[registry-facade] pulls twice on cache miss

[kylos101]

Bug description

When we pull from US, this does so from EU. Therefore, pulling twice due to a cache miss (and missing cache often) becomes expensive.

Steps to reproduce

This could be related a cache miss for a workspace-image (perhaps because we pushed new base images), or a cache miss on a newly built workspace image (for which we just built and pushed via image-builder-mk3).

Workspace affected

No response

Expected behavior

@aledbf were you able to socialize possible options with @csweichel for this last week? It's okay to say no, I know you two were working the S3 bucket issue instead.

Example repository

No response

Anything else?

Need analysis and design work, not ready to be picked up.

This is tech debt from the introduction of IPFS to cache images (to accelerate image pulls).

[aledbf]

were you able to socialize possible options with @csweichel for this last week?

No, but one option would be the use of io.MultiWriter here, checking first if IPFS is configured or not.

[aledbf]

aspect cost efficiency

Technical aspects should drive this change. The IPFS feature was added a long time after the component, and the additional interzonal costs were a side effect of the IPFS implmentation.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.