feat: add agent module with Nginx service commands and routes

- Introduced a new agent module with commands for managing Nginx configurations.
- Implemented `NginxService` for handling reload, validation, and configuration writing.
- Added routes for status, validation, and configuration writing using Axum.
- Created necessary command files: `reload.rs`, `run.rs`, `validate.rs`, `write_config.rs`.
- Updated `Cargo.toml` and `Cargo.lock` to include new dependencies.
- Added `.gitignore` for the agent module.
- Updated `justfile` to include OpenAPI generation for the agent.

apps/agent/src/commands.rs

@@ -0,0 +1,292 @@
+mod reload;
+mod run;
+mod validate;
+mod write_config;
+
+use std::{
+    collections::HashMap,
+    sync::{
+        Arc,
+        atomic::{AtomicU64, Ordering},
+    },
+};
+
+use tokio::sync::{Mutex, RwLock};
+use tokio_cron_scheduler::{Job, JobScheduler};
+use tracing::{error, info};
+
+use crate::commands::write_config::INTERNAL_CONFIG_FOLDER_NAME;
+
+const OLD_CONFIG_CLEANUP_THRESHOLD: u64 = 3600;
+
+pub struct NginxService {
+    // lock for nginx reload, and timestamp tracking
+    nginx_lock: Mutex<()>,
+    last_applied: AtomicU64,
+    // lock for write_config per (config_name, timestamp)
+    #[allow(clippy::type_complexity)]
+    write_config_lock: RwLock<HashMap<(String, u64), Arc<RwLock<()>>>>,
+    // commands
+    reload_cmd: Arc<reload::ReloadCommand>,
+    validate_cmd: Arc<validate::ValidateCommand>,
+    write_config_cmd: Arc<write_config::WriteConfigCommand>,
+}
+
+impl NginxService {
+    pub async fn new(
+        scheduler: Arc<JobScheduler>,
+        nginx_config_dir: std::path::PathBuf,
+    ) -> Result<Arc<Self>, Box<dyn std::error::Error + Send + Sync>> {
+        let nginx_service = Arc::new(NginxService {
+            nginx_lock: Mutex::new(()),
+            last_applied: AtomicU64::new(0),
+            write_config_lock: RwLock::new(HashMap::new()),
+            // commands
+            reload_cmd: Arc::new(reload::ReloadCommand::default()),
+            validate_cmd: Arc::new(validate::ValidateCommand::new(nginx_config_dir.clone())),
+            write_config_cmd: Arc::new(write_config::WriteConfigCommand::new(nginx_config_dir)),
+        });
+        let mut nginx_service_clone = nginx_service.clone();
+
+        scheduler
+            .clone()
+            // cleanup every 10 minutes
+            .add(Job::new_async("0 */10 * * * *", move |_uuid, _l| {
+                info!("Running nginx_service cleanup job");
+                let nginx_service_clone = nginx_service_clone.clone();
+                let job = Box::pin(async move {
+                    nginx_service_clone.cleanup_unused_lock().await;
+                });
+                info!("NginxService cleanup job completed");
+                job
+            })?)
+            .await?;
+
+        nginx_service_clone = nginx_service.clone();
+
+        scheduler
+            .clone()
+            // cleanup every hour
+            .add(Job::new_async("0 0 */1 * * *", move |_uuid, _l| {
+                info!("Running nginx_service old config cleanup job");
+                let nginx_service_clone = nginx_service_clone.clone();
+                let job = Box::pin(async move {
+                    nginx_service_clone.cleanup_old_configs().await;
+                });
+                info!("NginxService old config cleanup job completed");
+                job
+            })?)
+            .await?;
+
+        Ok(nginx_service)
+    }
+
+    pub async fn validate_and_reload(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+    ) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+        let cur = self.last_applied.load(Ordering::SeqCst);
+        if cur > timestamp {
+            return Err("Another operation is in progress with higher timestamp value".into());
+        }
+
+        // acquire write lock to update nginx_lock
+        let _nginx_guard = self.nginx_lock.lock().await;
+        // acquire write lock for this config+timestamp
+        let rw_lock = self.acquire_file_write_lock(config_name, timestamp).await;
+        let _guard = rw_lock.write().await;
+
+        match self
+            .reload_cmd
+            .validate_and_reload(config_name, timestamp, self.validate_cmd.clone())
+            .await
+        {
+            Ok((code, output)) => {
+                // update last_applied
+                self.last_applied.store(timestamp, Ordering::SeqCst);
+                Ok((code, output))
+            }
+            Err(e) => Err(e),
+        }
+    }
+
+    pub async fn write_config(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+        content: &str,
+    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let rw_lock = self.acquire_file_write_lock(config_name, timestamp).await;
+        let _guard = rw_lock.write().await;
+        // call the write_config command
+        self.write_config_cmd
+            .write_config(config_name, timestamp, content)
+            .await
+    }
+
+    pub async fn validate(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+    ) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+        self.validate_cmd.validate(config_name, timestamp).await
+    }
+
+    async fn cleanup_unused_lock(&self) {
+        let mut _write_lock = self.write_config_lock.write().await;
+        (*_write_lock).retain(|_, lock| {
+            // retain only locks that are currently held (readers or writers)
