Org-level defaults and shared reusable workflows for the EduIDE organisation.
Builds a container image for linux/amd64 and linux/arm64 on GitHub-hosted runners and publishes one multi-arch manifest. Every EduIDE image goes through this workflow.
jobs:
build:
uses: EduIDE/.github/.github/workflows/build-and-push-docker-image.yml@v1
with:
image-name: eduide/eduide-cloud/service
docker-file: dockerfiles/service/Dockerfile
secrets: inheritTags published (both point at the same manifest):
| Trigger | base_tag |
also published |
|---|---|---|
push to main |
latest |
latest-<sha7> |
| pull request | pr-<N> |
pr-<N>-<sha7> |
release v2.3.0 |
2.3.0 |
2.3.0-<sha7> |
| other branch | <branch-slug> |
<branch-slug>-<sha7> |
image-tag input |
that value | <value>-<sha7> |
The leading v is stripped from release tags so that a Helm chart's
appVersion and the image tag it refers to are the same string.
Inputs (all optional except image-name): image-name, docker-file,
docker-context, ref, build-args, labels, image-tag, registry,
cache-image, no-cache, free-disk-space, build-amd64, build-arm64.
Outputs: image_tag, base_tag, sha_tag, cache_tag, image_repo.
Use sha_tag when one image must build on top of another:
build-args: |
BASE_IDE_TAG=${{ needs.build-base.outputs.sha_tag }}Secrets: registry-user, registry-password (both default to the GitHub
token), and docker-secrets for build-time secrets such as SENTRY_AUTH_TOKEN.
- Both architectures are always built, including on pull requests. A PR image
that exists only for amd64 cannot be scheduled onto an arm64 node, which our
test clusters have. Building one architecture is possible via
build-amd64/build-arm64but is intended for debugging only. - GitHub-hosted runners only. There is deliberately no
execution-modeinput and no self-hosted or ARC path, so what CI proves is what ships. free-disk-spacedefaults totrue. The runner image ships ~30GB of toolchains we never use, and the Theia IDE images are large enough that reclaiming it is the difference between a green build andENOSPC.- Push-by-digest, then merge. Each architecture pushes an untagged digest; a final job assembles the manifest list. A partial failure therefore never publishes a half-built tag.
- The manifest is verified after publishing. A silently single-architecture image is the failure this workflow exists to prevent, so the merge job asserts the platforms are actually present rather than trusting the build.
printf, notecho, when slugging tags.echo | tr -cturns the trailing newline into a hyphen. That bug published1.1.0-and1.1.0--375ef32to GHCR before this workflow existed;tests/test-derive-tags.shguards it.
release.yml validates a version, takes the release notes from
CHANGELOG.md and creates a draft GitHub Release on the dispatched
commit. Each repository has a thin release.yml (workflow_dispatch, input
version) that calls it, then runs its own build, publish and verify jobs with
needs: release, and ends with release-publish.yml, which publishes the
draft. The tag only appears at that last step, so a failed build leaves a
draft and no tag.
jobs:
release:
uses: EduIDE/.github/.github/workflows/release.yml@main
permissions:
contents: write
with:
version: ${{ inputs.version }} # X.Y.Z or X.Y.Z-rc.N, no leading v
tag_prefix: "v"
changelog_heading: "## [{version}]"
# build, publish and verify jobs with needs: release
publish-release:
needs: [release, verify]
uses: EduIDE/.github/.github/workflows/release-publish.yml@main
permissions:
contents: write
with:
tag: ${{ needs.release.outputs.tag }}Inputs of release.yml: version (required), tag_prefix (default v),
changelog_path (default CHANGELOG.md), changelog_heading (default
## [{version}], {version} is replaced), version_check (optional bash run
in the checkout with $VERSION set; a non-zero exit fails; skipped for release
candidates). Pass version_check only where a file in the repository states
the version (a Chart.yaml) or another invariant of the released commit must
hold; repositories that stamp the version at build time pass none.
Outputs: tag, prerelease (true/false), sha.
Refuses: a malformed version, an existing tag, a version not newer than the
last release with the same prefix, a stable version dispatched from a branch
other than main, and a missing or empty CHANGELOG section. A release
candidate without a section of its own falls back to ## [Unreleased].
Reruns: "Re-run failed jobs" keeps the draft of the first attempt. A new
dispatch of the same version replaces a leftover draft of that tag, so drafts
never pile up. Asset uploads (gh release upload) and gh release view work
on a draft by tag name, but only for a job with contents: write: a draft is
invisible to a read-only token.
A tag or Release created with GITHUB_TOKEN starts no other workflow, so the
caller must run its publish jobs itself.
Runs gitleaks over the commits a pull request or push adds, not the whole history. The binary is downloaded from the release and verified against a pinned sha256 (gitleaks-action needs a paid licence for organisation repositories). Findings are redacted in the log. No inputs, no secrets.
on:
pull_request:
push:
branches: [main]
jobs:
secret-scan:
uses: EduIDE/.github/.github/workflows/secret-scan.yml@mainA PR scans base..head, a push scans before..after. A new branch, a force
push, a manual or scheduled run scans everything not yet on the default branch.
merge_group events are not supported: call the workflow on pull_request and
push only.
Repo-specific allowlists go in .gitleaks.toml and .gitleaksignore in the
caller. On pull requests both are read from the base commit, so a PR cannot
weaken its own scan; changes to them take effect after they are merged.
To bump gitleaks, change GITLEAKS_VERSION and GITLEAKS_SHA256 together.
Every third-party action in this repo is pinned by commit SHA with the tag in a
trailing comment (uses: owner/action@<sha> # v1.2.3). The shared Renovate
preset extends helpers:pinGitHubActionDigests, so all EduIDE repos get the same
treatment and Renovate keeps the digests current.
Actions owned by EduIDE/**, including the reusable workflows above, are
deliberately excluded: they stay on @main or @v1. Pinning them by digest
would turn every change in this repo into a Renovate PR in every consumer. The
trust boundary is the same organisation, so protect main and the v1 tag here
(branch protection, immutable releases) instead.
./tests/test-derive-tags.sh # requires yqThe test extracts the tag-derivation shell directly out of the workflow YAML and executes it, so it cannot drift from what runs in CI.
v1 is a moving tag. A maintainer advances it by hand to a reviewed commit on
main after a non-breaking change (git tag -f v1 <sha> && git push -f origin v1
by a maintainer allowed to update it by the tag ruleset). Branch and tag rulesets
on main and v1 should require review for main and restrict v1 updates to
maintainers.
Consumers should pin the v1 tag. Breaking changes to inputs or outputs get a
new major tag.