Add "--preserve-digest" option to podman push

[StarpTech]

Is this a BUG REPORT or FEATURE REQUEST? (leave only one on its own line)

/kind feature

Description

When running podman build and podman push multiple times, layers are modified and pushed to the registry even when the image hasn't changed or in other terms was fully cached. To avoid changing digests containers/storage introduce the --preserve-digest flag. See containers/image#1413 I'd like to add this flag to the podman push command

Describe the results you expected:

See #15538

https://github.com/containers/podman/blob/main/pkg/domain/infra/abi/images.go#L285

[mheon]

@flouthoc PTAL

[vrothberg]

AFAICS @mtrmac was not in favor of adding the option (see #15538 (reply in thread)). As outlined in the conversation, the --preserve-digest flag may also push uncompressed data.

I'd also feel more comfortable keeping that in skopeo copy only to avoid users falling into the trap of pushing uncompressed layers.

[mtrmac]

Yes, I’m not at all enthusiastic about podman push --preserve-digests. It’s a trap for inattentive readers.

Overall optimizing this already-pushed case in single seconds to a bit fewer single seconds just seems very low-priority to me.

---

The directions to investigate from #15538 (reply in thread) would help this case, and other cases:

• For some historical reason, we disable HTTP keep-alive, perhaps increasing the per-request latency — although I would guess, without data, that to only matter for images with rather more layers, because keep-alives only help if there are idle connections, and we already run 6 sessions in parallel.
• It might be interesting to discuss looking for a known-compressed variant first, before looking for the original uncompressed one. That could ~halve the number of HEAD requests in this case.

[github-actions]

A friendly reminder that this issue had no activity for 30 days.

[rhatdan]

Since we are not going to work on this issue, closing.

[mtrmac]

FWIW:

The directions to investigate from #15538 (reply in thread) would help this case, and other cases:

• For some historical reason, we disable HTTP keep-alive

This is no longer the case.

Also, unfinished containers/image#1968 might reduce the number of necessary round-trips, by up to about a third for images with several but not too many layers.