Add GPU Numbers Support

[peiniliu]

fix issue 12

[Thor-wl]

Is there any UT and E2E for this feature?

[peiniliu]

Not really, I think.
This is a volcano device plugin, the functionality should be tested through volcano. NVIDIA GPUs numbers can now be defined via container level resource requirements using the 'volcano.sh/gpu-numbers'.

[Thor-wl]

OK. Thanks, Peini!

[shinytang6]

Hi @peiniliu, btw there is another important point to think about: whether to support gpu number & gpu share at the same time or according to the configuration.

e.g. if a user applies for one single gpu, theoretically the available memory of the gpu will decrease, but it cannot be perceived at the device plugin level right now.

[william-wang]

Does anyone have gpu env to have a testing against #12? We can not merge the pr without the testing on GPU env.

[peiniliu]

Hi,
@shinytang6 As discussed, we plan to do the GPU number and GPU share separately, we support configuration on the device plugin to specify GPUStrategy whether be number or share. The design doc now can be reviewed.
@william-wang I bought a GKE cluster to have a test. I ll test it in the following weeks. I ll let you know if all things are done.
Best,
Peini

[peiniliu]

Hi all,
gpu share and gpu number allocation features are tested through GKE with a single GPU node. Preliminary results can be found with in this doc.
https://docs.google.com/document/d/1Q8obEBpYkIgFGQvvjnwTEBvTvWtXzMfpQSIdjYCWXeo/edit?usp=sharing
Pls let me know if there are some misunderstandings or any other comments. Also it will be better to fix this ASAP given the GKE nodes need to be paid :)
Best,
Peini

[william-wang]

@peiniliu Thank you so much for the testing. We'll review the change as soon as possible:)

[Thor-wl]

It's generally OK for me now. Please rebase all the commits into one and add --signoff flag when execute git commit.

[Thor-wl]

Please make sure the commits you submit are ready.

[Thor-wl]

/hold

[peiniliu]

Tomorrow I ll do a test with the reviewed commits. Peini Thor-wl ***@***.***> 于 2022年7月12日周二 15:49写道：

…

***@***.**** commented on this pull request. Please make sure the commits you submit are ready. — Reply to this email directly, view it on GitHub <#15 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AF23Z6N3BQY2ZABCMMI2QXLVTVZW7ANCNFSM5CU3DKAA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[shinytang6]

In addition to the above comments, generally the implementation LGTM

[william-wang]

@peiniliu Overall, Look good to me now. Please combine 21 commits to one commit.

[Thor-wl]

@peiniliu Overall, Look good to me now. Please combine 21 commits to one commit.

+1

[peiniliu]

Thanks all,
I think it is ready now.

[william-wang]

/approve

[shinytang6]

Thanks for the effort!
/lgtm
/approve