Replace systemctl CLI parsing with D-Bus API

[delano]

Summary

Replace the current systemctl CLI output parsing in systemd.py with proper D-Bus API calls for interacting with systemd.

Background

The current implementation parses text output from systemctl list-units, which is fragile:

• We recently shipped a bug where failed units were returned as "running" because we weren't parsing the state columns correctly
• CLI output format can change between systemd versions
• Text parsing is error-prone and hard to test reliably

systemd's official, stable API is D-Bus. The CLI tools are convenience wrappers, not the primary interface.

Technical Details

Current Implementation (src/ots_containers/systemd.py)

def discover_instances() -> list[int]:
    result = subprocess.run(
        ["systemctl", "list-units", "onetime@*", "--plain", "--no-legend"],
        capture_output=True,
        text=True,
    )
    # Parses text: "onetime@7043.service loaded active running ..."

Target Implementation

Use pystemd or dasbus to query systemd via D-Bus:

# Example with pystemd
from pystemd.systemd1 import Manager

def discover_instances() -> list[int]:
    with Manager() as manager:
        units = manager.list_units()
        ports = []
        for unit in units:
            name, _, _, _, active_state, sub_state, *_ = unit
            if name.startswith(b"onetime@") and name.endswith(b".service"):
                if active_state == b"active" and sub_state == b"running":
                    port = int(name.split(b"@")[1].split(b".")[0])
                    ports.append(port)
        return sorted(ports)

D-Bus API Properties

From org.freedesktop.systemd1.Unit:

• ActiveState: "active", "inactive", "failed", etc.
• SubState: "running", "dead", "failed", etc.
• LoadState: "loaded", "not-found", etc.

Library Options

Library	Pros	Cons
pystemd	Facebook/Meta maintained, widely used	Requires libsystemd
dasbus	Pure Python, modern async	Less systemd-specific
dbus-python	Standard library feel	Older, callback-based

Scope

Functions to convert in systemd.py:

• discover_instances() - Query unit list and states
• daemon_reload() - Call Reload() method
• start(unit) - Call StartUnit() method
• stop(unit) - Call StopUnit() method
• restart(unit) - Call RestartUnit() method
• status(unit) - Query unit properties
• unit_exists(unit) - Check LoadState property

Acceptance Criteria

• All systemd interactions use D-Bus instead of subprocess calls
• Unit tests mock D-Bus interface, not subprocess
• Works on systems with systemd 245+ (Debian 11+, Ubuntu 20.04+)
• Graceful fallback or clear error if D-Bus unavailable
• No change to public API of systemd.py module

References

• systemd D-Bus API
• pystemd GitHub
• systemd-mcp architecture - Example of proper D-Bus integration