CI
Partly builtSome of this works today; the page says which parts.
Run each CI job in a fresh full VM, with Docker, systemd and nested KVM, billed by the second.
Status: harebox run works from any CI system today. GitHub Actions runs-on labels need a registered GitHub App, which isn't connected yet.
One command starts a fresh sandbox, runs the job in it, deletes the sandbox, and exits with the job's own exit code:
harebox run -- echo running tests
| Flag | Does |
|---|---|
--from TEMPLATE |
Start from a saved template with your toolchain and caches |
--size SIZE |
small, default, large, xlarge, or a shape like 8c16g |
--environment NAME |
Bring an environment's variables, secrets and setup script |
--ttl SECONDS |
Auto-stop, so a hung job can't run forever. One hour by default |
--keep-on-failure |
Keep the sandbox when the job fails, and print its ID, so you can inspect it |
--keep |
Always keep it |
--env K=V, --cwd DIR |
Passed to the command |
An interrupt, such as a cancelled CI job, still deletes the sandbox. In GitHub Actions:
jobs:
test:
runs-on: ubuntu-latest
steps:
- run: curl -fsSL https://harebox.com/install.sh | sh
- run: ~/.local/bin/harebox run --from ci-base --size large -- "git clone $REPO app && cd app && make test"
env:
HAREBOX_API_KEY: ${{ secrets.HAREBOX_API_KEY }}
The installer checks that HareBox signed the release before installing anything, with ssh-keygen, which GitHub's Ubuntu runners include.
Give CI its own key with the ci preset, which can manage only the sandboxes it creates: harebox api-key new --name ci --preset ci.
To keep a sandbox between steps, drive it yourself:
ID=$(harebox new --ttl 1800 --name ci-job | jq -r .id)
harebox exec "$ID" -- echo running tests
harebox delete "$ID"
harebox exec exits with the job's own code. --ttl stops a forgotten job. Spot costs the same as Shared and may be preempted with two minutes' notice, so keep deployments on Shared.
jobs:
test:
runs-on: harebox-spot
harebox ci