+            lock.try_write().is_err()
+        });
+    }
+
+    async fn cleanup_old_configs(&self) {
+        // list all files within nginx_config_dir/YANPM that is older than now - OLD_CONFIG_CLEANUP_THRESHOLD
+        let cutoff = std::time::SystemTime::now()
+            .duration_since(std::time::UNIX_EPOCH)
+            .unwrap()
+            .as_secs()
+            - OLD_CONFIG_CLEANUP_THRESHOLD;
+
+        let nginx_config_dir = self.validate_cmd.nginx_config_dir();
+        let yanpm_dir = nginx_config_dir.join(INTERNAL_CONFIG_FOLDER_NAME);
+
+        let read_dir = match tokio::fs::read_dir(&yanpm_dir).await {
+            Ok(rd) => rd,
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
+                // directory does not exist, nothing to clean up
+                return;
+            }
+            Err(e) => {
+                error!(
+                    "Error reading {} config directory {}: {}",
+                    INTERNAL_CONFIG_FOLDER_NAME,
+                    yanpm_dir.display(),
+                    e
+                );
+                return;
+            }
+        };
+
+        tokio::pin!(read_dir);
+        while let Some(entry) = read_dir.next_entry().await.unwrap_or(None) {
+            let metadata = match entry.metadata().await {
+                Ok(md) => md,
+                Err(e) => {
+                    error!(
+                        "Error getting metadata for file {}: {}",
+                        entry.path().display(),
+                        e
+                    );
+                    continue;
+                }
+            };
+            if let Ok(modified) = metadata.modified()
+                && let Ok(duration) = modified.duration_since(std::time::UNIX_EPOCH)
+            {
+                let mtime_secs = duration.as_secs();
+                if mtime_secs < cutoff {
+                    // file is older than cutoff, remove it
+                    if let Err(e) = tokio::fs::remove_file(entry.path()).await {
+                        error!(
+                            "Error removing old config file {}: {}",
+                            entry.path().display(),
+                            e
+                        );
+                    } else {
+                        info!("Removed old config file {}", entry.path().display());
+                    }
+                }
+            }
+        }
+    }
+
+    async fn acquire_file_write_lock(&self, config_name: &str, timestamp: u64) -> Arc<RwLock<()>> {
+        let mut write_lock = self.write_config_lock.write().await;
+        write_lock
+            .entry((config_name.to_string(), timestamp))
+            .or_insert_with(|| Arc::new(RwLock::new(())))
+            .clone()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::error::Error;
+    use std::sync::Arc as StdArc;
+    use tokio::time::{Duration, sleep};
+
+    impl NginxService {
+        // Test helper that simulates a long-running reload without invoking external commands.
+        pub async fn test_simulated_reload(
+            &self,
+            config_name: &str,
+            timestamp: u64,
+            delay_ms: u64,
+        ) -> Result<(), Box<dyn Error + Send + Sync>> {
+            // pre-check
+            let cur = self.last_applied.load(Ordering::SeqCst);
+            if cur >= timestamp {
+                return Err("stale".into());
+            }
+
+            // acquire exclusive lock and re-check
+            let _nginx_guard = self.nginx_lock.lock().await;
+            let cur2 = self.last_applied.load(Ordering::SeqCst);
+            if cur2 >= timestamp {
+                return Err("stale".into());
+            }
+
+            // per-file lock
+            let rw_lock = self.acquire_file_write_lock(config_name, timestamp).await;
+            let _guard = rw_lock.write().await;
+
+            // simulate operation
+            sleep(Duration::from_millis(delay_ms)).await;
+
+            // on success update last_applied
+            let mut prev = self.last_applied.load(Ordering::SeqCst);
+            while prev < timestamp {
+                match self.last_applied.compare_exchange(
+                    prev,
+                    timestamp,
+                    Ordering::SeqCst,
+                    Ordering::SeqCst,
+                ) {
+                    Ok(_) => break,
+                    Err(next) => prev = next,
+                }
+            }
+
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn concurrent_stale_is_rejected() {
+        let scheduler = StdArc::new(JobScheduler::new().await.unwrap());
+        let svc = NginxService::new(scheduler.clone(), std::env::temp_dir())
+            .await
+            .unwrap();
+
+        let s1 = svc.clone();
+        let h1 = tokio::spawn(async move { s1.test_simulated_reload("cfg", 2, 200).await });
+
+        // let second start shortly after first so it will wait for the mutex
+        sleep(Duration::from_millis(20)).await;
+
+        let s2 = svc.clone();
+        let h2 = tokio::spawn(async move { s2.test_simulated_reload("cfg", 1, 10).await });
+
+        let r1 = h1.await.unwrap();
+        assert!(r1.is_ok(), "first (newer) task should succeed");
+
+        let r2 = h2.await.unwrap();
+        assert!(
+            r2.is_err(),
+            "second (older) task should be rejected as stale"
+        );
+    }
+}

