pobsync/docs/development.md

# Development Notes

This document contains development and optional Docker workflows. The recommended production path is the native
systemd installer documented in the README.

## Local Development

```
python3 -m venv .venv
. .venv/bin/activate
python3 -m pip install -e .
mkdir -p var
python3 manage.py migrate
python3 manage.py createsuperuser
python3 manage.py runserver
```

The admin is available at:

- http://127.0.0.1:8000/
- http://127.0.0.1:8000/admin/

Staff-only JSON endpoints are available at:

- http://127.0.0.1:8000/api/
- http://127.0.0.1:8000/api/status/

## Running Tests

The project test suite is currently run through the Docker image so the runtime dependencies match deployment:

```
docker compose build web scheduler worker
docker compose run --rm web python manage.py test pobsync_backend --verbosity 2
```

## Maintainer CLI

The Django UI is the normal operating surface. The `pobsync` entrypoint and direct `manage.py` commands are kept for
debugging, automated maintenance, and migrations. Prefer using the control panel for day-to-day host configuration,
schedule changes, manual backup queueing, snapshot discovery, retention planning, and SSH credential management.

Useful checks:

```
pobsync django check
python3 manage.py showmigrations pobsync_backend
```

The short `pobsync` aliases are limited to operational actions that are useful while debugging a running install.
Configuration aliases are intentionally not public commands; use the Django UI or explicit management commands instead.

## UI Refresh Pattern

The control panel stays Django-template-first. Pages that need live status should expose a small server-rendered partial
view and opt into refresh with `data-refresh-url` and `data-refresh-interval` on the container that should be replaced.
The shared script in `base.html` polls only those explicit regions, skips refreshes while the browser tab is hidden, and
lets the partial response turn polling off with the `X-Pobsync-Refresh-Active: false` header.

Use this for operational status surfaces such as running backup details. Avoid refreshing form-heavy sections while an
operator might be typing.

Worker and scheduler commands are normally run by systemd services:

```
pobsync worker --loop --interval 15
pobsync scheduler --loop --interval 60
```

One-off maintenance commands are still available when the UI is not the right tool:

```
pobsync backup <host> --dry-run
pobsync discover-snapshots --host <host>
pobsync retention <host>
```

For scripted configuration changes, call the Django management command explicitly so it is clear that this is an
automation/debugging path rather than the normal UI workflow:

```
pobsync django configure_pobsync_host <host> --address <host.example>
pobsync django configure_pobsync_schedule <host> --schedule-expression "15 2 * * *"
```

## Installer Development

The native installer is interactive by default when stdin is a terminal. It should keep every prompt backed by a command
line flag or environment variable so production installs remain scriptable.

Useful modes:

```
sudo scripts/install-systemd
sudo scripts/install-systemd --non-interactive
sudo scripts/install-systemd --verbose
sudo scripts/install-systemd --create-superuser --superuser-username admin
sudo scripts/update-systemd
```

The installer should print a short completion summary with the control panel URL, Self Check reminder, and service log
commands. Keep normal output user-facing: pobsync step names with OK, FAILED, or SKIPPED. Full apt, pip, Django, and
systemd output belongs behind `--verbose` or in the failed step output.

The updater is intentionally a small wrapper around the installer for routine production deploys. It should stay
non-interactive, preserve the existing environment file, skip OS package installation, skip superuser creation, and still
run the Django/runtime refresh steps needed after a code update.

## Docker With SQLite

Docker Compose is useful for local development and disposable test installs. Native systemd is preferred for production
backup servers.

```
docker compose up --build web
```

This starts Django on:

- http://127.0.0.1:8010/
- http://127.0.0.1:8010/admin/
- http://127.0.0.1:8010/api/
- http://127.0.0.1:8010/api/status/

Run the scheduler alongside the web admin:

```
docker compose up --build web scheduler worker
```

The web service runs Django through Gunicorn and serves static files with WhiteNoise. The container persists
`/opt/pobsync` and the SQLite database in Docker volumes.

Backup data is always available at `/backups` inside the containers. By default this uses `./backups` on the host.
Override the host-side mount with `POBSYNC_BACKUP_ROOT`:

```
POBSYNC_BACKUP_ROOT=/mnt/backups/pobsync docker compose up --build web scheduler worker
```

## Docker With MariaDB

```
docker compose --profile mariadb up --build web-mariadb
```

With the scheduler:

```
docker compose --profile mariadb up --build web-mariadb scheduler-mariadb worker-mariadb
```

SQLite remains the default because it is enough for a single backup server and keeps deployment simple.

For native systemd installs with MariaDB client support, run the installer with:

```
sudo scripts/install-systemd --install-extras mariadb
```

## Current Architecture

The public operating surface is Django-first. The CLI is now a maintainer layer around Django management commands and
the old YAML/cron workflow has been retired from the `pobsync` entrypoint.
Discovered snapshots are stored in `SnapshotRecord`, including the base snapshot metadata and a nullable SQL link to the
base record when it is known.

The Django retention command plans from `SnapshotRecord` instead of rediscovering snapshots from the filesystem.
Post-backup pruning from Django also uses the SQL retention service after the completed snapshot is recorded.
Staff-only JSON endpoints expose service status, hosts, snapshots, and backup runs for lightweight inspection.
Staff-only dashboard views expose the same operational state through Django templates.
Host pages include a safe snapshot discovery action that records existing snapshots into SQL.
Host pages also include a read-only SQL retention plan view before any destructive pruning action.
Schedules can be created or updated from host pages using the same SQL-backed scheduler model.
Host config can be edited from host pages while keeping host identity stable.

The remaining internal engine code still contains reusable backup primitives:

- snapshot naming and metadata
- rsync command construction and execution
- retention planning and pruning
- host locking

Next refactor targets:

- Move more snapshot lifecycle details into typed domain objects.
- Replace remaining dictionary-shaped config at engine boundaries.