Trait MergeStrategy

Source

pub trait MergeStrategy<S>: Send + Sync {
    // Required methods
    fn merge(branches: Vec<S>) -> Result<S, WorkflowError>;
    fn default_instance() -> Self;
}

Expand description

并行分支合并策略 — Graph 层职责，非 State 内建属性。

将多个并行分支执行后产生的状态合并为一个。合并规则由 Graph 编排层决定，而非 State 自身。

§职责边界

State = 数据
BranchState = Overlay
ChangeLog = Observability + Checkpoint
MergeStrategy = 并行语义
Executor = 调度
Node = Effect Producer

§示例

// 为 AgentState 定义合并策略
pub struct AgentStateMerge;
impl MergeStrategy<AgentState> for AgentStateMerge {
    fn merge(branches: Vec<AgentState>) -> Result<AgentState, WorkflowError> {
        // messages: concat, iterations: max, tokens: sum
    }
}

// ParallelNode 使用
ParallelNode::builder()
    .merge_strategy(AgentStateMerge)
    .branch("search", search_node)
    .branch("analyze", analyze_node)
    .build();