zesdex/apps/infrastructure/src/bgbash/job.rs

//! Background bash job — spawns a `bash -c` subprocess and tracks its life.

use std::process::{Child, Command, Stdio};
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::{Arc, Mutex};

use tracing::error;

/// A handle to a spawned background bash job.
pub struct BashJob {
    pub id: String,
    pub command: String,
    pub process: Mutex<Option<Child>>,
    pub cancelled: AtomicBool,
}

/// Spawn a background bash job and return a handle.
///
/// The job runs until completion or until `cancel()` is called.
pub fn spawn_bash_job(cmd: String) -> Arc<BashJob> {
    let child = Command::new("bash")
        .arg("-c")
        .arg(&cmd)
        .stdout(Stdio::piped())
        .stderr(Stdio::piped())
        .spawn()
        .ok();

    let job = Arc::new(BashJob {
        id: uuid::Uuid::new_v4().to_string(),
        command: cmd,
        process: Mutex::new(child),
        cancelled: AtomicBool::new(false),
    });

    // Spawn a monitor thread using try_wait() polling so the lock is never
    // held across a blocking wait, allowing cancel() to acquire the lock.
    let job_clone = Arc::clone(&job);
    std::thread::spawn(move || {
        loop {
            let mut exited = false;
            {
                let mut guard = match job_clone.process.lock() {
                    Ok(g) => g,
                    Err(poisoned) => {
                        error!("bgbash job mutex poisoned, recovering");
                        poisoned.into_inner()
                    }
                };
                if let Some(ref mut child) = *guard {
                    match child.try_wait() {
                        Ok(Some(_)) => exited = true,
                        Ok(None) => {} // still running
                        Err(e) => {
                            error!("bgbash wait error: {e}");
                            exited = true;
                        }
                    }
                } else {
                    exited = true; // no child process
                }
            } // lock is dropped here — cancel() can now acquire it

            if exited || job_clone.cancelled.load(Ordering::SeqCst) {
                break;
            }
            std::thread::sleep(std::time::Duration::from_millis(50));
        }
    });

    job
}

impl BashJob {
    pub fn cancel(&self) {
        self.cancelled.store(true, Ordering::SeqCst);
        if let Ok(mut guard) = self.process.lock() {
            if let Some(ref mut child) = *guard {
                let _ = child.kill();
                let _ = child.wait();
            }
        }
    }

    pub fn is_running(&self) -> bool {
        if self.cancelled.load(Ordering::SeqCst) {
            return false;
        }
        let Ok(mut guard) = self.process.lock() else {
            return false;
        };
        guard.as_mut().is_some_and(|c| matches!(c.try_wait(), Ok(None)))
    }
}