apps/agent/src/commands/reload.rs

@@ -0,0 +1,98 @@
+use std::path::Path;
+use std::sync::Arc;
+use std::time::{SystemTime, UNIX_EPOCH};
+
+use tokio::sync::Mutex;
+use tracing::error;
+
+use crate::commands::write_config::INTERNAL_CONFIG_FOLDER_NAME;
+use crate::commands::{run::run_cmd, validate::ValidateCommand};
+
+pub struct ReloadCommand {
+    is_reloading: Mutex<bool>,
+}
+
+struct ReloadResetGuard<'a> {
+    guard: tokio::sync::MutexGuard<'a, bool>,
+}
+
+impl<'a> Drop for ReloadResetGuard<'a> {
+    fn drop(&mut self) {
+        *self.guard = false;
+    }
+}
+
+impl Default for ReloadCommand {
+    fn default() -> Self {
+        Self {
+            is_reloading: Mutex::new(false),
+        }
+    }
+}
+
+impl ReloadCommand {
+    pub async fn validate_and_reload(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+        validate_cmd: Arc<ValidateCommand>,
+    ) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+        // ensure the written fragment exists
+        validate_cmd.validate(config_name, timestamp).await?;
+
+        // Now atomically swap the YANPM.conf symlink to point to the new fragment
+        // so nginx -t validates the composed main config. If validation fails,
+        // attempt to restore the previous symlink.
+        let filename = crate::commands::run::to_file_name(config_name, timestamp)?;
+        let nginx_dir = validate_cmd.nginx_config_dir();
+        let symlink_path = nginx_dir.join("YANPM.conf");
+        let now = SystemTime::now().duration_since(UNIX_EPOCH)?.as_nanos();
+        let tmp_name = format!("YANPM.conf.tmp.{}.{}", std::process::id(), now);
+        let tmp_path = nginx_dir.join(&tmp_name);
+
+        // prepare relative target: INTERNAL_CONFIG_FOLDER_NAME/<filename>
+        let rel_target = Path::new(INTERNAL_CONFIG_FOLDER_NAME).join(&filename);
+
+        // read previous target if exists
+        let previous_target = std::fs::read_link(&symlink_path).ok();
+
+        // Acquire reload guard before mutating the symlink to avoid races
+        let reloading_lock = self.is_reloading.lock().await;
+        if *reloading_lock {
+            return Err("Reload already in progress".into());
+        }
+        // set flag to true and ensure it is reset on drop
+        let mut mut_guard = reloading_lock;
+        *mut_guard = true;
+        let _reset_guard = ReloadResetGuard { guard: mut_guard };
+
+        // create temporary symlink and atomically rename into place
+        std::os::unix::fs::symlink(&rel_target, &tmp_path)?;
+        tokio::fs::rename(&tmp_path, &symlink_path).await?;
+
+        // validate composed main config now that symlink points to new fragment
+        if let Err(e) = validate_cmd.validate_all().await {
+            // restore previous symlink state while still holding the guard
+            if let Some(prev) = previous_target {
+                let restore_tmp =
+                    nginx_dir.join(format!("YANPM.conf.restore.{}.{}", std::process::id(), now));
+                std::os::unix::fs::symlink(&prev, &restore_tmp)?;
+                if let Err(err) = tokio::fs::rename(&restore_tmp, &symlink_path).await {
+                    error!(
+                        "Failed to restore previous YANPM.conf symlink after validation error: {}",
+                        err
+                    );
+                }
+            } else if let Err(err) = tokio::fs::remove_file(&symlink_path).await {
+                error!(
+                    "Failed to remove YANPM.conf symlink after validation error: {}",
+                    err
+                );
+            }
+            return Err(e);
+        }
+
+        // reload the running nginx master process (no -c) so it reloads its configured main config
+        run_cmd("nginx", &["-s", "reload"], 10).await
+    }
+}

