config(self.config, config)
    callback_manager = get_callback_manager_for_config(config)
    run_manager = callback_manager.on_chain_start(
        None,
        input,
        name=config.get("run_name", self.get_name()),
        run_id=config.get("run_id"),
    )
    try:
        # assign defaults
        (
            debug,
            stream_modes,
            output_keys,
            interrupt_before_,
            interrupt_after_,
            checkpointer,
            store,
        ) = self._defaults(
            config,
            stream_mode=stream_mode,
            output_keys=output_keys,
            interrupt_before=interrupt_before,
            interrupt_after=interrupt_after,
            debug=debug,
        )
        # set up messages stream mode
        if "messages" in stream_modes:
            run_manager.inheritable_handlers.append(
                StreamMessagesHandler(stream.put)
            )
        # set up custom stream mode
        if "custom" in stream_modes:
            config[CONF][CONFIG_KEY_STREAM_WRITER] = lambda c: stream.put(
                ((), "custom", c)
            )
        with SyncPregelLoop(
            input,
            stream=StreamProtocol(stream.put, stream_modes),
            config=config,
            store=store,
            checkpointer=checkpointer,
            nodes=self.nodes,
            specs=self.channels,
            output_keys=output_keys,
            stream_keys=self.stream_channels_asis,
            debug=debug,
        ) as loop:
            # create runner
            runner = PregelRunner(
                submit=loop.submit,
                put_writes=loop.put_writes,
            )
            # enable subgraph streaming
            if subgraphs:
                loop.config[CONF][CONFIG_KEY_STREAM] = loop.stream
            # enable concurrent streaming
            if subgraphs or "messages" in stream_modes or "custom" in stream_modes:
                # we are careful to have a single waiter live at any one time
                # because on exit we increment semaphore count by exactly 1
                waiter: Optional[concurrent.futures.Future] = None
                # because sync futures cannot be cancelled, we instead
                # release the stream semaphore on exit, which will cause
                # a pending waiter to return immediately
                loop.stack.callback(stream._count.release)

                def get_waiter() -> concurrent.futures.Future[None]:
                    nonlocal waiter
                    if waiter is None or waiter.done():
                        waiter = loop.submit(stream.wait)
                        return waiter
                    else:
                        return waiter
            else:
                get_waiter = None  # type: ignore[assignment]
            # Similarly to Bulk Synchronous Parallel / Pregel model
            # computation proceeds in steps, while there are channel updates
            # channel updates from step N are only visible in step N+1
            # channels are guaranteed to be immutable for the duration of the step,
            # with channel updates applied only at the transition between steps
            while loop.tick(
                input_keys=self.input_channels,
                interrupt_before=interrupt_before_,
                interrupt_after=interrupt_after_,
                manager=run_manager,
            ):
                for _ in runner.tick(
                    loop.tasks.values(),
                    timeout=self.step_timeout,
                    retry_policy=self.retry_policy,
                    get_waiter=get_waiter,
                ):
                    # emit output
                    yield from output()
        # emit output
        yield from output()
        # handle exit
        if loop.status == "out_of_steps":
            raise GraphRecursionError(
                f"Recursion limit of {config['recursion_limit']} reached "
                "without hitting a stop condition. You can increase the "
                "limit by setting the `recursion_limit` config key."
            )
        # set final channel values as run output
        run_manager.on_chain_end(loop.output)
    except BaseException as e:
        run_manager.on_chain_error(e)
        raise

astream(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None, output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, subgraphs: bool = False) -> AsyncIterator[Union[dict[str, Any], Any]] 异步

为单个输入流式传输图步骤。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入。

• config (Optional[RunnableConfig], default: None ) –

  用于运行的配置。

• stream_mode (Optional[Union[StreamMode, list[StreamMode]]], default: None ) –

  流式传输输出的模式，默认为 self.stream_mode。选项包括“values”、“updates”和“debug”。values：为每一步发出状态的当前值。updates：仅为每一步发出状态的更新。输出是一个字典，节点名称作为键，更新后的值作为值。debug：为每一步发出调试事件。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  要流式传输的键，默认为所有非上下文通道。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之前中断的节点，默认为图中的所有节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  要在其之后中断的节点，默认为图中的所有节点。

