集群模式部署，如果重启supervisor，必须重启所有woker吗？ #2402
Description
System Info / 系統信息
docker
Running Xinference with Docker? / 是否使用 Docker 运行 Xinfernece？
- docker / dockerpip install / 通过 pip install 安装installation from source / 从源码安装
Version info / 版本信息
0.15.2
The command used to start Xinference / 用以启动 xinference 的命令
分布式场景，正常启动 supervisor 和 worker
supervisor启动指定了supervisor-port
在worker上启动一个模型，如：bge-m3
Reproduction / 复现过程
重启supervisor，前端无法查看正在运行的模型 bge-m3；模型服务不可用；
Expected behavior / 期待表现
- supervisor重启后，已经运行的模型正常
- 模型服务正常
Metadata
Metadata
Assignees
Type
Projects
Relationships
Participants
Activity
pkunight commentedon Oct 10, 2024
我也发现了这个问题, 必须先启动supervisor, 后启动worker, 而且此后连接不能中断. 否则即使supervisor成功重启了, worker依然会持续报错连不上supervisor的ip地址.
paradin commentedon Oct 10, 2024
启动supervisor时指定supervisor-port的话，重启supervisor后是能够让worker连上的（因为supervisor端口固定了）
但是问题是现在supervisor是有状态的，重启后woker虽然能report_status，但是却没有report running models status
如果supervisor能实现无状态（比如通过redis共享），还能解决目前supervisor单点问题
ak47947 commentedon Oct 12, 2024
我也发现这个问题了，如果这个问题不解决，是没法真正集群使用的
github-actions commentedon Oct 19, 2024
This issue is stale because it has been open for 7 days with no activity.
github-actions commentedon Oct 24, 2024
This issue was closed because it has been inactive for 5 days since being marked as stale.
github-actions commentedon Dec 19, 2024
This issue is stale because it has been open for 7 days with no activity.
github-actions commentedon Dec 25, 2024
This issue was closed because it has been inactive for 5 days since being marked as stale.