apps/agent/src/commands/run.rs

@@ -0,0 +1,85 @@
+use std::time::Duration;
+
+use tokio::{process::Command, time::timeout};
+use tracing::error;
+
+pub fn to_file_name(
+    config_name: &str,
+    timestamp: u64,
+) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
+    // reject empty or unsafe names to avoid path traversal or invalid filesystem chars
+    if config_name.is_empty() {
+        return Err("config_name is empty".into());
+    }
+    if config_name.len() > 255 {
+        return Err("config_name too long".into());
+    }
+    if config_name.contains('/') || config_name.contains('\\') || config_name.contains("..") {
+        return Err("config_name contains invalid path characters".into());
+    }
+    if !config_name
+        .chars()
+        .all(|c| c.is_ascii_alphanumeric() || "-._".contains(c))
+    {
+        return Err("config_name contains invalid characters".into());
+    }
+
+    Ok(format!("{}_{}.conf", timestamp, config_name))
+}
+
+pub async fn run_cmd(
+    cmd: &str,
+    args: &[&str],
+    dur_s: u64,
+) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+    let mut c = Command::new(cmd);
+    c.args(args);
+    let res = timeout(Duration::from_secs(dur_s), c.output()).await;
+    let out = match res {
+        Ok(Ok(out)) => out,
+        Ok(Err(e)) => return Err(Box::new(e)),
+        Err(_) => {
+            return Err(Box::new(std::io::Error::new(
+                std::io::ErrorKind::TimedOut,
+                "command timeout",
+            )));
+        }
+    };
+    let code = out.status.code().unwrap_or(-1);
+    let output = String::from_utf8_lossy(&[out.stdout, out.stderr].concat()).to_string();
+    if code != 0 {
+        error!("command failed ({}): {}", code, output);
+        return Err(format!("command failed ({}): {}", code, output).into());
+    }
+    Ok((code, output))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::to_file_name;
+
+    #[test]
+    fn to_file_name_valid() {
+        let res = to_file_name("myconf", 1234).expect("should succeed");
+        assert_eq!(res, "1234_myconf.conf");
+    }
+
+    #[test]
+    fn to_file_name_empty() {
+        assert!(to_file_name("", 1).is_err());
+    }
+
+    #[test]
+    fn to_file_name_invalid_chars() {
+        assert!(to_file_name("bad/name", 1).is_err());
+        assert!(to_file_name("bad\\name", 1).is_err());
+        assert!(to_file_name("bad..name", 1).is_err());
+        assert!(to_file_name("bad$name", 1).is_err());
+    }
+
+    #[test]
+    fn to_file_name_too_long() {
+        let long = "a".repeat(300);
+        assert!(to_file_name(&long, 1).is_err());
+    }
+}