• debug (Optional[bool], default: None ) –

  是否在执行期间打印调试信息，默认为 False。

• subgraphs (bool, default: False ) –

  是否流式传输子图，默认为 False。

Yields

• AsyncIterator[Union[dict[str, Any], Any]] –

  图中每一步的输出。输出形状取决于 stream_mode。

Examples

使用不同的 stream 模式和一个图

>>> import operator
>>> from typing_extensions import Annotated, TypedDict
>>> from langgraph.graph import StateGraph
>>> from langgraph.constants import START
...
>>> class State(TypedDict):
...     alist: Annotated[list, operator.add]
...     another_list: Annotated[list, operator.add]
...
>>> builder = StateGraph(State)
>>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
>>> builder.add_node("b", lambda _state: {"alist": ["there"]})
>>> builder.add_edge("a", "b")
>>> builder.add_edge(START, "a")
>>> graph = builder.compile()

使用 stream_mode="values"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
...     print(event)
{'alist': ['Ex for stream_mode="values"'], 'another_list': []}
{'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
{'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}

使用 stream_mode="updates"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
...     print(event)
{'a': {'another_list': ['hi']}}
{'b': {'alist': ['there']}}

使用 stream_mode="debug"

>>> async for event in graph.astream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
...     print(event)
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
{'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
{'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def astream(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: Optional[Union[StreamMode, list[StreamMode]]] = None,
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    subgraphs: bool = False,
) -> AsyncIterator[Union[dict[str, Any], Any]]:
    """Stream graph steps for a single input.

    Args:
        input: The input to the graph.
        config: The configuration to use for the run.
        stream_mode: The mode to stream output, defaults to self.stream_mode.
            Options are 'values', 'updates', and 'debug'.
            values: Emit the current values of the state for each step.
            updates: Emit only the updates to the state for each step.
                Output is a dict with the node name as key and the updated values as value.
            debug: Emit debug events for each step.
        output_keys: The keys to stream, defaults to all non-context channels.
        interrupt_before: Nodes to interrupt before, defaults to all nodes in the graph.
        interrupt_after: Nodes to interrupt after, defaults to all nodes in the graph.
        debug: Whether to print debug information during execution, defaults to False.
        subgraphs: Whether to stream subgraphs, defaults to False.

    Yields:
        The output of each step in the graph. The output shape depends on the stream_mode.

    Examples:
        Using different stream modes with a graph:
        ```pycon
        >>> import operator
        >>> from typing_extensions import Annotated, TypedDict
        >>> from langgraph.graph import StateGraph
        >>> from langgraph.constants import START
        ...
        >>> class State(TypedDict):
        ...     alist: Annotated[list, operator.add]
        ...     another_list: Annotated[list, operator.add]
        ...
        >>> builder = StateGraph(State)
        >>> builder.add_node("a", lambda _state: {"another_list": ["hi"]})
        >>> builder.add_node("b", lambda _state: {"alist": ["there"]})
        >>> builder.add_edge("a", "b")
        >>> builder.add_edge(START, "a")
        >>> graph = builder.compile()
        ```
        With stream_mode="values":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="values"']}, stream_mode="values"):
        ...     print(event)
        {'alist': ['Ex for stream_mode="values"'], 'another_list': []}
        {'alist': ['Ex for stream_mode="values"'], 'another_list': ['hi']}
        {'alist': ['Ex for stream_mode="values"', 'there'], 'another_list': ['hi']}
        ```
        With stream_mode="updates":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="updates"']}, stream_mode="updates"):
        ...     print(event)
        {'a': {'another_list': ['hi']}}
        {'b': {'alist': ['there']}}
        ```
        With stream_mode="debug":

        ```pycon
        >>> async for event in graph.astream({"alist": ['Ex for stream_mode="debug"']}, stream_mode="debug"):
        ...     print(event)
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': []}, 'triggers': ['start:a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 1, 'payload': {'id': '...', 'name': 'a', 'result': [('another_list', ['hi'])]}}
        {'type': 'task', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'input': {'alist': ['Ex for stream_mode="debug"'], 'another_list': ['hi']}, 'triggers': ['a']}}
        {'type': 'task_result', 'timestamp': '2024-06-23T...+00:00', 'step': 2, 'payload': {'id': '...', 'name': 'b', 'result': [('alist', ['there'])]}}
        ```
    """

    stream = AsyncQueue()
    aioloop = asyncio.get_running_loop()

