Hi,

In CRI-O, mounting OCI Images is already available (#8317 ). AFAIK (See a slack comment @saschagrunert ), only the layer with the media type application/vnd.oci.image.layer.v1.tar{+gzip|zstd} is supported.

However, according to:

• Guidelines for OCI Artifact Usage, it is allowed to use customized media types beyond application/vnd.oci.image.layer.v1.tar.
• VolumeSource: OCI Artifact and/or Image, it seems that OCI Artifacts are intended to be valid VolumeSources for mounting in Kubernetes.

Given that only layers with the media type application/vnd.oci.image.layer.v1.tar{+gzip|zstd} are currently pulled, I assume that CRI-O does not fully support general OCI artifacts.

Background: We are currently working on a spec for AI/ML models as OCI artifacts aimed at efficiently delivering models to Kubernetes. We've defined several custom media types at different levels:

• For layer: application/vnd.cnai.model.layer.v1.tar, application/vnd.cnai.doc.layer.v1.tar, etc.
• For manifest: application/vnd.cnai.model.manifest.v1+json
• For config: application/vnd.cnai.model.config.v1+json
  Currently, we prefer to use our custom-defined media type for config instead of application/vnd.oci.image.config.v1+json because we think this might be a more idiomatic usage of OCI artifacts. However, we also have some concerns about things like diff_ids. See (issue Defining the diffID Generation Order in the Model Spec modelpack/model-spec#22) for details.

We are interested in understanding the plans of container runtime like cri-o regarding support for generic OCI artifacts, which would facilitate our next steps.

I've initiated this issue to discuss this topic.

Thank you!