apps/agent/src/commands/validate.rs

@@ -0,0 +1,47 @@
+use crate::commands::{run::run_cmd, write_config::INTERNAL_CONFIG_FOLDER_NAME};
+use std::path::PathBuf;
+
+pub struct ValidateCommand {
+    nginx_config_dir: PathBuf,
+}
+
+impl ValidateCommand {
+    pub fn new(nginx_config_dir: PathBuf) -> Self {
+        Self { nginx_config_dir }
+    }
+
+    pub fn nginx_config_dir(&self) -> PathBuf {
+        self.nginx_config_dir.clone()
+    }
+
+    pub async fn validate_all(
+        &self,
+    ) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+        run_cmd("nginx", &["-t"], 10).await
+    }
+
+    pub async fn validate(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+    ) -> Result<(i32, String), Box<dyn std::error::Error + Send + Sync>> {
+        let filename = crate::commands::run::to_file_name(config_name, timestamp)?;
+        // fragments are written into the YANPM subdirectory
+        let full_path = self
+            .nginx_config_dir
+            .join(INTERNAL_CONFIG_FOLDER_NAME)
+            .join(&filename);
+
+        // ensure the fragment file exists
+        if tokio::fs::metadata(&full_path).await.is_err() {
+            return Err(format!("Config file not found: {}", full_path.display()).into());
+        }
+
+        run_cmd(
+            "nginx",
+            &["-t", "-c", full_path.to_str().ok_or("invalid config path")?],
+            10,
+        )
+        .await
+    }
+}

apps/agent/src/commands/write_config.rs

@@ -0,0 +1,131 @@
+use std::os::unix::fs::PermissionsExt;
+use std::path::PathBuf;
+use std::time::{SystemTime, UNIX_EPOCH};
+use tokio::io::AsyncWriteExt;
+
+use crate::commands::run::to_file_name;
+
+pub const INTERNAL_CONFIG_FOLDER_NAME: &str = "YANPM";
+const FILE_SIZE_LIMIT: usize = 10 * 1024 * 1024; // 10MB
+
+pub struct WriteConfigCommand {
+    nginx_config_dir: PathBuf,
+}
+
+impl WriteConfigCommand {
+    pub fn new(nginx_config_dir: PathBuf) -> Self {
+        Self { nginx_config_dir }
+    }
+    pub async fn write_config(
+        &self,
+        config_name: &str,
+        timestamp: u64,
+        content: &str,
+    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let filename = to_file_name(config_name, timestamp)?;
+        let path = self.nginx_config_dir.clone();
+        // ensure main config dir exists
+        tokio::fs::create_dir_all(&path).await?;
+
+        // create YANPM subdir where fragment files live
+        let yanpm_dir = path.join(INTERNAL_CONFIG_FOLDER_NAME);
+        tokio::fs::create_dir_all(&yanpm_dir).await?;
+        let final_path = yanpm_dir.join(&filename);
+
+        // limit size to 10MB
+        if content.len() > FILE_SIZE_LIMIT {
+            return Err(format!(
+                "content exceeds {}MB size limit",
+                FILE_SIZE_LIMIT / (1024 * 1024)
+            )
+            .into());
+        }
+
+        // create a temporary filename in the same directory for atomic replace
+        let now = SystemTime::now().duration_since(UNIX_EPOCH)?.as_nanos();
+        let tmp_filename = format!("{}.tmp.{}.{}", filename, std::process::id(), now);
+        // create tmp file in the same directory as final file to ensure atomic rename
+        let tmp_path = yanpm_dir.join(tmp_filename);
+
+        let mut file = tokio::fs::OpenOptions::new()
+            .create(true)
+            .write(true)
+            .truncate(true)
+            .open(&tmp_path)
+            .await?;
+        file.write_all(content.as_bytes()).await?;
+        // ensure data is flushed to disk; propagate errors
+        file.sync_all().await?;
+
+        // atomically move the tmp file into the YANPM dir
+        tokio::fs::rename(&tmp_path, &final_path).await?;
+
+        // set explicit permissions (rw-r-----)
+        tokio::fs::set_permissions(&final_path, std::fs::Permissions::from_mode(0o640)).await?;
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{INTERNAL_CONFIG_FOLDER_NAME, WriteConfigCommand};
+    use std::time::SystemTime;
+    use std::time::UNIX_EPOCH;
+
+    #[tokio::test]
+    async fn write_config_success_and_cleanup() {
+        let base = std::env::temp_dir().join(format!(
+            "yanpm_test_{}_{}",
+            std::process::id(),
+            SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .unwrap()
+                .as_nanos()
+        ));
+        // ensure clean
+        let _ = tokio::fs::remove_dir_all(&base).await;
+        let cmd = WriteConfigCommand::new(base.clone());
+
+        let config_name = "unittest";
+        let timestamp = 42u64;
+        let content = "hello world";
+
+        cmd.write_config(config_name, timestamp, content)
+            .await
+            .expect("write should succeed");
+
+        let filename = super::to_file_name(config_name, timestamp).unwrap();
+        let final_path = base.join(INTERNAL_CONFIG_FOLDER_NAME).join(&filename);
+        let data = tokio::fs::read_to_string(&final_path)
+            .await
+            .expect("file should exist");
+        assert_eq!(data, content);
+
+        // cleanup
+        tokio::fs::remove_dir_all(&base).await.expect("cleanup");
+    }
+
+    #[tokio::test]
+    async fn write_config_size_limit() {
+        let base = std::env::temp_dir().join(format!(
+            "yanpm_test_{}_{}",
+            std::process::id(),
+            SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .unwrap()
+                .as_nanos()
+        ));
+        let _ = tokio::fs::remove_dir_all(&base).await;
+        let cmd = WriteConfigCommand::new(base.clone());
+
+        // exceed 10MB limit
+        let large = vec![b'a'; 10 * 1024 * 1024 + 1];
+        let large_str = String::from_utf8_lossy(&large).to_string();
+
+        let res = cmd.write_config("big", 1, &large_str).await;
+        assert!(res.is_err());
+
+        let _ = tokio::fs::remove_dir_all(&base).await;
+    }
+}

