panel/README.md
agent e11fc00840 Package the panel as a Nix flake with a NixOS module
The panel now lives in its own repository instead of the nixos config.
The flake provides:

- packages.<system>.default: panelctl plus a panel-api wrapper, with the
  frontend under share/panel and runtime tools appended to PATH
- nixosModules.default: services.reudy-panel (service, directories,
  shared routes group, Caddy vhost with optional Authelia forward_auth,
  routes import and reload path unit, Forgejo integration)
- overlays.default, and checks that build the package and evaluate a
  host using the module

panelctl takes the routes file owner from PANEL_USER/PANEL_GROUP
instead of hardcoding reudy:panelroutes (defaults unchanged).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UbWSNkXxZhYf7eqHTyx3Bf
2026-09-27 17:51:55 +00:00

212 lines
7.6 KiB
Markdown

# panelctl quickstart
Minimal container management panel for rootless Podman + Caddy.
## Installing on NixOS
This repository is a flake that provides the panel as a package
(`packages.<system>.default`) and a NixOS module (`nixosModules.default`).
Add it to your system flake:
```nix
{
inputs.panel = {
url = "git+https://git.reudy.net/reudy-net/panel";
inputs.nixpkgs.follows = "nixpkgs";
};
outputs = { nixpkgs, panel, ... }: {
nixosConfigurations.vps = nixpkgs.lib.nixosSystem {
modules = [
panel.nixosModules.default
{
services.reudy-panel = {
enable = true;
domain = "panel.example.com"; # Caddy virtual host
autheliaAddress = "127.0.0.1:9091"; # forward_auth in front of it
};
}
];
};
};
}
```
The module sets up the `panel-api` service, the directories below, the
`panelroutes` group shared with Caddy, the Caddy import of the generated routes
and a path unit that reloads Caddy when they change. With Forgejo enabled on
the same host, its URLs are passed to the panel automatically
(`services.reudy-panel.forgejo.enable`). See `nix/module.nix` for all options.
To deploy a new panel version, bump the input and rebuild:
```bash
nix flake update panel
sudo nixos-rebuild switch --flake .#vps
```
Checks (package build and a module evaluation) run with `nix flake check`.
## Base directory
`/var/lib/containers`
## Generated structure
```
/var/lib/containers/
├── stacks/<app>/compose.yaml # Compose file per app
├── volumes/<app>/data # Persistent volumes
├── routes/routes.caddy # Single aggregate Caddy routes file
├── backups/<app>-<timestamp>.zip # Volume backups
└── state/apps/<app>.env # App manifest
```
All app routes are written to a single `routes/routes.caddy` file that Caddy imports.
## Quick workflow
```bash
# Routes are "domain|upstream[|path]" entries, comma-separated.
# Create a new app (single route, protected by Authelia)
panelctl init whoami "whoami.reudy.net|127.0.0.1:18080" true
# Create with multiple routes (different ports, optional path)
panelctl init myapp "app.reudy.net|127.0.0.1:18081,api.reudy.net|127.0.0.1:18082|/api/*" true
# Create with wildcard domain (requires DNS challenge in Caddy)
panelctl init wild "*.reudy.net|127.0.0.1:18083" false
# Change routes later (Caddy reloads automatically)
panelctl set-routes whoami "whoami.reudy.net|127.0.0.1:18080,who.reudy.net|127.0.0.1:18080"
# Deploy (compose up + caddy reload)
panelctl deploy whoami
# Check container status
panelctl status whoami
# View logs
panelctl logs whoami --tail 50
# Restart containers
panelctl restart whoami
# Stop containers
panelctl stop whoami
# Validate compose file
panelctl validate-compose whoami
# Backup volumes to zip
panelctl backup whoami
# List backups
panelctl list-backups whoami
# Restore from backup
panelctl restore whoami whoami-20260101-120000.zip
# List all apps
panelctl list
# Show app manifest
panelctl show whoami
# Remove app (keeps volumes)
panelctl remove whoami --keep-volumes
# Remove app and all data
panelctl remove whoami
```
## Notes
- The default compose file uses `traefik/whoami` for smoke testing — edit before production use.
- App names must be lowercase slugs (`[a-z0-9-]`).
- Wildcard domains (`*.example.com`) require DNS challenge in Caddy (provider-specific).
- Backups stop containers for consistency, then restart if they were running.
