- Dockerfile: deterministic build from pinned CONTAINER_APP_VERSION + FEATURES - Merges named feature branches at specific SHAs for reproducibility - Switch CronJob to custom image with --clone-url-base and --all-organizations - Add kingfisher to service-versions.yaml (version tracks upstream main SHA) - Document spork container builds in new how-to card - Document spork workflow in CLAUDE.md - Update kingfisher service docs for custom image Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3.4 KiB
| title | modified | last-reviewed | tags | |||
|---|---|---|---|---|---|---|
| Build a Spork Container | 2026-03-29 | 2026-03-29 |
|
Build a Spork Container
How to build a container image from a spork-strategy project with fully-pinned, reproducible inputs.
Why not use the deploy branch directly?
The deploy branch is force-pushed on every mirror-sync. Building from deploy is not reproducible — the same build a week later fetches different code (or fails because the old commit was garbage collected).
Instead, spork containers use Nix to fetch upstream main at a pinned SHA and generate patches from feature branches at pinned SHAs. Both upstream and feature commits are on stable branches that are never force-pushed.
How the Nix build works
The default.nix uses builtins.fetchGit (eval-time, network access) to fetch two source trees:
- Upstream source at the pinned
upstreamRevonmain - Feature branch source at the pinned
revfor each feature
Then a sandboxed diff -ruN generates a patch from upstream→feature for each feature branch. buildRustPackage applies the patches to the upstream source and builds.
This means:
- The upstream rev persists forever (main only fast-forwards)
- Feature revs are on your branches (you control them)
- No dependency on the
deploybranch - Fully reproducible given the same revs
Prerequisites
- Sporked project set up (see create-a-spork)
- Nix build runs on ringtail (
nix-container-builderrunner)
Get the SHAs
cd ~/code/3rd/kingfisher
git fetch origin
# Upstream SHA (main branch)
git rev-parse origin/main
# e.g., 1d37d2983cd4a58c12663dd8df0e79dfe89a5d75
# Feature branch SHAs
git rev-parse origin/feature/upstream/clone-url-base
# e.g., 677c7a5d5fc42b655d38fbf95dc8b814d89ceabb
Update default.nix
Edit containers/kingfisher/default.nix:
version— short upstream SHA (for container tag)upstreamRev— full upstream main SHAfeatures[].rev— full feature branch SHA
If dependencies changed, update Cargo.lock too:
cd ~/code/3rd/kingfisher
git checkout origin/main
cargo update
cp Cargo.lock ~/code/personal/blumeops/containers/kingfisher/Cargo.lock
Build and push
The build is triggered via the standard container build workflow on ringtail's nix-container-builder runner, or manually:
mise run container-build-and-release kingfisher
Update the deployment
- Update
argocd/manifests/kingfisher/kustomization.yamlwith the new tag - Update
service-versions.yamlif the upstream SHA changed - Sync the ArgoCD app
Note on CONTAINER_APP_VERSION
The default.nix includes version which maps to CONTAINER_APP_VERSION for the container-version-check hook. For sporked containers this is a git SHA, not a release version. Don't confuse it with an upstream release number.
Reproducibility
The upstream rev must be an ancestor of each feature rev. If you bump the upstream rev without rebasing your feature branches, the generated patch will conflict and the build fails — which is the correct behavior.
The invariant: feature revs are descendants of the upstream rev. Mirror-sync maintains this automatically. You just need to update the revs in default.nix after an upgrade.
See also
- create-a-spork — initial spork setup
- manage-spork-branches — feature branch workflow
- kingfisher — first sporked project