apps/agent/src/main.rs

@@ -0,0 +1,83 @@
+#![forbid(unsafe_code)]
+
+mod commands;
+mod routes;
+
+use axum::routing::get;
+use axum::{Router, routing::post};
+use clap::Parser;
+use std::os::unix::fs::PermissionsExt;
+use std::path::PathBuf;
+use std::sync::Arc;
+use tokio::net::UnixListener;
+use tracing::{error, info};
+
+use crate::commands::NginxService;
+use crate::routes::{status, validate, validate_and_reload, write_config};
+
+#[derive(Parser)]
+struct Args {
+    /// Unix socket path to bind the daemon to
+    sock: String,
+
+    /// Directory where generated nginx config files will be written
+    #[arg(long, default_value = "/etc/nginx/conf.d")]
+    nginx_config_dir: PathBuf,
+}
+
+#[tokio::main]
+async fn main() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let args = Args::parse();
+    let sock = args.sock;
+
+    let path = PathBuf::from(&sock);
+    if let Some(dir) = path.parent() {
+        tokio::fs::create_dir_all(dir).await?;
+        // permissive; set tighter perms in production via image/build steps
+        tokio::fs::set_permissions(dir, std::fs::Permissions::from_mode(0o770)).await?;
+    }
+    // If an existing path exists at the socket location, ensure it's a socket
+    match tokio::fs::metadata(&path).await {
+        Ok(md) => {
+            use std::os::unix::fs::FileTypeExt;
+            if md.file_type().is_socket() {
+                tokio::fs::remove_file(&path).await?;
+            } else {
+                return Err(
+                    format!("Socket path {} exists and is not a socket", path.display()).into(),
+                );
+            }
+        }
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+        Err(e) => return Err(e.into()),
+    }
+
+    // bind using tokio's UnixListener (avoids converting a blocking std listener)
+    let listener = UnixListener::bind(&path)?;
+    // set socket perms to 0660 (best-effort)
+    if let Err(err) =
+        tokio::fs::set_permissions(&path, std::fs::Permissions::from_mode(0o660)).await
+    {
+        error!(
+            "Warning: failed to set permissions on socket {}: {}",
+            path.display(),
+            err
+        );
+    }
+
+    let scheduler = Arc::new(tokio_cron_scheduler::JobScheduler::new().await?);
+
+    let app = Router::new()
+        .route("/status", get(status))
+        .route("/validate_and_reload", post(validate_and_reload))
+        .route("/validate", post(validate))
+        .route("/write_config", post(write_config))
+        .with_state(NginxService::new(scheduler.clone(), args.nginx_config_dir).await?);
+
+    scheduler.start().await?;
+
+    info!("Starting yanpm-daemon on unix socket: {}", sock);
+    axum::serve::serve(listener, app).await?;
+
+    Ok(())
+}