- If deploy reports `XDG_RUNTIME_DIR` missing, enable lingering:
```
sudo loginctl enable-linger reudy
```
## Web UI & API
- Nix runs `panel-api` as a systemd service on `127.0.0.1:9911`.
- Caddy proxies `https://panel.reudy.net` → panel-api with Authelia forward_auth.
- Open `https://panel.reudy.net` for the web UI.
- API docs: [API.md](API.md)
### Web UI features
- Live status: one `/status` poll every few seconds (faster while something is
running, paused when the tab is hidden) updates cards in place, so open tabs,
unsaved edits and scroll positions are never lost. The header shows when the
panel last synced and warns when the Authelia session has expired.
- Per-app status (running / partial / stopped), container list, and a busy
indicator that is shared between browsers while an operation runs.
- New-app dialog: starter container, pasted compose file or git repository;
suggests the next free port and a domain based on the app name.
- Compose editor with unsaved-changes tracking, Ctrl+S, save & deploy, validate.
- Logs with follow mode, routes editor with validation, file browser with
drag-and-drop upload, backups with restore (and optional redeploy).
- Git source tab: deployed commit, "check for updates", and sync & deploy.
- Activity drawer with the output of every operation (e.g. why a deploy failed).
- Keyboard: `/` search, `N` new app, `Esc` closes menus. Deep links like
`#/whoami/logs` open an app on a specific tab.
### Git-backed apps
Apps created from a repository (Forgejo, GitHub or any git host, over https or
ssh) are cloned to `stacks/<app>/repo`.
**Forgejo.** `panel.nix` points the panel at the local Forgejo
(`PANEL_FORGEJO_URL`, `PANEL_FORGEJO_API_URL`, `PANEL_FORGEJO_SSH_URL`, taken
from `forgejo.nix`). In the panel's **Settings** you can connect a Forgejo
access token (read access to repositories and user). With it, the new-app
dialog lists your repositories and branches, and private ones are cloned over
https with the token. Without it, public repositories are listed and private
ones are cloned over ssh with the deploy key. Commit and compare links point at
Forgejo. The token is stored in `state/panel/forgejo-token` (mode 0600).
**SSH / deploy key.** The panel generates an ed25519 key pair in
`state/panel/ssh/` the first time it is needed. Its public half is shown in
Settings (and next to ssh URLs); add it as a read-only deploy key to a
repository — or to your Forgejo account for access to all repositories — to
clone `ssh://git@git.reudy.net:14922/owner/repo.git` style URLs.
**Sync** fetches the configured branch and
hard-resets the checkout to it before redeploying, so the repository is the
source of truth: compose edits made in the panel are discarded on the next sync
(the UI warns about this). An access token for a private repository is stored in
the clone's `.git/config`; use a read-only token.
### Environment variables
Each app can have environment variables (the **Environment** tab, or when
creating the app; `.env` text can be pasted in). They are stored in
`state/env/<app>.env` as `KEY=VALUE` lines (mode 0600) — outside the repository
and stack directory, so git syncs never touch them — and `panelctl` passes them
to every compose command:
- They are always available for `${VAR}` interpolation in the compose file.
The UI points out variables the compose file uses without a default that
aren't set.
- With **Pass to every container** (the default), `deploy`/`restart` also
generate `stacks/<app>/.panel-env.yaml`, a compose override that lists the
keys under every service's `environment:`. Compose reads the values from its
own environment, so they are never quoted into YAML, and they take
precedence over values set in the compose file.
Values must be single-line. Names that would change how podman/compose run
(`PATH`, `HOME`, `XDG_*`, `DOCKER_*`, `COMPOSE_*`, `PODMAN_*`, …) are rejected.
Changes apply on the next deploy. Backups do not include variables.
### Concurrency
`panel-api` handles requests concurrently, so a long deploy never blocks status
or logs. Mutating operations are serialised per app — a second operation on a
busy app gets HTTP 409 — and `panelctl` takes a `flock` on the shared routes
file while rewriting it.