Krisztian Litkey
Krisztian Litkey
> ouch > > the default for enable_cdi devices is enabled (true)... Yes, and TBH that is partly why it took me a while to realize what went wrong. I...
> my scenario is PyTorch.... > > if a PyTorch enabled application finds the GPU device it will use it.. if not it will fall back to CPU. In this...
> Is there a way to know at this level if the CDI request is a DRA allocation vs the other uses of CDI and/or CDI annotation experiments. I *think*...
/cc @bart0sh @elezar
> @klihub to your question I was pointing out that if we have a pod with the update already applied.. on sync we probably need to pass over these new...
> @klihub Regarding your comment on "sync", are you referring to the kubelet's Pod sync loop, or a sync loop within containerd? If it's the former, is the suggestion that...
> @mikebrow @mxpv @klihub I added a second "work-in-progress" commit that adds support for persistence of sandbox resource updates. > > Note that the unit recovery unit tests will fail....
@champtar AFAICT, this is an alternative implementation of #124.
> Ci is broken: `/bin/sh: 2: protoc: not found` It's not CI per se IMO, rather I think some github-related PITA that hits us with regular irregularity. What happens is...
> @klihub indeed ... Now that I've written it, mine is more fine grained, ie you can add / modify / remove individual namespaces Currently there is a pair of...