feat(ci): save cache on build-main workflows and only restore cache on docker-build-and-push-main workflows

[youtalk]

Description

• fix(docker): fix CCACHE_DIR in Dockerfile #4863

The reproducible-containers/buildkit-cache-dance action introduced in #4854, unfortunately, might not support extracting the cache from the docker buildx bake and only supports docker build.

I confirmed that replacing docker/bake-action with docker/build-and-push-action as per the official sample code does indeed enable the cache to work.
https://github.com/reproducible-containers/buildkit-cache-dance?tab=readme-ov-file#examples

Therefore, in this PR, we will create a cache by running docker/build-and-push-action in the build-main and build-main-self-hosted workflows, which are scheduled to run daily, and in the sequentially executed docker-build-and-push-main and docker-build-and-push-main-self-hosted workflows, we will run docker/bake-action as in the past and only load the cache.

It turns out that the cache was effective in #4854 because I forgot to delete the cache from the commit during my experiment, which caused the cache to be effective. I apologize for the oversight.

Tests performed

Using this commit to temporarily run the build-main workflows in this PR, we confirmed that the execution time from the second run onwards was drastically reduced from the previous 2.5 hours to 1 hour.

• first try (without cache): https://github.com/autowarefoundation/autoware/actions/runs/9495072771/usage
• second try (with cache): https://github.com/autowarefoundation/autoware/actions/runs/9506736727/usage

https://autowarefoundation.github.io/autoware-ci-metrics/

Effects on system behavior

Not applicable.

Interface changes

Pre-review checklist for the PR author

The PR author must check the checkboxes below when creating the PR.

• I've confirmed the contribution guidelines.
• The PR follows the pull request guidelines.

In-review checklist for the PR reviewers

The PR reviewers must check the checkboxes below before approval.

• The PR follows the pull request guidelines.

Post-review checklist for the PR author

The PR author must check the checkboxes below before merging.

• There are no open discussions or they are tracked via tickets.

After all checkboxes are checked, anyone who has write access can merge the PR.

[youtalk]

The workflow namings are unclear, so I would like to refactor and rename them in another PR.
ref. #4747

[youtalk]

Unfortunately, this PR still has an issue where the docker-build-and-push-main workflows cannot restore the cache, even though a fully matching cache is already available. However, since this issue can be resolved in a separate PR, we would like to consider this PR is ready for review for now.

Cache not found for input keys: cache-amd64-cuda-0187c3f0f9a23c6232cfe50f211e9deee563e665bcd440c1d921e03bb67898cc, cache-amd64-cuda-, cache-amd64-

https://github.com/autowarefoundation/autoware/actions/runs/9500467708/job/26183604143#step:6:13

https://github.com/autowarefoundation/autoware/actions/caches

[xmfcx]

Unfortunately, this PR still has an issue where the docker-build-and-push-main workflows cannot restore the cache, even though a fully matching cache is already available. However, since this issue can be resolved in a separate PR, we would like to consider this PR is ready for review for now.

This is happens because the workflow trigger of schedule isolates the cache.
Only scheduled build-main can get the cache of another scheduled build-main job, nothing else.
I guess not even workflow_dispatch event can reach it.

See: https://docs.github.com/en/actions/using-workflows/caching-dependencies-to-speed-up-workflows#restrictions-for-accessing-a-cache

For this to work as intended, docker-build-and-push-main should maintain its own cache.

Also

• instead of: cache-${{ inputs.platform }}-${{ inputs.name }}-${{ hashFiles('autoware.repos') }}
• I propose: ccache-build-main-${{ runner.arch }}-humble-${{ github.sha }}

Reasons:

• We don't need ${{ inputs.platform }} because ${{ runner.arch }} gives the similar result by using more generic variables.
  • Possible values are X86, X64, ARM, or ARM64.
• ${{ inputs.name }} is cuda or no-cuda
  • We should only generate the cache for cuda version. Because no-cuda packages are a subset of cuda packages.
  • no-cuda version can restore and use the same cache.
• Strictly separate actions/cache/save@v4 and actions/cache/restore@v4 actions instead of just using actions/cache@v4 action.
  • This way we know who and when updates the cache and who just uses it.

The reason for -${{ github.sha }} is to know where the cache was created from.

restore-keys: |
  ccache-build-main-${{ runner.arch }}-humble-

Will fetch the most fresh version anyways. As long as it is unique, it will be enough.

@youtalk -san, I know you are busy this week and I would like to implement all I've mentioned until wednesday, is that OK?

[youtalk]

@xmfcx It's my pleasure! Please go ahead.