973333f0dd
This uses https://github.com/actions/cache, this should avoid repulling packages from the Fedora repositories every time we rebuild, downloading them from Github is minimally faster from my testing. This pulls in the package cache before every build and uploads the new cache again (with new packages) so it can be reused on later runs. It's kinda janky as the first image i.e. aurora-dx-nvidia-open :stable and :latest are competing with each other basically as they have the same image name. Only one job run can write to the same cache (we have four of those). So there will be some inefficiencies as some runs won't upload packages. But the only difference should be nvidia specific things and rocm which we only install on our regular images. So it's not perfect in any way. But this is easy to revert so I say we just try it. All this is possible because we set --mount=type=cache,dst=/var/cache/libdnf5 and keepcache=1 for dnf. docs: https://docs.github.com/en/actions/reference/workflows-and-actions/dependency-caching |
||
|---|---|---|
| .. | ||
| ISSUE_TEMPLATE | ||
| workflows | ||
| changelogs.py | ||
| CODEOWNERS | ||
| FUNDING.yml | ||
| pull.yml | ||
| pull_request_template.md | ||
| renovate.json5 | ||
| semantic.yml | ||