agent-exec transport 5b732130457a04d9a20ff740653539b8/feb6b3f216f8

tests/integration/multi-instance/multi_instance_test.go

@@ -725,10 +725,28 @@ func (e *env) setupConversation(t *testing.T, uc *userClient) (string, string) {
 
 // postMessage 先等会话空闲（上一 run 终态），再提交消息——#hitl 需要一个
 // 可占用的空闲会话；普通消息同样避免撞上上一 run 的收口窗口。
+//
+// waitConversationIdle 与 POST 之间存在 TOCTOU：观察 idle 后、POST 到达前，
+// 上一条消息的 run 可能才异步开始（新依赖管线启动更慢），服务端正确返回
+// 409 "conversation already has an in-flight run"。对这类 409 有界重试
+// （再 wait + POST，最多 3 次）；其他状态码仍立即 Fatal，不吞错误。
 func (e *env) postMessage(t *testing.T, c *http.Client, convID, text string) {
 	t.Helper()
+	const maxInFlightRetries = 3
+	path := "/api/conversations/" + convID + "/messages"
+	body := map[string]any{"text": text}
 	e.waitConversationIdle(t, c, convID, 20*time.Second)
-	e.do(t, c, "POST", "/api/conversations/"+convID+"/messages", map[string]any{"text": text}, nil)
+	for attempt := 0; ; attempt++ {
+		code, raw := e.raw(t, c, "POST", path, body)
+		if code == http.StatusOK {
+			return
+		}
+		if code == http.StatusConflict && strings.Contains(raw, "in-flight") && attempt < maxInFlightRetries {
+			e.waitConversationIdle(t, c, convID, 20*time.Second)
+			continue
+		}
+		t.Fatalf("POST %s: status %d, body %.300s", path, code, raw)
+	}
 }
 
 func (e *env) waitConversationIdle(t *testing.T, c *http.Client, convID string, within time.Duration) {