search

microsoft / pai

Resource scheduling and cluster management for AI
https://openpai.readthedocs.io
MIT License
2.63k stars 548 forks source link

关于openpai部署对应的k8s版本 #4733

Open damon008 opened 4 years ago

damon008 commented 4 years ago

你好，想请问下，对于基于k8s部署openpai (hadoop集群，非k8s的资源调度)时，从v0.9以后，对应的k8s的版本要求是怎样的？k8s1.9以上都可以吗？还是每个openpai版本对应的k8s的版本有所不同？

fanyangCS commented 4 years ago

We've tested the k8s up to 1.15 on yarn version. We have no plan to test higher version.

damon008 commented 4 years ago

@fanyangCS

请问意思是说openpai的0.14及以上版本，现在支持基于k8s 到ver1.15。

还想问的第二个问题是：OpenPAI部署的hadoop版本一直是2.9版本，请问支持2.10 甚至更上的么？比如：2.10.0 、3.1.3等

fanyangCS commented 4 years ago

0.14是基于yarn调度的。0.14以上就是1.0版本，是基于k8s调度的。openpai 1.0版本需要k8s ver1.15或以上版本。OpenPAI 1.0及以上版本不依赖于Hadoop

OpenPAI version0.14 不支持hadoop 2.10, 3.x.

damon008 commented 4 years ago

@fanyangCS 好的，多谢。另外想请问下，1.0的基于k8s调度的，其中framework-controller是定制化任务，这类任务目前是仅用于训练么？可以用于其他场景么？比如大数据分析等。

fanyangCS commented 4 years ago

for framework controller, please raise the issue in its own repo. OpenPAI itself is mainly designed for deep learning training.

damon008 commented 4 years ago

@fanyangCS ok,thanks.

scarlett2018 commented 4 years ago

Thanks @damon008 for following openpai and interact with us. I noticed you had been a long time fan of OpenPAI, thank you!

Would you mind share more info about what's the major usage for you when considering openPAI for yourself or your organization. if you don't mind, would you please share the org name? so that we can clustering issues together by org names.