Open Stack

So I explored how exactly skopeo copy works with the WIP zstd:chunked 
compression format.

The main case here is when copying *new* layers of the *same* image. 
What is not the case in TripleO CI. We build images from scratch and tag 
it only once, like current-tripleo. Here ends that build's lifecycle.

Had we decided to change that to update existing images, we could come 
back and try that new feature of skopeo, or may be even consume it 
directly via buildah bud, when it's there.

Another caveat is that only OCI compliant registries could accept such 
zstd:chunked compressed layers.

But anyway that was a good try...


Bogdan Dobrelya bdobreli at redhat.com writes:
> Note that skopeo will shortly have the layers diffs [0] based "copy" > operations [1],[2],[3]. It also deduplicates things and uses local cache
> to not repeat redundant data transfers. Would be great if TripleO 
> container registry could start using that super cool feature as early 
> adoption.
> 
> We could make a build with these patches and try that for the subject 
> tooling for TCIB distribution over quay.io.
> 
> IIUC, the process is like:
> 
> * as input we have full layers from the source registry, built with TCIB;
> * skopeo uses its super powers to redo it on fly into layers diff
> * either skopeo as well, or maybe even tripleoclient prepare image CLI 
> pushes those optimized layers of the images to the target quay.io registry.
> * ???
> * profit!
> 
> [0] https://github.com/containers/image/pull/902
> [1] https://github.com/containers/image/pull/1084
> [2] https://github.com/containers/storage/pull/775
> [3] https://github.com/containers/skopeo/pull/1111



-- 
Best regards,
Bogdan Dobrelya,
Irc #bogdando