Repository navigation
Conversation
kmcginnes
marked this pull request as ready for review
November 14, 2026 00:15
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Docker on arm64, such as Apple Silicon, can't pull
3.3.1or later, so the publish workflow now pushes reallinux/amd64andlinux/arm64images under every tag.3.3.0was a single amd64 image that arm64 Docker ran under emulation. From3.3.1, the runner image moved from Ubuntu 24.04 to 26.04, and itsdocker buildwrites an image index with one amd64 image plus a provenance attestation entry. Docker on arm64 finds no matching platform in that index and refuses it. The workflow didn't change between the two releases. The runner logs show the switch: the 3.3.0 run sayswriting image sha256:…, and the 3.3.1 run saysexporting attestation manifestandexporting manifest list.Both version tags come from one build, and
imagetools createcopies that index unchanged, so every alias keeps one digest.The client build is the same on every platform, so it runs once on the runner's own platform. Under emulation it was the slowest stage. Only the prod install, the Node.js binary, and the runtime stage run per platform. Both base images already publish arm64 variants. The proxy server's prod dependencies have no native addons and no platform-specific packages.
--provenance=falsealone would bring back the 3.3.0 shape: amd64 only, run under emulation on arm64. With a real arm64 entry, arm64 hosts pick it, so the attestations stay. QEMU in one job beat a native arm runner plus a merge job, since runners come from a shared pool and the emulated part is small.test_build_docker.ymlnow also builds the arm64 image and checks its architecture, its Node.js version,/status,/explorer/, and that it listens on every interface. The wait for the server goes from 30 to 60 seconds, and the check for the listening log line now retries for up to 10 seconds, since under emulation the line can land after the server already answers. The binfmt and BuildKit images the two setup actions run are pinned by digest, like every other image and action here. Dependabot doesn't update those two pins.Check the tag, version, and latest tags before publishing a release image (#2603) also changes the push steps of
build_docker.yml. This PR touches only the build and push steps and adds the two setup steps before them, so whichever merges second rebases onto a small conflict. That PR's push loop becomes onebuildx buildwith a-tper tag.Evidence
Before:
public.ecr.aws/neptune/graph-explorer:3.3.2andlatestare an image index withlinux/amd64andunknown/unknownonly.After: a local
docker buildx build --platform linux/amd64,linux/arm64, pushed to a throwaway registry on127.0.0.1, gave<version>,sagemaker-<version>,latest, andsagemaker-latestthe same index digest, holdinglinux/amd64,linux/arm64, and two attestation entries.Verified live: on an arm64 host, a
--no-cachebuild of both platforms took 262 s before this change and 121 to 143 s after it, with amd64 emulated. Both images report Node.jsv24.22.2(arm64andx64) and answer/statusand/explorer/. The CI step scripts ran unchanged against the arm64 image. On this PR's first GitHub run of the Docker test workflow, the amd64 build took 69 s, the arm64 build under QEMU 4 min 2 s, and the whole job 6 min 15 s, inside its 15-minute limit. The publish job builds both platforms in one build, so it should take about as long as those two builds.Merge Danger
Door: two-way
Impact: publishing
The next push to
mainpublisheslatest-SNAPSHOTas a two-platform index. Reverting goes back to amd64-only images.Related
🤖 Written by an AI agent on my behalf.