Feature Request: Support exporting images to a containerd socket

[spkane]

Description

It is more and more common not to have a local Docker daemon available and during certain testing cycles pushing to a registry is not always easy or desirable.

It would be great if you could easily create a buildpack and then export the results directly into a containerd namespace.

Proposed solution

Extend pack and the underlying creator command so that it is possible to pass in something like -containerd k8s.io, which will then tell the export step to push the results into that containerd namespace instead of into a Docker registry or daemon.

Describe alternatives you've considered

Many hacks that require running Docker, exporting to docker, pull out the image, and then re-pushing it into containerd.

[spkane]

Cross-posted here: buildpacks/pack#1393

Feel free to close one or the other, but I wasn't sure which location was best, and whether both tools would need related changes.

[jjbustamante]

Hi @spkane, Thanks for creating the issue! I've been working on a couple of RFCs related to adding additional support for removing the Daemon or exporting to OCI format.

• RFC: Export to OCI format rfcs#203
• RFC: Deprecate the use of the Daemon rfcs#201

I will try to take this issue into the evaluation and see if we can move forward with the idea!

[spkane]

Thank you very much @jjbustamante ! That would be very much appreciated.

[natalieparellano]

With export to OCI layout supported in lifecycle 0.17.0, this is no longer blocked. We should research if we can use OCI layout as intermediate glue to export to the containerd socket. This research would probably need to be done in partnership with some folks from pack.

[natalieparellano]

cc @jjbustamante

[natalieparellano]

See buildpacks/pack#1519 (comment) - we have a short-term hack that degrades performance, but refactoring imgutil/local to use imgutil/layout/sparse will solve this cleanly.

[jjbustamante]

See buildpacks/pack#1519 (comment) - we have a short-term hack that degrades performance, but refactoring imgutil/local to use imgutil/layout/sparse will solve this cleanly.

Do we need to fix this issue first?