修复: B-03b-R10③工作流事件补execution_id治本并发串扰(emit填+消费exec_id守卫)
This commit is contained in:
@@ -73,11 +73,23 @@ pub enum WorkflowEvent {
|
||||
waiting_node: NodeId,
|
||||
},
|
||||
/// 工作流执行完成
|
||||
///
|
||||
/// B-03b-R10 ③(波17 治本): execution_id 字段标识本次终态事件归属的工作流执行。
|
||||
/// 此前该变体不带 execution_id,AppState.event_bus 全局单例下并发工作流的终态事件
|
||||
/// 会被 forward 循环的 matches! 完成判定(workflow.rs:159-163 只看变体不看 exec_id)误判,
|
||||
/// 致任一工作流终态触发所有 forward 循环 break(过早断他人链)。
|
||||
/// #[serde(default)] 向后兼容: 老事件/老 DB 反序列化(无 execution_id)填空串不炸。
|
||||
WorkflowCompleted {
|
||||
#[serde(default)]
|
||||
execution_id: String,
|
||||
total_duration_ms: u64,
|
||||
},
|
||||
/// 工作流执行失败
|
||||
///
|
||||
/// B-03b-R10 ③(波17 治本): execution_id 字段标识本次终态事件归属的工作流执行(同 WorkflowCompleted)。
|
||||
WorkflowFailed {
|
||||
#[serde(default)]
|
||||
execution_id: String,
|
||||
error: String,
|
||||
failed_node: NodeId,
|
||||
},
|
||||
|
||||
@@ -179,8 +179,11 @@ impl DagExecutor {
|
||||
}
|
||||
|
||||
let total = start.elapsed().as_millis() as u64;
|
||||
// B-03b-R10 ③: WorkflowCompleted 携带 execution_id,供 forward 循环按 exec_id 匹配
|
||||
// (全局 event_bus 单例下并发工作流的终态事件不再误触发他人 forward 的完成判定)。
|
||||
self.event_bus
|
||||
.send(WorkflowEvent::WorkflowCompleted {
|
||||
execution_id: self.execution_id.clone(),
|
||||
total_duration_ms: total,
|
||||
})
|
||||
.await;
|
||||
|
||||
@@ -156,10 +156,19 @@ pub async fn run_workflow_inner(
|
||||
loop {
|
||||
match rx.recv().await {
|
||||
Ok(event) => {
|
||||
// B-03b-R10 ③(波17 治本): 完成判定加 execution_id 匹配。
|
||||
// AppState.event_bus 是全局单例,并发工作流各自订阅同一总线;
|
||||
// 此前 matches! 只看变体不看 exec_id,任一工作流的 WorkflowCompleted/Failed
|
||||
// 都会触发本 forward 循环 break,过早断他人链(并发串扰根因)。
|
||||
// 现在按变体+exec_id 双重匹配:只处理属于本 forward_exec_id 的终态事件,
|
||||
// 他人终态事件忽略(emit 给前端后继续循环),不再误 break。
|
||||
// execution_id 字段 #[serde(default)] 向后兼容老事件(空串):
|
||||
// 自家终态事件本就带正确 exec_id;空串匹配只可能漏(自家老事件),不会误 break。
|
||||
let finished = matches!(
|
||||
event,
|
||||
WorkflowEvent::WorkflowCompleted { .. }
|
||||
| WorkflowEvent::WorkflowFailed { .. }
|
||||
&event,
|
||||
WorkflowEvent::WorkflowCompleted { execution_id, .. }
|
||||
| WorkflowEvent::WorkflowFailed { execution_id, .. }
|
||||
if execution_id == &forward_exec_id
|
||||
);
|
||||
let payload = WorkflowEventPayload {
|
||||
execution_id: forward_exec_id.clone(),
|
||||
@@ -197,14 +206,17 @@ pub async fn run_workflow_inner(
|
||||
Ok(Some(record)) => {
|
||||
let terminal = match record.status.as_str() {
|
||||
"completed" => Some(WorkflowEvent::WorkflowCompleted {
|
||||
execution_id: forward_exec_id.clone(),
|
||||
total_duration_ms: 0,
|
||||
}),
|
||||
"failed" => Some(WorkflowEvent::WorkflowFailed {
|
||||
execution_id: forward_exec_id.clone(),
|
||||
error: "工作流执行失败(Lagged 兜底补发,详情见 DB)"
|
||||
.to_string(),
|
||||
failed_node: String::new(),
|
||||
}),
|
||||
"cancelled" => Some(WorkflowEvent::WorkflowFailed {
|
||||
execution_id: forward_exec_id.clone(),
|
||||
error: "工作流被取消(Lagged 兜底补发)".to_string(),
|
||||
failed_node: String::new(),
|
||||
}),
|
||||
@@ -350,6 +362,7 @@ pub async fn run_workflow_inner(
|
||||
.unwrap_or_default();
|
||||
event_bus
|
||||
.send(WorkflowEvent::WorkflowFailed {
|
||||
execution_id: exec_id.clone(),
|
||||
error,
|
||||
failed_node,
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user