发表机构
AiShiWeiLai Co., Ltd.; College of Computer Science, Sichuan University（爱思唯来有限公司; 四川大学计算机学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
DHSched通过无状态对等Dispatch副本管理有状态实时虚拟形象会话，以代次机制消除控制面与执行权间隙，实现无冲突迁移，支持近5万并发会话。
AI 中文摘要
实时虚拟形象服务运行长期存在的、由GPU支持的会话，这些会话将连续的文本流转换为语音、面部动作、渲染视频和RTC媒体。弹性操作会反复将这些会话重新分配给不同的Worker。重新分配会更新共享的放置状态，而之前的GPU运行时和RTC连接仍然保持存活，从而在控制面的分配与执行权之间产生了间隙。有状态控制器通过保留每个会话的权限来串行化所有权，将该状态置于控制器的扩展和故障转移路径上。我们提出了DHSched，一个通过无状态的对等Dispatch副本管理长期存在的有状态会话的控制面。DHSched将会话所有权外部化为由(WorkerID, epoch)标识的代次。Dispatch有条件地提交每个代次。Infer-Controller在释放推进状态的输入之前，声明并重新验证相同的代次。过期的运行时即使仍然存活，也会失去执行权。DHSched从定期刷新的负载视图中对候选者进行排序，同时保持容量准入的权威性。恢复通过源epoch保护和源无关的重建，沿着相同的代次路径推进。DHSched运行在我们的生产虚拟形象服务栈中，该栈达到了49,987个并发在线会话。在2026年6月10日，它完成了9,927次会话迁移，没有观察到双所有者冲突或容量超调。在生产峰值时，CreateLive延迟在P99为98.3毫秒，PushText延迟在P99为16.5毫秒。在100,000次随机化所有权运行中，分配时、仅声明和仅WorkerID协议允许过期执行，而DHSched没有记录双所有者或过期代次释放。
英文摘要
Real-time avatar services run long-lived, GPU-backed sessions that transform a continuous text stream into speech, facial motion, rendered video, and RTC media. Elastic operation repeatedly reassigns these sessions across Workers. A reassignment updates shared placement state while the previous GPU runtime and RTC connection remain alive, opening a gap between control-plane assignment and execution authority. A stateful controller serializes ownership by retaining per-session authority, placing that state on the controller's scaling and failover path. We present DHSched, a control plane that manages long-lived stateful sessions through stateless peer Dispatch replicas. DHSched externalizes session ownership as a generation identified by (WorkerID, epoch). Dispatch conditionally commits each generation. Infer-Controller claims and revalidates the same generation before releasing state-advancing input. A stale runtime loses execution authority even if it remains alive. DHSched ranks candidates from periodically refreshed load views while keeping capacity admission authoritative. Recovery advances the same generation path via source-epoch guards and source-independent reconstruction. DHSched runs in our production avatar-serving stack, which reached 49,987 concurrent live sessions. On 10 June 2026, it completed 9,927 session migrations with no observed dual-owner conflicts or capacity overshoots. At the production peak, CreateLive latency was 98.3 ms at P99 and PushText latency was 16.5 ms at P99. Across 100,000 randomized ownership runs, assignment-time, claim-only, and WorkerID-only protocols permit stale execution, while DHSched records no dual-owner or stale-generation release.