apps/agent/src/routes.rs

@@ -0,0 +1,130 @@
+use axum::Json;
+use axum::extract::State;
+use axum::http::StatusCode;
+use axum::response::IntoResponse;
+use serde::{Deserialize, Serialize};
+use serde_json::{Value, from_value};
+use std::sync::Arc;
+use tracing::warn;
+
+use crate::commands::NginxService;
+
+#[derive(Serialize)]
+pub struct StatusResp {
+    pub ok: bool,
+}
+
+pub async fn status() -> impl IntoResponse {
+    let resp = StatusResp { ok: true };
+    (axum::http::StatusCode::OK, axum::Json(resp))
+}
+
+#[derive(Serialize)]
+pub struct ValidateAndReloadResp {
+    pub rc: i32,
+    pub ro: String,
+}
+
+#[derive(Deserialize)]
+pub struct ValidateBody {
+    config_name: String,
+    timestamp: u64,
+}
+
+pub async fn validate(
+    State(nginx_controller): State<Arc<NginxService>>,
+    Json(payload): Json<Value>,
+) -> impl IntoResponse {
+    let params: ValidateBody = match from_value(payload) {
+        Ok(req) => req,
+        Err(e) => {
+            warn!("Invalid validate request: {}", e);
+            return (StatusCode::BAD_REQUEST).into_response();
+        }
+    };
+
+    let (_code, _output) = match nginx_controller
+        .validate(&params.config_name, params.timestamp)
+        .await
+    {
+        Ok(res) => res,
+        Err(e) => {
+            let resp = serde_json::json!({ "error": e.to_string() });
+            return (StatusCode::INTERNAL_SERVER_ERROR, axum::Json(resp)).into_response();
+        }
+    };
+
+    (axum::http::StatusCode::OK,).into_response()
+}
+
+#[derive(Deserialize)]
+pub struct ValidateAndReloadBody {
+    config_name: String,
+    timestamp: u64,
+}
+
+pub async fn validate_and_reload(
+    State(nginx_controller): State<Arc<NginxService>>,
+    Json(payload): Json<Value>,
+) -> impl IntoResponse {
+    let params: ValidateAndReloadBody = match from_value(payload) {
+        Ok(req) => req,
+        Err(e) => {
+            warn!("Invalid validate_and_reload request: {}", e);
+            return (StatusCode::BAD_REQUEST).into_response();
+        }
+    };
+
+    let (code, output) = match nginx_controller
+        .validate_and_reload(&params.config_name, params.timestamp)
+        .await
+    {
+        Ok(res) => res,
+        Err(e) => {
+            let resp = ValidateAndReloadResp {
+                rc: -1,
+                ro: e.to_string(),
+            };
+            return (StatusCode::INTERNAL_SERVER_ERROR, axum::Json(resp)).into_response();
+        }
+    };
+
+    let resp = ValidateAndReloadResp {
+        rc: code,
+        ro: output,
+    };
+    (axum::http::StatusCode::OK, axum::Json(resp)).into_response()
+}
+
+#[derive(Deserialize)]
+pub struct WriteConfigBody {
+    config_name: String,
+    timestamp: u64,
+    content: String,
+}
+
+pub async fn write_config(
+    State(nginx_controller): State<Arc<NginxService>>,
+    Json(payload): Json<Value>,
+) -> impl IntoResponse {
+    let body: WriteConfigBody = match from_value(payload) {
+        Ok(req) => req,
+        Err(e) => {
+            warn!("Invalid write_config request: {}", e);
+            return (StatusCode::BAD_REQUEST).into_response();
+        }
+    };
+
+    match nginx_controller
+        .write_config(&body.config_name, body.timestamp, &body.content)
+        .await
+    {
+        Ok(_) => (),
+        Err(e) => {
+            let resp = serde_json::json!({ "error": e.to_string() });
+            return (StatusCode::INTERNAL_SERVER_ERROR, axum::Json(resp)).into_response();
+        }
+    };
+
+    (axum::http::StatusCode::OK,).into_response()
+}