FelixYBW opened a new pull request, #13428: URL: https://github.com/apache/gluten/pull/13428
## What changes are proposed in this pull request? Part of #12288 (see the amd64 vs. arm64 analysis in https://github.com/apache/gluten/issues/12288#issuecomment-6378740235). `docker_image.yml` builds every `centos-8-jdk*` and `centos-9-jdk*` image for both amd64 and arm64. These are the images each arm64 job actually uses: | arm64 job | Image | |---|---| | `velox_backend_arm` `build-native-lib-centos-9`, `velox_backend_cache` `cache-static-lib-centos-9` | `vcpkg-centos-9` | | `velox_backend_arm` `cpp-test-udf-test`, `velox_backend_cache` `cache-shared-lib-centos-9` | `centos-9-jdk8` | | `velox_nightly` `build-native-lib-centos-8-arm64` | `vcpkg-almalinux-8-gcc13` | | ARM TPC, nightly arm64 bundle jobs, weekly openEuler | raw OS images (`stream9`, `centos:8`, openEuler) | No arm64 job uses `centos-8-jdk8/11/17` or `centos-9-jdk11/17`. This PR: - builds `build-centos-8` for amd64 only; - excludes the arm64 legs for `jdk11` and `jdk17` from `build-centos-9`. arm64 builds remain for `centos-9-jdk8`, `vcpkg-centos-9` and `vcpkg-almalinux-8-gcc13`. The first two are also what the dev containers (`docs/developers/dev-container.md`) use on Apple-silicon Macs. **Savings:** on the 2026-10-18 build, these arm64 legs took about 5.5 runner-hours per `docker_image.yml` run: `centos-8` jdk8/11/17 at 55/54/53 min and `centos-9` jdk11/17 at 54/58 min. That's roughly 11 hours a week at the current ~2 full runs a week. **Effect on the tags:** the `merge-dynamic` job still creates each tag, now from a single amd64 digest. The existing multi-arch tags become amd64-only on the next publish. Anyone pulling `centos-8-jdk*`, `centos-9-jdk11` or `centos-9-jdk17` on an arm64 machine needs `--platform linux/amd64` (emulated) or `centos-9-jdk8` instead. `velox-backend-build-in-docker.md` uses `centos-8-jdk8` in its dynamic-build example, which then only works on amd64 hosts. **Overlap with #13422:** that PR drops `centos-8-jdk11` and `centos-9-jdk11` entirely and touches the same matrix lines. Whichever lands second needs a trivial rebase. ## How was this patch tested? - `actionlint`: clean. - I listed the container image of every job that runs on an arm64 runner, in all workflows, to confirm none uses the images being dropped. ## Was this patch authored or co-authored using generative AI tooling? Generated-by: Claude claude-opus-5-5 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
