More details on etherpad: https://etherpad.opendev.org/p/cyborg-yoga-ptg brinzhang 发件人: Brin Zhang(张百林) 发送时间: 2021年10月28日 19:37 收件人: 'openstack-discuss@lists.openstack.org' <openstack-discuss@lists.openstack.org> 抄送: 'xin-ran.wang@intel.com' <xin-ran.wang@intel.com>; Alex Song (宋文平) <songwenping@inspur.com>; Jorhson Deng (邓兆森) <dengzhaosen@inspur.com>; Juntingqiu Qiujunting (邱军婷) <qiujunting@inspur.com>; 'eric_xiett@163.com' <eric_xiett@163.com> 主题: [cyborg][ptg] Yoga PTG Summary Hi everyone! First of all I would like to thank everyone for taking time and attending session. I think we had pretty productive time and discussions. You may find discussion summaries below: * With nova-cyborg interaction ** Cyborg vGPU support, we write a spec that adds the prefilter and the traits against every Nova RP and then cyborg contributors to provide a subsequent spec for Cyborg using their own trait ** Continue to work on resume/suspend feature, add the unit tests and update the PoC codes ** Works on PMEM instance cold migration in Nova, the spec is already merged in Xena release, and it need to re-propose it in Yoga release * Introducing some new accelerators driver ** xilin FPGA Driver ** PMEM Driver ** Optimization the exist device function, such as FPGA program interface, GPU/vGPU support * New feature will be support in Yoga release ** Get device profile get by name. It’s need to add a microversion, because of the request path_url changed, proposed the spec but need to update *** SPEC URL: https://review.opendev.org/c/openstack/cyborg-specs/+/813183 ** Add disable/enable device status to mark the device whether can be use or not ***SPEC URL: https://review.opendev.org/c/openstack/cyborg-specs/+/815460 ** We would like to improve the parameter validation, consider checking the api parameters with schema ** Add batch query ARQs for more than one instance support in Get *One* Accelerator Request API * Docs improving, such as nova-cyborg interaction manual and API ref docs * Improve the exception mechanism, improve the efficiency of abnormal judgment in unit testing * Improve the abnormal instance handling scenarios, for example, when the host is disconnected and the device is damaged, how should we set the device status and the accelerator instance state at this time. brinzhang