    def output() -> Iterator:
        while True:
            try:
                ns, mode, payload = stream.get_nowait()
            except asyncio.QueueEmpty:
                break
            if subgraphs and isinstance(stream_mode, list):
                yield (ns, mode, payload)
            elif isinstance(stream_mode, list):
                yield (mode, payload)
            elif subgraphs:
                yield (ns, payload)
            else:
                yield payload

    config = ensure_config(self.config, config)
    callback_manager = get_async_callback_manager_for_config(config)
    run_manager = await callback_manager.on_chain_start(
        None,
        input,
        name=config.get("run_name", self.get_name()),
        run_id=config.get("run_id"),
    )
    # if running from astream_log() run each proc with streaming
    do_stream = next(
        (
            cast(_StreamingCallbackHandler, h)
            for h in run_manager.handlers
            if isinstance(h, _StreamingCallbackHandler)
        ),
        None,
    )
    try:
        # assign defaults
        (
            debug,
            stream_modes,
            output_keys,
            interrupt_before_,
            interrupt_after_,
            checkpointer,
            store,
        ) = self._defaults(
            config,
            stream_mode=stream_mode,
            output_keys=output_keys,
            interrupt_before=interrupt_before,
            interrupt_after=interrupt_after,
            debug=debug,
        )
        # set up messages stream mode
        if "messages" in stream_modes:
            run_manager.inheritable_handlers.append(
                StreamMessagesHandler(stream.put_nowait)
            )
        # set up custom stream mode
        if "custom" in stream_modes:
            config[CONF][CONFIG_KEY_STREAM_WRITER] = lambda c: stream.put_nowait(
                ((), "custom", c)
            )
        async with AsyncPregelLoop(
            input,
            stream=StreamProtocol(stream.put_nowait, stream_modes),
            config=config,
            store=store,
            checkpointer=checkpointer,
            nodes=self.nodes,
            specs=self.channels,
            output_keys=output_keys,
            stream_keys=self.stream_channels_asis,
        ) as loop:
            # create runner
            runner = PregelRunner(
                submit=loop.submit,
                put_writes=loop.put_writes,
                use_astream=do_stream is not None,
            )
            # enable subgraph streaming
            if subgraphs:
                loop.config[CONF][CONFIG_KEY_STREAM] = loop.stream
            # enable concurrent streaming
            if subgraphs or "messages" in stream_modes or "custom" in stream_modes:

                def get_waiter() -> asyncio.Task[None]:
                    return aioloop.create_task(stream.wait())
            else:
                get_waiter = None  # type: ignore[assignment]
            # Similarly to Bulk Synchronous Parallel / Pregel model
            # computation proceeds in steps, while there are channel updates
            # channel updates from step N are only visible in step N+1
            # channels are guaranteed to be immutable for the duration of the step,
            # with channel updates applied only at the transition between steps
            while loop.tick(
                input_keys=self.input_channels,
                interrupt_before=interrupt_before_,
                interrupt_after=interrupt_after_,
                manager=run_manager,
            ):
                async for _ in runner.atick(
                    loop.tasks.values(),
                    timeout=self.step_timeout,
                    retry_policy=self.retry_policy,
                    get_waiter=get_waiter,
                ):
                    # emit output
                    for o in output():
                        yield o
        # emit output
        for o in output():
            yield o
        # handle exit
        if loop.status == "out_of_steps":
            raise GraphRecursionError(
                f"Recursion limit of {config['recursion_limit']} reached "
                "without hitting a stop condition. You can increase the "
                "limit by setting the `recursion_limit` config key."
            )
        # set final channel values as run output
        await run_manager.on_chain_end(loop.output)
    except BaseException as e:
        await asyncio.shield(run_manager.on_chain_error(e))
        raise

invoke(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: StreamMode = 'values', output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, **kwargs: Any) -> Union[dict[str, Any], Any]

运行图，使用单个输入和配置。

参数

• input (Union[dict[str, Any], Any]) –

  图的输入数据。它可以是字典或任何其他类型。

• config (Optional[RunnableConfig], default: None ) –

  可选。图运行的配置。

• stream_mode (StreamMode, default: 'values' ) –

  Optional[str]。图运行的流模式。默认值为 "values"。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  可选。要从图运行中检索的输出键。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。在中断图运行之前的节点。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。在中断图运行之后的节点。

• debug (Optional[bool], default: None ) –

  可选。启用图运行的调试模式。

• **kwargs (Any, default: {} ) –

  传递给图运行的附加关键字参数。

返回值

• Union[dict[str, Any], Any] –

  图运行的输出。如果 stream_mode 为 "values"，则返回最新的输出。

• Union[dict[str, Any], Any] –

  如果 stream_mode 不是 "values"，则返回输出块的列表。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

def invoke(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: StreamMode = "values",
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    **kwargs: Any,
) -> Union[dict[str, Any], Any]:
    """Run the graph with a single input and config.

