Fix bug causing GPU unavaliable after updating container resource

[tkanng]

No description provided.

[Davidrjx]

@tkanng hi, why closing the pr? i also ran into the same problem.

[tkanng]

Hi, @Davidrjx The PR can fix the problem, just merge it ! I close the PR, because it seems there's no plan to merge it. :) If you want to know the reason causing the problem and how to fix it , feel free to contact with me!

[dragonofac]

@tkanng
So, what cause the nvml unknown error in the container when '--cpu-manager-policy=static' option is open ?
And how could I fixed it if the patch not merged?

[tkanng]

Sorry for the late reply. Just merge the patch I submitted，it works fine.

…

On Tue, Jul 16, 2019 at 3:31 PM Daniel ***@***.***> wrote: @tkanng <https://github.com/tkanng> So, why it cause the nvml error in the container why '--cpu-manager-policy=static' option is open ? And how could I fixed it if the patch not merged? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#55?email_source=notifications&email_token=AI2ZLPD7XRJGZ36RZHXY7DLP7V2OJA5CNFSM4HJSHJ5KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODZ77ALI#issuecomment-511701037>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AI2ZLPENDS2BIQ7VB4EWJCDP7V2OJANCNFSM4HJSHJ5A> .

[Davidrjx]

@tkanng thanks

[Davidrjx]

@ruster-cn @tkanng's PR should be right, my local test currently is ok. Just add above similar logic into newer Docker runtime, like v19.03.*, then have related nvidia-runtime updated along with docker. At this time, dockerd and nvidia-container should reference same runtime.