langgraph-parallel（语言图并行）

LangGraph Parallel Execution

Run independent nodes concurrently for performance.

Fan-Out/Fan-In Pattern

from langgraph.graph import StateGraph

def fan_out(state):
    """Split work into parallel tasks."""
    state["tasks"] = [{"id": 1}, {"id": 2}, {"id": 3}]
    return state

def worker(state):
    """Process one task."""
    task = state["current_task"]
    result = process(task)
    return {"results": [result]}

def fan_in(state):
    """Combine parallel results."""
    combined = aggregate(state["results"])
    return {"final": combined}

workflow = StateGraph(State)
workflow.add_node("fan_out", fan_out)
workflow.add_node("worker", worker)
workflow.add_node("fan_in", fan_in)

workflow.add_edge("fan_out", "worker")
workflow.add_edge("worker", "fan_in")  # Waits for all workers

Using Send API

from langgraph.constants import Send

def router(state):
    """Route to multiple workers in parallel."""
    return [
        Send("worker", {"task": task})
        for task in state["tasks"]
    ]

workflow.add_conditional_edges("router", router)

Complete Send API Example (2026 Pattern)

from langgraph.graph import StateGraph, START, END
from langgraph.constants import Send
from typing import TypedDict, Annotated
from operator import add

class OverallState(TypedDict):
    subjects: list[str]
    jokes: Annotated[list[str], add]  # Accumulates from parallel branches

class JokeState(TypedDict):
    subject: str

def generate_topics(state: OverallState) -> dict:
    """Initial node: create list of subjects."""
    return {"subjects": ["cats", "dogs", "programming", "coffee"]}

def continue_to_jokes(state: OverallState) -> list[Send]:
    """Fan-out: create parallel branch for each subject."""
    return [
        Send("generate_joke", {"subject": s})
        for s in state["subjects"]
    ]

def generate_joke(state: JokeState) -> dict:
    """Worker: process one subject, return to accumulator."""
    joke = llm.invoke(f"Tell a short joke about {state['subject']}")
    return {"jokes": [f"{state['subject']}: {joke.content}"]}

# Build graph
builder = StateGraph(OverallState)
builder.add_node("generate_topics", generate_topics)
builder.add_node("generate_joke", generate_joke)

builder.add_edge(START, "generate_topics")
builder.add_conditional_edges("generate_topics", continue_to_jokes)
builder.add_edge("generate_joke", END)  # All branches converge automatically

graph = builder.compile()

# Invoke
result = graph.invoke({"subjects": [], "jokes": []})
# result["jokes"] contains all 4 jokes

Parallel Agent Analysis

from typing import Annotated
from operator import add

class AnalysisState(TypedDict):
    content: str
    findings: Annotated[list[dict], add]  # Accumulates

async def run_parallel_agents(state: AnalysisState):
    """Run multiple agents in parallel."""
    agents = [security_agent, tech_agent, quality_agent]

    # Run all concurrently
    tasks = [agent.analyze(state["content"]) for agent in agents]
    results = await asyncio.gather(*tasks, return_exceptions=True)

    # Filter successful results
    findings = [r for r in results if not isinstance(r, Exception)]

    return {"findings": findings}

Map-Reduce Pattern (True Parallel)

import asyncio

async def parallel_map(items: list, process_fn) -> list:
    """Map: Process all items concurrently."""
    tasks = [asyncio.create_task(process_fn(item)) for item in items]
    return await asyncio.gather(*tasks, return_exceptions=True)

def reduce_results(results: list) -> dict:
    """Reduce: Combine all results."""
    successes = [r for r in results if not isinstance(r, Exception)]
    failures = [r for r in results if isinstance(r, Exception)]

    return {
        "total": len(results),
        "passed": len(successes),
        "failed": len(failures),
        "results": successes,
        "errors": [str(e) for e in failures]
    }

async def map_reduce_node(state: State) -> dict:
    """Combined map-reduce in single node."""
    results = await parallel_map(state["items"], process_item_async)
    summary = reduce_results(results)
    return {"summary": summary}

# Alternative: Using Send API for true parallelism in graph
def fan_out_to_mappers(state: State) -> list[Send]:
    """Fan-out pattern for parallel map."""
    return [
        Send("mapper", {"item": item, "index": i})
        for i, item in enumerate(state["items"])
    ]

# All mappers write to accumulating state key
# Reducer runs after all mappers complete (automatic fan-in)

Error Isolation

async def parallel_with_isolation(tasks: list):
    """Run parallel tasks, isolate failures."""
    results = await asyncio.gather(*tasks, return_exceptions=True)

    successes = []
    failures = []

    for task, result in zip(tasks, results):
        if isinstance(result, Exception):
            failures.append({"task": task, "error": str(result)})
        else:
            successes.append(result)

    return {"successes": successes, "failures": failures}

Timeout per Branch

import asyncio

async def parallel_with_timeout(agents: list, content: str, timeout: int = 30):
    """Run agents with per-agent timeout."""
    async def run_with_timeout(agent):
        try:
            return await asyncio.wait_for(
                agent.analyze(content),
                timeout=timeout
            )
        except asyncio.TimeoutError:
            return {"agent": agent.name, "error": "timeout"}

    tasks = [run_with_timeout(a) for a in agents]
    return await asyncio.gather(*tasks)

Key Decisions

Common Mistakes

• No error isolation (one failure kills all)
• No timeout (one slow branch blocks)
• Sequential where parallel possible
• Forgetting to wait for all branches

Evaluations

See references/evaluations.md for test cases.

Related Skills

• langgraph-state - Accumulating state with Annotated[list, add] reducer
• langgraph-supervisor - Supervisor dispatching to parallel workers
• langgraph-subgraphs - Parallel subgraph execution
• langgraph-streaming - Stream progress from parallel branches
• langgraph-checkpoints - Checkpoint parallel execution for recovery
• multi-agent-orchestration - Higher-level coordination patterns

Capability Details

fanout-pattern

Keywords: fanout, parallel, concurrent, scatter Solves:

• Run agents in parallel
• Implement fan-out pattern
• Distribute work across workers

fanin-pattern

Keywords: fanin, gather, aggregate, collect Solves:

• Aggregate parallel results
• Implement fan-in pattern
• Collect worker outputs

parallel-template

Keywords: template, implementation, parallel, agent Solves:

• Parallel agent fanout template
• Production-ready code
• Copy-paste implementation