    Args:
        input: The input data for the graph. It can be a dictionary or any other type.
        config: Optional. The configuration for the graph run.
        stream_mode: Optional[str]. The stream mode for the graph run. Default is "values".
        output_keys: Optional. The output keys to retrieve from the graph run.
        interrupt_before: Optional. The nodes to interrupt the graph run before.
        interrupt_after: Optional. The nodes to interrupt the graph run after.
        debug: Optional. Enable debug mode for the graph run.
        **kwargs: Additional keyword arguments to pass to the graph run.

    Returns:
        The output of the graph run. If stream_mode is "values", it returns the latest output.
        If stream_mode is not "values", it returns a list of output chunks.
    """
    output_keys = output_keys if output_keys is not None else self.output_channels
    if stream_mode == "values":
        latest: Union[dict[str, Any], Any] = None
    else:
        chunks = []
    for chunk in self.stream(
        input,
        config,
        stream_mode=stream_mode,
        output_keys=output_keys,
        interrupt_before=interrupt_before,
        interrupt_after=interrupt_after,
        debug=debug,
        **kwargs,
    ):
        if stream_mode == "values":
            latest = chunk
        else:
            chunks.append(chunk)
    if stream_mode == "values":
        return latest
    else:
        return chunks

ainvoke(input: Union[dict[str, Any], Any], config: Optional[RunnableConfig] = None, *, stream_mode: StreamMode = 'values', output_keys: Optional[Union[str, Sequence[str]]] = None, interrupt_before: Optional[Union[All, Sequence[str]]] = None, interrupt_after: Optional[Union[All, Sequence[str]]] = None, debug: Optional[bool] = None, **kwargs: Any) -> Union[dict[str, Any], Any] async

异步在单个输入上调用图。

参数

• input (Union[dict[str, Any], Any]) –

  计算的输入数据。它可以是字典或任何其他类型。

• config (Optional[RunnableConfig], default: None ) –

  可选。计算的配置。

• stream_mode (StreamMode, default: 'values' ) –

  可选。计算的流模式。默认值为 "values"。

• output_keys (Optional[Union[str, Sequence[str]]], default: None ) –

  可选。要包含在结果中的输出键。默认值为 None。

• interrupt_before (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。要在其之前中断的节点。默认值为 None。

• interrupt_after (Optional[Union[All, Sequence[str]]], default: None ) –

  可选。要在其之后中断的节点。默认值为 None。

• debug (Optional[bool], default: None ) –

  可选。是否启用调试模式。默认值为 None。

• **kwargs (Any, default: {} ) –

  附加关键字参数。

返回值

• Union[dict[str, Any], Any] –

  计算的结果。如果 stream_mode 为 "values"，则返回最新的值。

• Union[dict[str, Any], Any] –

  如果 stream_mode 为 "chunks"，则返回块列表。

源代码在 libs/langgraph/langgraph/pregel/__init__.py

async def ainvoke(
    self,
    input: Union[dict[str, Any], Any],
    config: Optional[RunnableConfig] = None,
    *,
    stream_mode: StreamMode = "values",
    output_keys: Optional[Union[str, Sequence[str]]] = None,
    interrupt_before: Optional[Union[All, Sequence[str]]] = None,
    interrupt_after: Optional[Union[All, Sequence[str]]] = None,
    debug: Optional[bool] = None,
    **kwargs: Any,
) -> Union[dict[str, Any], Any]:
    """Asynchronously invoke the graph on a single input.

    Args:
        input: The input data for the computation. It can be a dictionary or any other type.
        config: Optional. The configuration for the computation.
        stream_mode: Optional. The stream mode for the computation. Default is "values".
        output_keys: Optional. The output keys to include in the result. Default is None.
        interrupt_before: Optional. The nodes to interrupt before. Default is None.
        interrupt_after: Optional. The nodes to interrupt after. Default is None.
        debug: Optional. Whether to enable debug mode. Default is None.
        **kwargs: Additional keyword arguments.

