search

microsoft / pai

Resource scheduling and cluster management for AI
https://openpai.readthedocs.io
MIT License
2.61k stars 546 forks source link

Prebuilt docker image for aarch64 #5807

Open huww98 opened 1 year ago

huww98 commented 1 year ago

Short summary about the issue/question: Missing prebuilt docker images for aarch64

Brief what process you are following:

We want to deploy an OpenPAI cluster on Huawei Kunpeng CPUs. But find we can't.

$ docker run ... openpai/dev-box:v1.8.0
WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested
exec /usr/local/start-script.sh: exec format error

Could you provide the prebuilt docker images for aarch64 architecture? Or are there any "install from source" docs?

Binyang2014 commented 1 year ago

We don't know the base image for Kunpeng CPUs. If you want to build image for Kunpeng manually, you need to change https://github.com/microsoft/pai/blob/master/src/<servicename>/build/<servicename>.common.dockerfile. And change the FROM ... sentence.

Since we don't verity if the components work in arm architecture, this change may cause some components failed.

huww98 commented 1 year ago

@Binyang2014 Thanks. I think the ubuntu base image would work. It has arm64 version. Do we have any documents about how every docker image is built?

Binyang2014 commented 1 year ago

please refer this line https://github.com/microsoft/pai/blob/8449892080abb949d481965d1520dcf34a75331e/tests/jenkins/stage_build.sh#L29