udmin/docs/flow_architecture.md

Flow 架构与执行图（udmin）

模块架构

graph LR
  subgraph flow
    dsl[flow/dsl]
    domain[flow/domain]
    context[flow/context]
    engine[flow/engine]
    task[flow/task]
    executors[flow/executors/*]
    mappers[flow/mappers]
    log_handler[flow/log_handler]
  end

  subgraph services
    svc_flow[services/flow_service]
    svc_logs[services/flow_run_log_service]
  end

  subgraph routes
    r_flows[routes/flows]
    r_run_logs[routes/flow_run_logs]
  end

  subgraph middlewares
    mw_sse[middlewares/sse]
    mw_ws[middlewares/ws]
    mw_jwt[middlewares/jwt]
    mw_http[middlewares/http_client]
  end

  subgraph infra
    db[db]
    redis[redis]
    models_flow[models/flow]
    models_run[models/flow_run_log]
  end

  r_flows --> svc_flow
  r_run_logs --> svc_logs
  svc_flow --> dsl
  svc_flow --> mappers
  dsl --> domain
  mappers --> context
  svc_flow --> engine
  engine --> task
  task --> executors
  executors --> mw_http
  engine --> log_handler
  log_handler --> svc_logs
  mw_sse -.events.-> r_flows
  mw_ws  -.events.-> r_flows
  svc_flow --> models_flow
  svc_logs --> models_run
  models_flow --> db
  models_run --> db
  mw_jwt --> r_flows
  redis -.token check.-> mw_jwt

引用：

• DSL/Design 构建 backend/src/flow/dsl.rs:60-93,138-170,172-203,246-303
• 领域模型 backend/src/flow/domain.rs:20-28,31-36,39-47,60-68
• 上下文与事件 backend/src/flow/context.rs:29-45
• 引擎驱动 backend/src/flow/engine.rs:117-209,213-577
• 编排服务 backend/src/services/flow_service.rs:285-305,342-349,351-365,366-399
• 路由入口 backend/src/routes/flows.rs:26-35,101-133
• 运行日志服务 backend/src/services/flow_run_log_service.rs:46-63,74-131

请求/运行编排时序

sequenceDiagram
  participant C as Client
  participant R as routes.flows
  participant S as services.flow_service
  participant D as flow.dsl & mappers
  participant E as FlowEngine
  participant L as flow.log_handler
  participant M as models(flow/flow_run_log)
  participant SSE as middlewares.sse/ws

  C->>R: POST /flows/{id}/run
  R->>S: run(id, input)
  S->>M: get flow doc
  alt design_json
    S->>D: chain_from_design_json(design)
    D-->>S: ChainDef
    S->>D: ctx_from_design_json(design)
    D-->>S: ctx supplement
  else YAML
    S->>D: parse FlowDSL (YAML)
    D-->>S: ChainDef
  end
  S->>E: drive(chain, ctx, opts)
  E-->>S: (ctx, logs)
  S->>L: log_success/log_error
  L->>M: insert flow_run_log
  E-->>SSE: emit node/done/error (stream)
  S-->>R: RunResult(ok, ctx, logs)
  R-->>C: ApiResponse

关键实现：

• 解析与补充 backend/src/services/flow_service.rs:285-305
• 引擎驱动 backend/src/services/flow_service.rs:342-349
• 日志入库与事件推送 backend/src/flow/log_handler.rs

引擎执行图（含执行模式）

flowchart TD
  A[选择起点: Start/入度=0/首节点] --> B{合流屏障}
  B -- 未达成/去重 --> Z[跳过当前节点]
  B -- 达成 --> C{任务?}
  C -- 否 --> D[选择后继]
  C -- 是 --> E{执行模式}
  E -- Sync --> ES[同步执行任务; 写回 ctx]
  E -- Async(Fire&Forget) --> EA[异步执行; 追踪句柄]
  E -- Queued(组队列) --> EQ[组信号量=1; 获取后执行]
  E -- Bounded(限并发) --> EB[组信号量=n; 获取后执行]
  ES --> D
  EA --> D
  EQ --> D
  EB --> D
  D -->|条件节点| F{条件评估 JSON/Rhai}
  F -- 通过 --> G[进入后继]
  F -- 不通过/无匹配 --> H[挑选无条件后继或停止]
  D -->|非条件| G
  G --> I{是否 End}
  I -- 是 --> J[FlowEnd 等待组策略; 推送事件; 结束]
  I -- 否 --> K{多分支?}
  K -- 单分支 --> L[计算组离开; BranchExit 等待]
  K -- 多分支 --> M[并行驱动其他分支]
  L --> A
  M --> L --> A

参考：

• 模式枚举与选项 backend/src/flow/context.rs:9-15,29-45
• 合流与分支 backend/src/flow/engine.rs:245-268,502-574
• 组等待策略 backend/src/flow/engine.rs:62-115,559-573

DSL/Design 转换

flowchart LR
  X0[Design JSON 输入] --> X1[校验: 唯一ID/Start&End/合法边]
  X1 --> X2[兼容 sourcePortID → source_port_id]
  X2 --> X3[推断节点 kind/name/task]
  X3 --> X4[解析组: members/parentID/awaitPolicy]
  X4 --> X5[组装条件: AND组/端口匹配/启发式]
  X5 --> X6["生成 ChainDef: nodes links groups"]
  X0 --> Y1[mappers: 节点配置提取]
  Y1 --> Y2[ctx.nodes.<id>.<executor>]

实现位置：backend/src/flow/dsl.rs:138-170,172-203,246-303 与 backend/src/flow/mappers.rs:27-95

数据模型（ER）

erDiagram
  FLOWS {
    bigint id PK
    varchar name
    text yaml
    text design_json
    varchar code
    varchar remark
    timestamp created_at
    timestamp updated_at
  }
  FLOW_RUN_LOGS {
    bigint id PK
    bigint flow_id FK
    varchar flow_code
    text input
    text output
    bool ok
    text logs
    bigint user_id
    varchar username
    timestamp started_at
    bigint duration_ms
    timestamp created_at
  }
  FLOWS ||--o{ FLOW_RUN_LOGS : has

事件与日志通道

graph LR
  engine_push[engine.push_and_emit] --> sse_emit[middlewares.sse.emit_*]
  engine_push --> ws_emit[middlewares.ws.forward]
  sse_emit --> client[前端]
  ws_emit  --> client
  engine_push --> log_handler_db[DatabaseLogHandler]
  engine_push --> log_handler_sse[SseLogHandler]
  log_handler_db --> run_log_service
  log_handler_sse --> run_log_service
  run_log_service --> db

执行器生态

graph LR
  subgraph executors
    http
    db
    variable
    script_rhai
    script_js
    script_python
    condition
  end
  ctx_nodes[ctx.nodes.<id>.*] --> executors
  ctx_global[ctx.* 顶层] --> executors
  http --> http_out[(http_response)]
  db   --> db_out[(db_response)]
  variable --> var_out[(ctx键值写回)]
  condition --> route[分支选择]

注意点：写回策略与幂等，参考 backend/src/flow/engine.rs:294-375

---

以上图示与说明用于快速理解 Flow 的完整链路：从 DSL/Design 解析到引擎驱动与事件/日志通道，再到数据持久化与执行器生态。结合上文的代码引用，可在 IDE 中跳转到具体实现进行深度阅读。