    Returns:
        The result of the computation. If stream_mode is "values", it returns the latest value.
        If stream_mode is "chunks", it returns a list of chunks.
    """

    output_keys = output_keys if output_keys is not None else self.output_channels
    if stream_mode == "values":
        latest: Union[dict[str, Any], Any] = None
    else:
        chunks = []
    async for chunk in self.astream(
        input,
        config,
        stream_mode=stream_mode,
        output_keys=output_keys,
        interrupt_before=interrupt_before,
        interrupt_after=interrupt_after,
        debug=debug,
        **kwargs,
    ):
        if stream_mode == "values":
            latest = chunk
        else:
            chunks.append(chunk)
    if stream_mode == "values":
        return latest
    else:
        return chunks

get_graph(config: Optional[RunnableConfig] = None, *, xray: Union[int, bool] = False) -> DrawableGraph

返回计算图的可绘制表示形式。

源代码在 libs/langgraph/langgraph/graph/graph.py 中

def get_graph(
    self,
    config: Optional[RunnableConfig] = None,
    *,
    xray: Union[int, bool] = False,
) -> DrawableGraph:
    """Returns a drawable representation of the computation graph."""
    graph = DrawableGraph()
    start_nodes: dict[str, DrawableNode] = {
        START: graph.add_node(self.get_input_schema(config), START)
    }
    end_nodes: dict[str, DrawableNode] = {}
    if xray:
        subgraphs = {
            k: v for k, v in self.get_subgraphs() if isinstance(v, CompiledGraph)
        }
    else:
        subgraphs = {}

    def add_edge(
        start: str,
        end: str,
        label: Optional[Hashable] = None,
        conditional: bool = False,
    ) -> None:
        if end == END and END not in end_nodes:
            end_nodes[END] = graph.add_node(self.get_output_schema(config), END)
        return graph.add_edge(
            start_nodes[start],
            end_nodes[end],
            str(label) if label is not None else None,
            conditional,
        )

    for key, n in self.builder.nodes.items():
        node = n.runnable
        metadata = n.metadata or {}
        if key in self.interrupt_before_nodes and key in self.interrupt_after_nodes:
            metadata["__interrupt"] = "before,after"
        elif key in self.interrupt_before_nodes:
            metadata["__interrupt"] = "before"
        elif key in self.interrupt_after_nodes:
            metadata["__interrupt"] = "after"
        if xray:
            subgraph = (
                subgraphs[key].get_graph(
                    config=config,
                    xray=xray - 1
                    if isinstance(xray, int)
                    and not isinstance(xray, bool)
                    and xray > 0
                    else xray,
                )
                if key in subgraphs
                else node.get_graph(config=config)
            )
            subgraph.trim_first_node()
            subgraph.trim_last_node()
            if len(subgraph.nodes) > 1:
                e, s = graph.extend(subgraph, prefix=key)
                if e is None:
                    raise ValueError(
                        f"Could not extend subgraph '{key}' due to missing entrypoint"
                    )
                if s is not None:
                    start_nodes[key] = s
                end_nodes[key] = e
            else:
                nn = graph.add_node(node, key, metadata=metadata or None)
                start_nodes[key] = nn
                end_nodes[key] = nn
        else:
            nn = graph.add_node(node, key, metadata=metadata or None)
            start_nodes[key] = nn
            end_nodes[key] = nn
    for start, end in sorted(self.builder._all_edges):
        add_edge(start, end)
    for start, branches in self.builder.branches.items():
        default_ends = {
            **{k: k for k in self.builder.nodes if k != start},
            END: END,
        }
        for _, branch in branches.items():
            if branch.ends is not None:
                ends = branch.ends
            elif branch.then is not None:
                ends = {k: k for k in default_ends if k not in (END, branch.then)}
            else:
                ends = cast(dict[Hashable, str], default_ends)
            for label, end in ends.items():
                add_edge(
                    start,
                    end,
                    label if label != end else None,
                    conditional=True,
                )
                if branch.then is not None:
                    add_edge(end, branch.then)

    return graph

add_messages(left: Messages, right: Messages) -> Messages

将两个消息列表合并，根据 ID 更新现有消息。

默认情况下，这将确保状态为“仅追加”，除非新消息与现有消息具有相同的 ID。

参数

• left (Messages) –

  基本消息列表。

• right (Messages) –

  要合并到基本列表中的消息列表（或单个消息）。

返回值

• Messages –

  一个新的消息列表，其中 right 中的消息已合并到 left 中。

• Messages –

  如果 right 中的消息与 left 中的消息具有相同的 ID，则

• Messages –

  来自 right 的消息将替换来自 left 的消息。

Examples

>>> from langchain_core.messages import AIMessage, HumanMessage
>>> msgs1 = [HumanMessage(content="Hello", id="1")]
>>> msgs2 = [AIMessage(content="Hi there!", id="2")]
>>> add_messages(msgs1, msgs2)
[HumanMessage(content='Hello', id='1'), AIMessage(content='Hi there!', id='2')]

