Struct DaemonManager 

pub struct DaemonManager { /* private fields */ }

Manages the oxios background daemon.

Implementations

impl DaemonManager

pub fn new(pid_file: &str, log_dir: &str) -> Self

Create a daemon manager from config paths.

pub fn set_probe_port(&mut self, port: u16)

Set the port used by orphan-detection probes. Called by the CLI after assembling a manager; without it status/stop cannot catch --foreground debug instances that escape the pidfile.

pub fn with_probe_port(self, port: u16) -> Self

Consuming builder for the probe port. Lets CLI sites write DaemonManager::new(..).with_probe_port(port) without binding to mut, which is awkward in expression position.

pub fn status(&self) -> DaemonStatus

Check daemon status by reading every liveness source we have.

Resolution order — every source that can prove an oxios daemon is alive is consulted in turn, because each one alone is unreliable: the pidfile can be stale or absent (a direct --foreground run), the lockfile-based daemon can be invisible to a future daemon’s own write, and the launchd-managed daemon can be invisible to both. We surface the strongest signal.

pub fn start(&self, config_path: &Path, port: u16) -> Result<()>

Start the daemon in the background and wait for it to begin accepting connections on port (RFC-024 SP4: verifies the listener came up so a port-bind failure is reported immediately instead of masked by a started message that never resolves).

pub fn stop(&self) -> Result<()>

Stop the daemon.

Resolution order — every source that can prove an oxios daemon is alive is consulted, because each one alone is unreliable:

1. Instance lock (<pid_file>.lock): held by flock by every daemon (cmd_serve/--foreground AND start()-spawned). The PID recorded in the file is the authoritative owner.
2. PID file (<pid_file>): the legacy spawn-and-fork channel. Kept for backwards compat with prior daemons that wrote the pidfile but never took the instance lock.
3. Orphan port probe: when neither (1) nor (2) identifies a daemon, lsof -ti tcp:PORT -sTCP:LISTEN is consulted. This path catches a --foreground debug instance launched without writing a pidfile — the exact symptom the user reported.
4. System service: launchd plist or systemd unit. If loaded, unload it for this session; otherwise KeepAlive (macOS) or Restart=on-failure (Linux) will silently undo our kill within milliseconds. We do NOT remove the plist/unit here; that’s uninstall_service’s job.

User contract: an explicit daemon install is the only way to opt into the supervisor; in that mode stop must reverse it for this session, otherwise the daemon respawns within seconds and stop lies. A user who never ran daemon install and never spawned a daemon in another shell simply gets “not running”.

pub fn restart(&self, config_path: &Path, port: u16) -> Result<()>

Restart the daemon.

pub fn install_service(&self) -> Result<()>

Install as a system service (launchd on macOS, systemd on Linux).

After this returns successfully, stop() MUST be used to halt the daemon in this session — a plain kill -TERM will be undone by KeepAlive (macOS) / Restart=on-failure (Linux) within milliseconds.

pub fn uninstall_service(&self) -> Result<()>

Uninstall the system service.

Tears down any live supervisor registration BEFORE removing the file. On macOS, removing a loaded plist has no effect on the running job — bootout must run first.

pub fn cleanup(&self) -> Result<()>