How to enable zstd:chunked?

[TheCodeLamp]

Hello, I've used bootc for a while now, and while I like it, I've been a bit frustrated that I need to download such large updates even for small image changes. Is it possible (or even advisable) to enable zstd:chunked when pushing? And how would one do this?

[yudin-s]

It is possible to publish an image with zstd:chunked, but I would separate two things: creating the chunked image and getting bootc to benefit from chunked/deduplicated pulls.

For the push/build side, this is controlled by the container tooling, not by bootc itself. The usual knob is the compression format on the final push/copy step, for example:

podman push --compression-format=zstd:chunked IMAGE REGISTRY/IMAGE:TAG
buildah push --compression-format=zstd:chunked IMAGE docker://REGISTRY/IMAGE:TAG

The registry also has to preserve/accept those layer media types. If it rejects or rewrites them, clients will not see a chunked image even if your local push command used that format.

For the bootc side, the current docs make this an experimental-storage question rather than a normal update setting. The unified-storage docs list direct support for zstd:chunked as a goal, and mark unified storage as experimental. The documented path is:

bootc install to-disk --experimental-unified-storage /dev/sdX
# or on an existing system
bootc image set-unified

After onboarding, future bootc upgrade / bootc switch operations can use the unified path when the image is detected in bootc-owned container storage. That is the path intended to make chunked pulls and deduplication useful.

So my decision rule would be:

• To produce a chunked image: push/copy with --compression-format=zstd:chunked and inspect the pushed manifest/layers.
• To reduce bootc update transfer/import cost: test it with experimental unified storage first.
• For production hosts: treat this as experimental until unified storage is no longer experimental for your use case.

If this answers the enablement question, please mark it as the answer so other bootc users can find the current caveat quickly.

[TheCodeLamp]

You have a very AI-way of talking, lol

I've read the unified-storage issue too, however, I was hoping to understand the potential issues that might come up. Like, if I enable the chunked compression, is it impossible to fetch the image without unified storage enabled?

The feature description of unified storage describes zstd:chunked as a "goal", does that mean that it's not actually supported yet?

I've seen that u-blue has started using zstd:chunked, does this mean that it's stable enough for daily use?

[renner0e]

I've seen that u-blue has started using zstd:chunked, does this mean that it's stable enough for daily use?

To note that ghcr currently does not support partial pulls. If you push these images to other registries, yes partial pulls will actually work.

partial pull of blob sha256:1952fd19b5c7b2a38604fde068b824ffda59f1c103264b7f74af5505ae1bd691: read zstd:chunked manifest: fetching partial blob: received unexpected HTTP status: 501 Unsupported client range

[TheCodeLamp]

I see. Thank you. That's one of those caveats that I wasn't aware of :3