Goal

`zstd:chunked` is a image transport mechanism to reduce the amount of data being pulled by image updates. It's the bootc support for what used to be binary deltas for ostree/commit. The goal of this initiaitive is to find out if we could switch container images used by microshift to this compression technique, to reduce the bandwidth needed for MicroShift updates. In the long run, this could be also enabled for full OpenShift, to improve SNO/MNO updates at the edge.

Benefit Hypothesis:

'zsdt:chunked' has the potential to drastically reduce the amount of data needed to be pulled for an image update, as it works below the image layer level, on the actual content of the layer. Esp. bandwidth limited edge deployments would greatly benefit from that.

Resources

Add any resources (docs, slides, etc.) pertinent to the definition of the work. These might not be known until later. Update as necessary.

RHEL Feature: https://issues.redhat.com/browse/RHELBU-2207 

RHEL test plan: https://docs.google.com/document/d/1aBjVmGinrEkDxJoGkyCZWatUf27wGWGb44z14OemKKY/edit?tab=t.0 

Responsibilities

Indicate which roles and/or teams will be responsible for contributing to the initiative and generally what they might be expected to do.

Success Criteria

Find answers to the following questions:

1. What is the potential saving in bandwidth for a Y stream update? Measure the amount of bytes downloaded for container images from e.g. MicroShfit 4.17 to 4.18. Then repeat the same with 'zsdt:chunked' images and measure the difference. A MicroShift air-gapped environment is needed for that, and zstd:chunked needs to be enabled/configured when copying/pushing the images into the local registry. Daniel can provide a local quay with some long term stability if needed.
2. What is the risk of enabling this for the required images, which are shared with OCP? Can other/older images registries which do not support 'zstd:chunked' still handle those images?
3. Could we support this only for MicroShift, e.g. documented in the "airgapped" instructions?
4. What  would be need to change the OCP image build process to support 'zsdt:chunked'? What is the temperature/attitude of the OCP release team to look at this topic?

Results

Based on the answers to the previous stated question, we need to make an informed decision if and how we want to pursue this topic.