mirror of
https://github.com/imjasonh/kontain.me
synced 2026-07-07 00:32:28 +00:00
No description
git.kontain.me used to clone+checkout a repo onto Cloud Run's tmpfs (memory) and then tar the result. Now it fetches the packfile into an in-memory git dir, walks the requested commit's tree, and streams each blob straight into the layer tar. Only the packfile bytes (plus go-git's delta cache) are buffered, so memory no longer scales with the size of the checked-out worktree. The shipped .git directory is preserved (with core.bare unset and a generated index) so the git CLI still works inside the checkout. Object SHAs change since the tar is rebuilt differently, but the contents are the same. https://claude.ai/code/session_01Jfhuuwb5haNaasc9hs53jw |
||
|---|---|---|
| .github | ||
| cmd | ||
| pkg | ||
| .gitignore | ||
| .terraform.lock.hcl | ||
| apps.tf | ||
| go.mod | ||
| go.sum | ||
| LICENSE | ||
| main.tf | ||
| prober.tf | ||
| README.md | ||
kontain.me
Serving container images generated on-demand, at the time they are requested.
These include:
random.kontain.me, which serves randomly-generated images.mirror.kontain.me, which pulls and caches images from other registries.flatten.kontain.me, which pulls and flattens images from other registries, so they contain only one layer.ko.kontain.me, which builds a Go binary into a container image usingko.apko.kontain.me, which builds a minimal base image containing APK packages, usingapko.git.kontain.me, which serves an image containing a shallow clone of a Git repo at a given branch, tag, or commit, usinggo-git.wait.kontain.me, which enqueues a background task to serve a random image after some amount of time.
This repo also serves viz.kontain.me, which visualizes shared
image layers using Graphviz.
There's also two other services in this random grab-bag:
infinite-git.kontain.megenerates a new Git commit every time the repo is pulledinfinite-go.kontain.me(also in theinfinite-gitrepo) generates new Go module version every time it's fetched
Caveats
- The registry does not accept pushes.
- This is a silly hack and probably isn't stable. Don't rely on it for anything serious.
- It could probably do a lot of smart things to be a lot faster. 🤷
- Blobs and manifests are cached for 24 hours wherever possible, but will be rebuilt from scratch after that time.
How it works
The service is implemented using Google Cloud Run.
When the service receives a request for an image manifest, it parses the
request and generates layers for the requested image, writing the manifest and
blobs to Google Cloud Storage. After it
receives the manifest, docker pull fetches the blobs. The app simply
redirects to Cloud Storage to serve manifests and blobs.