Skip to content

Conversation

@tariq1890
Copy link
Contributor

@tariq1890 tariq1890 commented Aug 8, 2024

In this PR, we change how the baseline container runtime configuration is fetched when using containerd/crio.

We now know that neither /etc/containerd/config.toml nor /etc/crio/crio.conf files provide a complete picture of the current configuration. To address this gap, we run the containerd config dump or crio status config commands to get the complete runtime configuration that reflects the current state.

This PR has been tested in the following scenarios

i) nvidia-ctk runtime configure --runtime=containerd - Standalone
ii) nvidia-container-toolkit daemonset - GPU Operator Stack

@elezar
Copy link
Member

elezar commented Aug 9, 2024

Thanks for starting on this @tariq1890. I was thinking a bit on how to implement this and created #643 (more from a refactoring perspective).

The idea is that we introduce an interface that returns a *toml.Tree when called. We then have different implementations of this for:

  1. a .toml file.
  2. a command such as containerd config dump
  3. a direct query to the cri socket (not yet implemented).

This has the benefit that we can separate the construction of this Source / Reference from its use in the various runtimes.

What are your thoughts?

@tariq1890
Copy link
Contributor Author

Thanks @elezar . I have opened a new PR with your refactoring changes: #686

@elezar
Copy link
Member

elezar commented Sep 19, 2024

Does #686 replace this? If so, can we close this PR?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants