docs: add ci.md to CLAUDE.md + clarify runner details
Follow-up to #905: adds CLAUDE.md index entries (file map + reading path) for the new docs/ci.md. Also clarifies runner details in ci.md: - curl/jq use absolute nix store paths (no systemPackages needed) - .runner credential reuse: script writes dummy token on subsequent boots, runner ignores it when .runner file exists
This commit is contained in:
lexis
parent
5e13fcbe94
commit
7f91d52569
2 changed files with 44 additions and 67 deletions
|
|
@ -381,6 +381,8 @@ read them à la carte.
|
|||
[`docs/swarm.md`](docs/swarm.md).
|
||||
- **"How does the rebuild queue work? What are queue kinds and sources?"** →
|
||||
[`docs/coordinator.md`](docs/coordinator.md).
|
||||
- **"How does the CI runner work? What's the auto-registration flow?"** →
|
||||
[`docs/ci.md`](docs/ci.md).
|
||||
|
||||
## Conventions & process
|
||||
|
||||
|
|
|
|||
111
docs/ci.md
111
docs/ci.md
|
|
@ -1,81 +1,56 @@
|
|||
# hive-ci — Forgejo Actions runner
|
||||
# hive-ci: Forgejo Actions Runner
|
||||
|
||||
The `hive-ci` module adds a Forgejo Actions runner to the hive so
|
||||
workflows in the hyperhive repo (and any operator-managed repos on
|
||||
the hive forge) can run on the same host. Disabled by default.
|
||||
The `hive-ci` module runs a Forgejo Actions runner in a `hive-ci` nixos-container, executing CI jobs from `.forgejo/workflows/ci.yml` (e.g., `nix flake check` on every PR).
|
||||
|
||||
## Enabling
|
||||
## Operator bootstrap
|
||||
|
||||
```nix
|
||||
services.hyperhive.forge.ci = {
|
||||
enable = true;
|
||||
name = "prod-hive"; # shows in Forgejo admin panel; default = hostname
|
||||
};
|
||||
```
|
||||
Set `services.hyperhive.ci.enable = true` in the host NixOS config. That's it — no manual token provisioning.
|
||||
|
||||
Requires `services.hyperhive.forge.enable = true` (the runner
|
||||
registers against the hive-forge Forgejo instance). The module
|
||||
asserts this at eval time.
|
||||
**Requirements:**
|
||||
- `services.hyperhive.forge.enable = true` must also be set (the runner registers against hive-forge).
|
||||
- Optional: tune `services.hyperhive.ci.name` (runner name in forge admin panel), `concurrency` (parallel job capacity), `labels` (workflow targeting).
|
||||
|
||||
## Options
|
||||
## Container design
|
||||
|
||||
| option | default | description |
|
||||
|---|---|---|
|
||||
| `forge.ci.enable` | `false` | opt-in; off by default |
|
||||
| `forge.ci.name` | hostname | runner name in Forgejo admin panel |
|
||||
| `forge.ci.concurrency` | `2` | maximum parallel jobs |
|
||||
| `forge.ci.labels` | `["hive-ci:host"]` | runner labels (`<name>:<scheme>`) |
|
||||
| `forge.ci.package` | `pkgs.gitea-actions-runner` | override the runner binary |
|
||||
- **Shared host netns**: container reaches hive-forge at `http://127.0.0.1:<httpPort>` (same as hive-gateway).
|
||||
- **Non-ephemeral**: runner credentials persist across restarts (written to container's stateDir on first registration, reused thereafter).
|
||||
- **Sandbox fallback**: nspawn containers can't create user-namespaces, so nix's sandboxing would always fail. Module sets `nix.settings.sandbox-fallback = true` in the container — nix builds run unsandboxed (safe because the container is already isolated). See `docs/gotchas.md`.
|
||||
|
||||
The `host` scheme in labels means jobs run directly inside the
|
||||
`hive-ci` container (no docker/podman). Workflow files target this
|
||||
runner with:
|
||||
## Auto-registration flow
|
||||
|
||||
1. On first container boot, `systemd` runs the `ExecStartPre` script (from `hive-ci.nix:autoRegisterScript`).
|
||||
2. Script checks for existing runner credentials (`/var/lib/gitea-runner/hive/.runner`):
|
||||
- If found: container was previously registered — write dummy token to satisfy module's path check, exit.
|
||||
- If not found: first boot — proceed to registration.
|
||||
3. Script reads hive-c0re's admin token from `/run/hive-ci/core-token` (bind-mounted from `/var/lib/hyperhive/forge-core-token`).
|
||||
4. Calls `POST /api/v1/admin/runners/registration-token` on the forge using `${pkgs.curl}/bin/curl` and `${pkgs.jq}/bin/jq` (absolute nix store paths — the container doesn't need these in systemPackages). Retries for 30s in case forge is still starting.
|
||||
5. Writes real token to `/run/hive-ci/runner-token`.
|
||||
6. `gitea-actions-runner` reads the token and registers itself, persisting credentials (`.runner` file) to stateDir.
|
||||
7. On subsequent boots, runner reuses the `.runner` credentials — the preStart script detects the existing `.runner` file and writes a dummy token instead. The runner ignores the token file when `.runner` already exists.
|
||||
|
||||
## CI workflow
|
||||
|
||||
The single CI job is defined in `.forgejo/workflows/ci.yml`:
|
||||
|
||||
```yaml
|
||||
runs-on: [hive-ci]
|
||||
name: CI
|
||||
on:
|
||||
pull_request:
|
||||
branches: ["**"]
|
||||
jobs:
|
||||
check:
|
||||
name: nix flake check
|
||||
runs-on: [hive-ci]
|
||||
steps:
|
||||
- uses: actions/checkout@v3
|
||||
- name: check
|
||||
run: nix flake check
|
||||
```
|
||||
|
||||
## Container shape
|
||||
This runs on every PR, executing all flake checks (treefmt, rustfmt, cargo test, cargo clippy, module evaluation). No `--no-build`: the checks' derivations are the canonical source of truth.
|
||||
|
||||
`hive-ci` is an ephemeral-false nixos-container sharing the host
|
||||
network namespace (`privateNetwork = false`) so it reaches the
|
||||
hive-forge Forgejo instance at `localhost:<forge-http-port>` without
|
||||
needing a routed address.
|
||||
## References
|
||||
|
||||
Bind mounts:
|
||||
- `/run/hive-ci/core-token` ← `coreTokenPath` (host's forge
|
||||
admin token, read-only). Used only on first boot for registration.
|
||||
|
||||
The container has `git` in `systemPackages`; everything else a
|
||||
workflow needs (rust tools, nix, etc.) is either pulled in by the
|
||||
workflow itself via `nix flake check` or is part of the base NixOS
|
||||
closure.
|
||||
|
||||
Nix sandboxing is disabled in the container (`sandbox-fallback =
|
||||
true`) because nspawn containers can't create the user namespaces
|
||||
that the Nix sandbox requires. See `docs/gotchas.md` for the
|
||||
general nspawn sandbox note.
|
||||
|
||||
## Auto-registration
|
||||
|
||||
On first boot an `ExecStartPre` script (`hive-ci-autoregister`) runs
|
||||
before the runner service:
|
||||
|
||||
1. Reads the hive-c0re admin token from `/run/hive-ci/core-token`.
|
||||
2. Calls `GET /api/v1/admin/runners/registration-token` on the forge
|
||||
(retrying for up to 30 s while forge starts).
|
||||
3. Writes the registration token to `/run/hive-ci/runner-token`.
|
||||
4. The runner service picks it up and registers, writing credentials
|
||||
to `/var/lib/gitea-runner/hive/.runner`.
|
||||
|
||||
On subsequent boots the `.runner` credentials file already exists;
|
||||
the script writes a dummy token and exits — the runner ignores it
|
||||
and reuses its stored credentials.
|
||||
|
||||
No manual token handling is needed. The whole flow is automatic on
|
||||
`nixos-container start`.
|
||||
|
||||
## Cross-references
|
||||
|
||||
- `docs/forge.md` — hive-forge setup (prerequisite)
|
||||
- `docs/gotchas.md` — nspawn sandbox-fallback note
|
||||
- `nix/modules/hive-ci.nix`: runner configuration, auto-registration script, container setup.
|
||||
- `.forgejo/workflows/ci.yml`: workflow definition.
|
||||
- `docs/gotchas.md`: nix sandboxing limitations in containers.
|
||||
|
|
|
|||
Loading…
Reference in a new issue