Evan Lezar comments

Results 419 comments of


                                            Evan Lezar

no runtime for "nvidia" is configured

Please restart the `nvidia-operator-validator-zk4mv` pod to start with. If this proceeds, then restart the other pods too.

fsnotify/fsnotify needs help

Thanks for highlighting this. What is the plan forward with this? Is the current maintainer open to adding additional maintainers? As is the case with @klihub and @thaJeztah, this is...

Proposal: Use a docker image for the worker nodes

One issue is that this does not allow custom worker node images to be used.

Bump github.com/mittwald/go-helm-client from 0.12.10 to 0.12.14 in /tests

@dependabot rebase

`nvidia.com/gpu.memory` capacity

/cc @klueska

Bump google.golang.org/grpc from 1.63.2 to 1.67.1

@dependabot rebase

Support for specifying CDI devices through an API?

@suptejas I am not sure whether podman supports this. If I recall correctly, there was some work done to extend the functionality beyond the existing CLI-based support. I would ask...

CDI spec file gets created with 600 access

We use the following logic to ensure that the spec is world-readable: https://github.com/NVIDIA/nvidia-container-toolkit/blob/f959c0daaa30e7079458e550ed867cd4faf4a102/pkg/nvcdi/spec/spec.go#L67-L69 It may be useful to consider using something like `renameio` though to ensure atomic writes: https://github.com/NVIDIA/k8s-device-plugin/blob/c181c46f5fdeb367f16c5f9ab8be83c3ae60a90e/internal/lm/output.go#L90

[gfd] Fixing logic of atomically file writing

@belo4ya thanks for the fix. I think we would be ok with using `renameio`. Feel free to update this PR to switch to that instead. Alternatively we can do so...

[gfd] Add option to disable automatic cleanup features file on gpu-feature-discovery exit

Thanks @belo4ya, I have created #899 to add this option and we can continue this discussion there. @ArangoGutierrez one thing that I noted is that we don't do any cleanup...