oci layers on top

olot is a python-based tool to append layers (files) onto an OCI-compatible image. It is meant to be used in conjunction with command-line based tools to fetch and upload these images. Tools such as skopeo or oras. It can be used either as a CLI tool, or via standard python interface. The package is published to pypy.

Usage

As a CLI

1. use simple tool like skopeo ( or oras cp, or ...) and produce an oci image layout of the base image for the Modelcar (ie base image could be: A. ubi-micro, B. busybox, or C. even simple base image for KServe Modelcar, etc.)
2. (this project) use pure python way to add layers of the ML models, and any metadata like ModelCarD
3. use simple tool from step 1 to push the resulting layout to the remote registry (i.e. simpletool cp ... quay.io/mmortari/model:car)
4. ... you now have a Modelcar inside of your OCI registry that can be used with KServe and more!

IMAGE_DIR=download
OCI_REGISTRY_SOURCE=quay.io/mmortari/hello-world-wait:latest
OCI_REGISTRY_DESTINATION=quay.io/mmortari/demo20241208:latest
rm -rf $IMAGE_DIR

# Downloads the image `/quay.io/mmortari/hello-world-wait:latest` to the folder `download` with tag `latest`
skopeo copy --multi-arch all docker://${OCI_REGISTRY_SOURCE} oci:${IMAGE_DIR}:latest

# If using oras, you will need to also need to add write permissions
# oras copy --to-oci-layout $OCI_REGISTRY_SOURCE ./${IMAGE_DIR}:latest
# chmod +w ${IMAGE_DIR}/blobs/sha256/*

# Appends to the image found in `download` the files `model.joblib` and `README.md`
poetry run olot $IMAGE_DIR tests/data/sample-model/model.joblib tests/data/sample-model/README.md

# Pushes the (updated) image found in `download` folder to the registry `quay.io/mmortari/demo20241208` with tag `latest`
skopeo copy --multi-arch all oci:${IMAGE_DIR}:latest docker://${OCI_REGISTRY_DESTINATION}

# If using oras
# oras cp --from-oci-layout ./${IMAGE_DIR}:latest $OCI_REGISTRY_DESTINATION

You can now test the image to validate the files exist in it

podman run --rm -it $OCI_REGISTRY_DESTINATION ls -la /models/

Expected Output:

Unable to find image 'quay.io/mmortari/demo20241208:latest' locally
latest: Pulling from mmortari/demo20241208
6163eebc9af4: Download complete 
1933e30a3373: Download complete 
3dc1903f1fc8: Download complete 
Digest: sha256:6effaec653c4ba4711c8bda5d18211014016e543e6657eb36ced186fb9aed9e4
Status: Downloaded newer image for quay.io/mmortari/demo20241208:latest
total 20
drwxr-xr-x    1 root     root          4096 Feb 13 15:17 .
drwxr-xr-x    1 root     root          4096 Feb 13 15:17 ..
-rw-rw-r--    1 root     root          6625 Dec 24 09:24 README.md
-rw-rw-r--    1 root     root          3299 Dec 24 09:24 model.joblib

Cleanup your local image

podman image rm quay.io/mmortari/demo20241208:latest

As a Python Package

Install the package

pip install olot
# poetry add olot

Import and add layers onto a locally available model (using skopeo):

from olot.oci_artifact import oci_layers_on_top
from olot.backend.skopeo import skopeo_pull, skopeo_push

model_dir = 'download'
oci_registry_source='quay.io/mmortari/hello-world-wait:latest'
oci_registry_destination='quay.io/mmortari/demo20241208:latest'

model_files = [
    'tests/data/sample-model/model.joblib',
    'tests/data/sample-model/README.md',
]

# Download the model
skopeo_pull(oci_registry_source, model_dir)

# Add the layers
oci_layers_on_top(model_dir, model_files)

# Push the model
skopeo_push(model_dir, oci_registry_destination)