langchain-ai/langgraphjs
1
0
Fork 0
mirror of https://github.com/langchain-ai/langgraphjs.git synced 2026-07-21 16:45:24 -04:00
Code Issues 108 Packages Projects Releases 20 Wiki Activity

Send API serialize/deserialize issue #247

New Issue
Open
opened 2026-02-15 18:15:01 -05:00 by yindo · 3 comments
yindo commented 2026-02-15 18:15:01 -05:00
Owner
Copy Link
Copy Source

Originally created by @brettshollenberger on GitHub (Apr 30, 2025).

Originally assigned to: @benjamincburns, @dqbd on GitHub.

Hey all, just noticed an issue with serializing/deserializing objects that are wrapped with some custom class when using the Send API, for example BaseMessage, HumanMessage, etc.

  • For contrast, return a BaseMessage from a regular node, and it will stay wrapped as a BaseMessage
  • But if you queue the state to separate workers using Send, the workers will receive an unwrapped (plain object) with the data.
  • This is easy enough to re-wrap as BaseMessage, but it causes some unexpected interactions with things like createReactAgent, which expect the MessageState interface to be applied (which fails when not wrapped as BaseMessage).

Minimal example:

export interface GraphState {
    userRequest: BaseMessage; 
    chatHistory: BaseMessage[];
    task: Task;
    queue: Task[];
    completedTasks: Task[];
}

const queueTasks = async(state: GraphState) => {
    // in here, state.userRequest and state.chatHistory are fine
    if (!state.queue) {
        return [];
    }
    const queue = state.queue;
    return queue.map(task => new Send("executeTasks", { ...state, task }));
}

const planTasks = async(state: GraphState): Promise<Partial<GraphState>> => {
  //   in here, state.userRequest and state.chatHistory are fine  
  const tasks = [ ... ] // invent some tasks
  return { queue: tasks }
}

const executeTask = async(state: GraphState): Promise<Partial<GraphState>> => {
  // in here, state.userRequest and state.chatHistory are unwrapped/plain objects
  const agent =  createReactAgent(... stuff);
    const agentState = {
    messages: [
      ...(state.chatHistory.map((message) => { return new BaseMessage(message); },
      new HumanMessage(state.userRequest.content) // If we don't do this, createReactAgent will complain
    ]
  }
  agent.invoke(agentState); // This would explode if we don't re-wrap our messages after Send API
}

export const graph = new StateGraph(GraphAnnotation, ConfigurationAnnotation)
    .addNode("planTasks", planTasks)
    .addNode("executeTask", executeTask)
    .addEdge(START, "planTasks")
    .addConditionalEdges("planTasks", queueTasks)
    .addEdge("executeTasks", END)

Just logging the issue, thank you!

Originally created by @brettshollenberger on GitHub (Apr 30, 2025). Originally assigned to: @benjamincburns, @dqbd on GitHub. Hey all, just noticed an issue with serializing/deserializing objects that are wrapped with some custom class when using the `Send` API, for example `BaseMessage`, `HumanMessage`, etc. * For contrast, return a `BaseMessage` from a regular node, and it will stay wrapped as a `BaseMessage` * But if you queue the state to separate workers using `Send`, the workers will receive an unwrapped (plain object) with the data. * This is easy enough to re-wrap as `BaseMessage`, but it causes some unexpected interactions with things like `createReactAgent`, which expect the `MessageState` interface to be applied (which fails when not wrapped as `BaseMessage`). Minimal example: ```typescript export interface GraphState { userRequest: BaseMessage; chatHistory: BaseMessage[]; task: Task; queue: Task[]; completedTasks: Task[]; } const queueTasks = async(state: GraphState) => { // in here, state.userRequest and state.chatHistory are fine if (!state.queue) { return []; } const queue = state.queue; return queue.map(task => new Send("executeTasks", { ...state, task })); } const planTasks = async(state: GraphState): Promise<Partial<GraphState>> => { // in here, state.userRequest and state.chatHistory are fine const tasks = [ ... ] // invent some tasks return { queue: tasks } } const executeTask = async(state: GraphState): Promise<Partial<GraphState>> => { // in here, state.userRequest and state.chatHistory are unwrapped/plain objects const agent = createReactAgent(... stuff); const agentState = { messages: [ ...(state.chatHistory.map((message) => { return new BaseMessage(message); }, new HumanMessage(state.userRequest.content) // If we don't do this, createReactAgent will complain ] } agent.invoke(agentState); // This would explode if we don't re-wrap our messages after Send API } export const graph = new StateGraph(GraphAnnotation, ConfigurationAnnotation) .addNode("planTasks", planTasks) .addNode("executeTask", executeTask) .addEdge(START, "planTasks") .addConditionalEdges("planTasks", queueTasks) .addEdge("executeTasks", END) ``` Just logging the issue, thank you!
yindo added the bug label 2026-02-15 18:15:01 -05:00
yindo commented 2026-02-15 18:15:03 -05:00
Author
Owner
Copy Link
Copy Source

@brettshollenberger commented on GitHub (May 7, 2025):

This seems to be related, but when Sending arrays of messages, I can't get them to serialize properly. When sending a single message, I can send successfully and then unwrap, but when sending messages: [], it blows up when trying to send.

    Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/MESSAGE_COERCION_FAILURE/

        at _constructMessageFromParams (file:///Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/core/dist/messages/utils.js:109:47)
        at coerceMessageLikeToMessage (file:///Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/core/dist/messages/utils.js:129:16)
        at Array.map (<anonymous>)
        at BinaryOperatorAggregate.messagesStateReducer (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/graph/message.ts:29:59)
        at BinaryOperatorAggregate.update (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/channels/binop.ts:57:27)
        at _applyWrites (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/algo.ts:340:38)
        at PregelLoop.tick (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/loop.ts:554:34)
        at CompiledStateGraph._runLoop (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/index.ts:2040:20)
        at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
        at createAndRunLoop (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/index.ts:1939:9) {
      lc_error_code: 'MESSAGE_COERCION_FAILURE',
      pregelTaskId: '1e24419c-1e62-55ec-aff7-a1f82fe9ff84'
    },
    Error: Unable to coerce message from array: only human, AI, system, developer, or tool message coercion is currently supported.

    Received: {
      "lc_serializable": true,
      "lc_kwargs": {
        "content": "# content here",
        "additional_kwargs": {
          "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G",
          "type": "message",
          "role": "assistant",
          "model": "claude-3-5-haiku-20241022",
          "stop_reason": "end_turn",
          "stop_sequence": null,
          "usage": {
            "input_tokens": 201,
            "cache_creation_input_tokens": 0,
            "cache_read_input_tokens": 0,
            "output_tokens": 435
          }
        },
        "usage_metadata": {
          "input_tokens": 201,
          "output_tokens": 435,
          "total_tokens": 636,
          "input_token_details": {
            "cache_creation": 0,
            "cache_read": 0
          }
        },
        "response_metadata": {
          "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G",
          "model": "claude-3-5-haiku-20241022",
          "stop_reason": "end_turn",
          "stop_sequence": null,
          "usage": {
            "input_tokens": 201,
            "cache_creation_input_tokens": 0,
            "cache_read_input_tokens": 0,
            "output_tokens": 435
          },
          "type": "message",
          "role": "assistant"
        },
        "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G",
        "tool_calls": [],
        "invalid_tool_calls": []
      },
      "lc_namespace": [
        "langchain_core",
        "messages"
      ],
@brettshollenberger commented on GitHub (May 7, 2025): This seems to be related, but when `Send`ing arrays of messages, I can't get them to serialize properly. When sending a single message, I can send successfully and then unwrap, but when sending `messages: []`, it blows up when trying to send. ```tsx Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/MESSAGE_COERCION_FAILURE/ at _constructMessageFromParams (file:///Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/core/dist/messages/utils.js:109:47) at coerceMessageLikeToMessage (file:///Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/core/dist/messages/utils.js:129:16) at Array.map (<anonymous>) at BinaryOperatorAggregate.messagesStateReducer (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/graph/message.ts:29:59) at BinaryOperatorAggregate.update (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/channels/binop.ts:57:27) at _applyWrites (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/algo.ts:340:38) at PregelLoop.tick (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/loop.ts:554:34) at CompiledStateGraph._runLoop (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/index.ts:2040:20) at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at createAndRunLoop (/Users/brettshollenberger/programming/business/tools/bolt_me/node_modules/@langchain/langgraph/src/pregel/index.ts:1939:9) { lc_error_code: 'MESSAGE_COERCION_FAILURE', pregelTaskId: '1e24419c-1e62-55ec-aff7-a1f82fe9ff84' }, Error: Unable to coerce message from array: only human, AI, system, developer, or tool message coercion is currently supported. Received: { "lc_serializable": true, "lc_kwargs": { "content": "# content here", "additional_kwargs": { "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G", "type": "message", "role": "assistant", "model": "claude-3-5-haiku-20241022", "stop_reason": "end_turn", "stop_sequence": null, "usage": { "input_tokens": 201, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0, "output_tokens": 435 } }, "usage_metadata": { "input_tokens": 201, "output_tokens": 435, "total_tokens": 636, "input_token_details": { "cache_creation": 0, "cache_read": 0 } }, "response_metadata": { "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G", "model": "claude-3-5-haiku-20241022", "stop_reason": "end_turn", "stop_sequence": null, "usage": { "input_tokens": 201, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0, "output_tokens": 435 }, "type": "message", "role": "assistant" }, "id": "msg_01EkxM5kYzR5yMDVMCAYEn2G", "tool_calls": [], "invalid_tool_calls": [] }, "lc_namespace": [ "langchain_core", "messages" ], ```
yindo commented 2026-02-15 18:15:03 -05:00
Author
Owner
Copy Link
Copy Source

@brettshollenberger commented on GitHub (May 7, 2025):

In terms of minimal example for this later one:

export interface GraphState {
    messages: BaseMessage[];
    queue: string[];
    task: string;
}

export const GraphAnnotation = Annotation.Root({
    messages: Annotation<BaseMessage[], BaseMessageLike[]>({ 
        default: () => [],
        reducer: messagesStateReducer
    })
})

const queueTasks = async(state: GraphState) => {
    const queue = state.queue;
    return queue.map(task => new Send("executeTasks", { ...state, task }));
}

The queue tasks function will give this last message

@brettshollenberger commented on GitHub (May 7, 2025): In terms of minimal example for this later one: ``` export interface GraphState { messages: BaseMessage[]; queue: string[]; task: string; } export const GraphAnnotation = Annotation.Root({ messages: Annotation<BaseMessage[], BaseMessageLike[]>({ default: () => [], reducer: messagesStateReducer }) }) const queueTasks = async(state: GraphState) => { const queue = state.queue; return queue.map(task => new Send("executeTasks", { ...state, task })); } ``` The queue tasks function will give this last message
yindo commented 2026-02-15 18:15:03 -05:00
Author
Owner
Copy Link
Copy Source

@yzhaoa commented on GitHub (Sep 25, 2025):

Just ran into this issue, got the exact same unable to coerce message ... error trying to use a message state passed through Send.

Looking through the codebase, the problem seems to be https://github.com/langchain-ai/langgraphjs/blob/d2b51aae9c900a5b36289cc8eb321c9f06357da8/libs/langgraph/src/constants.ts#L180, and particularly the recursive object reconstruction in https://github.com/langchain-ai/langgraphjs/blob/d2b51aae9c900a5b36289cc8eb321c9f06357da8/libs/langgraph/src/constants.ts#L479-L517.

This was partially fixed in 5fef2ed73a for lc_serializable objects, which includes BaseMessage. While I haven't tested this yet, I'm still using an older version of langgraphjs, so I suspect this bug was filed before the fix was installed.

The reason I say that this is partially fixed though, is that the _deserializeCommandSendObjectGraph function would still indiscriminantly break other objects that are not directly marked lc_serializable. For example, Set, or Map objects which are supported by langgraph (but not langchain), are erased to {} due to having no owned properties.

Workaround

Before this is fixed more thoroughly, or perhaps closed as is because lc_serializable is working now, you can exploit the nested Command exception, i.e.:

new Send('target', new Command({update: myObject}));

to protect myObject from clobbering.

I have not tested checkpoint restoration with the above workaround yet.

And obviously, write test cases to check whether future changes have broken your workaround.

@yzhaoa commented on GitHub (Sep 25, 2025): Just ran into this issue, got the exact same unable to coerce message ... error trying to use a message state passed through `Send`. Looking through the codebase, the problem seems to be https://github.com/langchain-ai/langgraphjs/blob/d2b51aae9c900a5b36289cc8eb321c9f06357da8/libs/langgraph/src/constants.ts#L180, and particularly the recursive object reconstruction in https://github.com/langchain-ai/langgraphjs/blob/d2b51aae9c900a5b36289cc8eb321c9f06357da8/libs/langgraph/src/constants.ts#L479-L517. This was partially fixed in 5fef2ed73adf0eb49edd9ed5365977243538a02b for lc_serializable objects, which includes BaseMessage. While I haven't tested this yet, I'm still using an older version of langgraphjs, so I suspect this bug was filed before the fix was installed. The reason I say that this is partially fixed though, is that the `_deserializeCommandSendObjectGraph` function would still indiscriminantly break other objects that are not directly marked lc_serializable. For example, `Set`, or `Map` objects which are supported by langgraph (but not langchain), are erased to `{}` due to having no owned properties. ## Workaround Before this is fixed more thoroughly, or perhaps closed as is because lc_serializable is working now, you can exploit the nested `Command` exception, i.e.: ```typescript new Send('target', new Command({update: myObject})); ``` to protect `myObject` from clobbering. I have not tested checkpoint restoration with the above workaround yet. And obviously, write test cases to check whether future changes have broken your workaround.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
checkpointer
docs-needed
documentation
enhancement
good first issue
good writeup
help wanted
invalid
langgraph-supervisor
open-swe-auto
platform
prebuilt
pull-request
Mirrored from GitHub Pull Request
 question
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langchain-ai/langgraphjs#247
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?