>>> msgs1 = [HumanMessage(content="Hello", id="1")]
>>> msgs2 = [HumanMessage(content="Hello again", id="1")]
>>> add_messages(msgs1, msgs2)
[HumanMessage(content='Hello again', id='1')]

>>> from typing import Annotated
>>> from typing_extensions import TypedDict
>>> from langgraph.graph import StateGraph
>>>
>>> class State(TypedDict):
...     messages: Annotated[list, add_messages]
...
>>> builder = StateGraph(State)
>>> builder.add_node("chatbot", lambda state: {"messages": [("assistant", "Hello")]})
>>> builder.set_entry_point("chatbot")
>>> builder.set_finish_point("chatbot")
>>> graph = builder.compile()
>>> graph.invoke({})
{'messages': [AIMessage(content='Hello', id=...)]}

源代码位于 libs/langgraph/langgraph/graph/message.py 中

def add_messages(left: Messages, right: Messages) -> Messages:
    """Merges two lists of messages, updating existing messages by ID.

    By default, this ensures the state is "append-only", unless the
    new message has the same ID as an existing message.

    Args:
        left: The base list of messages.
        right: The list of messages (or single message) to merge
            into the base list.

    Returns:
        A new list of messages with the messages from `right` merged into `left`.
        If a message in `right` has the same ID as a message in `left`, the
        message from `right` will replace the message from `left`.

    Examples:
        ```pycon
        >>> from langchain_core.messages import AIMessage, HumanMessage
        >>> msgs1 = [HumanMessage(content="Hello", id="1")]
        >>> msgs2 = [AIMessage(content="Hi there!", id="2")]
        >>> add_messages(msgs1, msgs2)
        [HumanMessage(content='Hello', id='1'), AIMessage(content='Hi there!', id='2')]

        >>> msgs1 = [HumanMessage(content="Hello", id="1")]
        >>> msgs2 = [HumanMessage(content="Hello again", id="1")]
        >>> add_messages(msgs1, msgs2)
        [HumanMessage(content='Hello again', id='1')]

        >>> from typing import Annotated
        >>> from typing_extensions import TypedDict
        >>> from langgraph.graph import StateGraph
        >>>
        >>> class State(TypedDict):
        ...     messages: Annotated[list, add_messages]
        ...
        >>> builder = StateGraph(State)
        >>> builder.add_node("chatbot", lambda state: {"messages": [("assistant", "Hello")]})
        >>> builder.set_entry_point("chatbot")
        >>> builder.set_finish_point("chatbot")
        >>> graph = builder.compile()
        >>> graph.invoke({})
        {'messages': [AIMessage(content='Hello', id=...)]}
        ```

    """
    # coerce to list
    if not isinstance(left, list):
        left = [left]  # type: ignore[assignment]
    if not isinstance(right, list):
        right = [right]  # type: ignore[assignment]
    # coerce to message
    left = [
        message_chunk_to_message(cast(BaseMessageChunk, m))
        for m in convert_to_messages(left)
    ]
    right = [
        message_chunk_to_message(cast(BaseMessageChunk, m))
        for m in convert_to_messages(right)
    ]
    # assign missing ids
    for m in left:
        if m.id is None:
            m.id = str(uuid.uuid4())
    for m in right:
        if m.id is None:
            m.id = str(uuid.uuid4())
    # merge
    left_idx_by_id = {m.id: i for i, m in enumerate(left)}
    merged = left.copy()
    ids_to_remove = set()
    for m in right:
        if (existing_idx := left_idx_by_id.get(m.id)) is not None:
            if isinstance(m, RemoveMessage):
                ids_to_remove.add(m.id)
            else:
                merged[existing_idx] = m
        else:
            if isinstance(m, RemoveMessage):
                raise ValueError(
                    f"Attempting to delete a message with an ID that doesn't exist ('{m.id}')"
                )

            merged.append(m)
    merged = [m for m in merged if m.id not in ids_to_remove]
    return merged

注释