corrosion-admin-panel/corrosion-host-agent/src/docker_compose.rs

//! Docker-compose instance supervision — the Dune: Awakening adapter.
//!
//! Dune does not ship as a SteamCMD-updated process like Rust/Conan/Soulmask;
//! it runs as Docker container(s) (game server + RabbitMQ broker + Postgres),
//! orchestrated as a compose stack (a "battlegroup"). So Dune lifecycle is
//! `docker compose up -d / stop / restart` against the instance's compose
//! project, not a spawned OS process. This supervisor implements the same
//! [`Supervisor`] trait `ProcessSupervisor` does, so the instance command
//! dispatch is identical — only the management model differs.
//!
//! Scope (first cut): lifecycle + cached state. Two parity items are deferred
//! to Phase 3b alongside process PID adoption: (1) crash detection (containers
//! give us no child handle — a `docker compose ps` poll loop would supply it);
//! (2) state adoption on agent restart (a running stack reports `stopped` until
//! the next lifecycle command). Both are reconcilable with a `ps` probe.
//!
//! Reference: docs/reference-repos/icehunter SETUP_DOCKER.md (the docker
//! control plane this mirrors).

use std::path::PathBuf;
use std::process::Stdio;
use std::sync::Arc;
use std::time::Instant;

use anyhow::{bail, Context, Result};
use serde::Deserialize;
use tokio::process::Command;
use tokio::sync::{watch, Mutex};

use crate::config::InstanceConfig;
use crate::supervisor::{InstanceState, Supervisor};

/// Per-instance docker-compose settings (`[instance.docker_compose]`). All
/// fields optional — defaults cover the common "one compose file in the
/// instance root" case.
#[derive(Debug, Clone, Default, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct DockerComposeConfig {
    /// Compose file (`-f`). Relative paths resolve against the run dir. Default:
    /// compose's own discovery (docker-compose.yml in the run dir).
    #[serde(default)]
    pub file: Option<PathBuf>,
    /// Compose project name (`-p`). Default: the instance id.
    #[serde(default)]
    pub project: Option<String>,
    /// Limit lifecycle ops to one service. Default: every service in the file.
    #[serde(default)]
    pub service: Option<String>,
    /// Override the compose binary invocation. Default: `["docker","compose"]`.
    /// Use `["docker-compose"]` for the legacy standalone binary.
    #[serde(default)]
    pub command: Option<Vec<String>>,
}

struct Inner {
    started_at: Option<Instant>,
}

pub struct DockerComposeSupervisor {
    instance_id: String,
    /// Directory the compose commands run in (relative `-f`/file paths resolve
    /// against it).
    run_dir: PathBuf,
    compose_file: Option<PathBuf>,
    project: String,
    service: Option<String>,
    /// Compose binary + leading args, e.g. `["docker","compose"]`.
    command: Vec<String>,
    inner: Mutex<Inner>,
    state_tx: watch::Sender<InstanceState>,
}

impl DockerComposeSupervisor {
    pub fn new(cfg: &InstanceConfig) -> Arc<Self> {
        let dc = cfg.docker_compose.clone().unwrap_or_default();
        let run_dir = cfg
            .working_dir
            .clone()
            .unwrap_or_else(|| cfg.root.clone());
        let command = dc
            .command
            .filter(|c| !c.is_empty())
            .unwrap_or_else(|| vec!["docker".to_string(), "compose".to_string()]);
        let (state_tx, _) = watch::channel(InstanceState::Stopped);
        Arc::new(Self {
            instance_id: cfg.id.clone(),
            run_dir,
            compose_file: dc.file,
            project: dc.project.unwrap_or_else(|| cfg.id.clone()),
            service: dc.service,
            command,
            inner: Mutex::new(Inner { started_at: None }),
            state_tx,
        })
    }

    fn set_state(&self, state: InstanceState) {
        let _ = self.state_tx.send_replace(state);
    }

    /// Run one compose subcommand (`up`/`stop`/`restart`/...), bailing with the
    /// captured stderr on non-zero exit. Global flags (`-f`, `-p`) precede the
    /// subcommand; the optional single service is appended last.
    async fn run(&self, action: &str, action_args: &[&str]) -> Result<()> {
        let mut cmd = Command::new(&self.command[0]);
        cmd.args(&self.command[1..]);
        if let Some(file) = &self.compose_file {
            cmd.arg("-f").arg(file);
        }
        cmd.arg("-p").arg(&self.project);
        cmd.arg(action);
        cmd.args(action_args);
        if let Some(service) = &self.service {
            cmd.arg(service);
        }
        cmd.current_dir(&self.run_dir)
            .stdin(Stdio::null())
            .stdout(Stdio::piped())
            .stderr(Stdio::piped());

        let output = cmd
            .output()
            .await
            .with_context(|| format!("running `{} {action}` (is docker installed and on PATH?)", self.command.join(" ")))?;

        if !output.status.success() {
            let stderr = String::from_utf8_lossy(&output.stderr);
            let stdout = String::from_utf8_lossy(&output.stdout);
            let detail = if !stderr.trim().is_empty() {
                stderr.trim()
            } else {
                stdout.trim()
            };
            bail!("compose {action} failed ({}): {detail}", output.status);
        }
        Ok(())
    }
}

#[async_trait::async_trait]
impl Supervisor for DockerComposeSupervisor {
    fn instance_id(&self) -> &str {
        &self.instance_id
    }

    fn state(&self) -> InstanceState {
        self.state_tx.borrow().clone()
    }

    fn watch_state(&self) -> watch::Receiver<InstanceState> {
        self.state_tx.subscribe()
    }

    async fn uptime_seconds(&self) -> u64 {
        let inner = self.inner.lock().await;
        match (&*self.state_tx.borrow(), inner.started_at) {
            (InstanceState::Running, Some(t)) => t.elapsed().as_secs(),
            _ => 0,
        }
    }

    async fn start(self: Arc<Self>) -> Result<()> {
        if matches!(
            *self.state_tx.borrow(),
            InstanceState::Running | InstanceState::Starting
        ) {
            bail!("instance '{}' is already running", self.instance_id);
        }
        self.set_state(InstanceState::Starting);
        match self.run("up", &["-d"]).await {
            Ok(()) => {
                self.inner.lock().await.started_at = Some(Instant::now());
                self.set_state(InstanceState::Running);
                tracing::info!("instance '{}' compose up -d", self.instance_id);
                Ok(())
            }
            Err(e) => {
                self.set_state(InstanceState::Stopped);
                Err(e)
            }
        }
    }

    async fn stop(self: Arc<Self>) -> Result<()> {
        self.set_state(InstanceState::Stopping);
        match self.run("stop", &[]).await {
            Ok(()) => {
                self.inner.lock().await.started_at = None;
                self.set_state(InstanceState::Stopped);
                tracing::info!("instance '{}' compose stop", self.instance_id);
                Ok(())
            }
            Err(e) => {
                // Stop failed — the stack is most likely still up.
                self.set_state(InstanceState::Running);
                Err(e)
            }
        }
    }

    async fn restart(self: Arc<Self>) -> Result<()> {
        self.set_state(InstanceState::Starting);
        match self.run("restart", &[]).await {
            Ok(()) => {
                self.inner.lock().await.started_at = Some(Instant::now());
                self.set_state(InstanceState::Running);
                tracing::info!("instance '{}' compose restart", self.instance_id);
                Ok(())
            }
            Err(e) => {
                self.set_state(InstanceState::Stopped);
                Err(e)
            }
        }
    }
}