use std::os::unix::process::CommandExt;
use std::path::PathBuf;
use std::process::Stdio;
use tokio::fs;
use tokio::process::Command as AsyncCommand;
use crate::model;
use crate::preset;
use crate::ui;
pub async fn verify_health(port: u16) -> bool {
model::verify_health("127.0.0.1", port).await
}
async fn is_llama_server_pid(pid: u32) -> bool {
let output = AsyncCommand::new("ps")
.args(["-p", &pid.to_string(), "-o", "comm="])
.output()
.await;
if let Ok(out) = output {
let comm = String::from_utf8_lossy(&out.stdout).trim().to_string();
comm.contains("llama-server")
} else {
false
}
}
pub async fn serve(
profile: Option<String>,
port: u16,
foreground: bool,
) -> Result<(), color_eyre::Report> {
let target_profile = match profile {
Some(p) => p,
None => {
let presets = preset::list_presets()?;
if presets.is_empty() {
eprintln!("[ERR] No presets found in ~/.config/muthr/presets/");
return Ok(());
}
let names: Vec<&str> = presets.iter().map(|p| p.name.as_str()).collect();
match ui::select_list(&names) {
Some(idx) => presets[idx].name.clone(),
None => {
println!("[INFO] Cancelled.");
return Ok(());
}
}
}
};
let preset_path = preset::resolve_preset(&target_profile)
.ok_or_else(|| color_eyre::eyre::eyre!("Preset not found: {}", target_profile))?;
apply_vram_limits().await;
let home = std::env::var("HOME")?;
let cache_dir = PathBuf::from(&home).join(".cache/muthr");
fs::create_dir_all(&cache_dir).await?;
let tmp_preset = cache_dir.join("active-preset.ini");
let raw_content = fs::read_to_string(&preset_path).await?;
let expanded = raw_content.replace("~", &home);
fs::write(&tmp_preset, expanded).await?;
let preset = preset::parse_preset(&preset_path)?;
let bind_host = preset
.global
.host
.unwrap_or_else(|| "127.0.0.1".to_string());
let server_port = preset.global.port.unwrap_or(port as u32) as u16;
let log_stdout = cache_dir.join("llama-server.log");
let log_stderr = cache_dir.join("llama-server-err.log");
let pid_file = cache_dir.join("llama-server.pid");
if !foreground && pid_file.exists() {
if let Ok(pid_bytes) = fs::read_to_string(&pid_file).await {
if let Ok(old_pid) = pid_bytes.trim().parse::<u32>() {
if is_llama_server_pid(old_pid).await {
eprintln!(
"[WARN] Server already running (PID {}). Stopping first.",
old_pid
);
let _ = stop().await;
fs::remove_file(&pid_file).await.ok();
tokio::time::sleep(std::time::Duration::from_millis(500)).await;
} else {
fs::remove_file(&pid_file).await.ok();
}
}
}
}
let mut args: Vec<String> = vec![
"--host".to_string(),
bind_host.clone(),
"--port".to_string(),
server_port.to_string(),
"--prio".to_string(),
"2".to_string(),
"--n-gpu-layers".to_string(),
preset.global.n_gpu_layers.to_string(),
];
if preset.global.flash_attn {
args.push("--flash-attn".to_string());
args.push("on".to_string());
}
if preset.global.mlock {
args.push("--mlock".to_string());
}
if let Some(slot) = preset.slots.first() {
if let Some(model_path) = &slot.model_path {
let expanded_model = preset::expand_home(model_path);
args.push("--model".to_string());
args.push(expanded_model.to_string_lossy().to_string());
}
if let Some(ctx) = slot.ctx_size {
args.push("--ctx-size".to_string());
args.push(ctx.to_string());
}
if let Some(k) = &slot.cache_type_k {
args.push("--cache-type-k".to_string());
args.push(k.clone());
}
if let Some(v) = &slot.cache_type_v {
args.push("--cache-type-v".to_string());
args.push(v.clone());
}
if let Some(cache_ram) = slot.cache_ram {
args.push("--cache-ram".to_string());
args.push(cache_ram.to_string());
}
if let Some(temp) = slot.temp {
args.push("--temperature".to_string());
args.push(temp.to_string());
}
if let Some(top_p) = slot.top_p {
args.push("--top-p".to_string());
args.push(top_p.to_string());
}
if let Some(min_p) = slot.min_p {
args.push("--min-p".to_string());
args.push(min_p.to_string());
}
if let Some(top_k) = slot.top_k {
args.push("--top-k".to_string());
args.push(top_k.to_string());
}
if let Some(rp) = slot.repeat_penalty {
args.push("--repeat-penalty".to_string());
args.push(rp.to_string());
}
if slot.jinja == Some(true) {
args.push("--jinja".to_string());
}
if let Some(parallel) = slot.parallel {
args.push("--parallel".to_string());
args.push(parallel.to_string());
}
}
if let Some(b) = preset.global.batch_size {
args.push("--batch-size".to_string());
args.push(b.to_string());
}
if let Some(ub) = preset.global.ubatch_size {
args.push("--ubatch-size".to_string());
args.push(ub.to_string());
}
if let Some(t) = preset.global.threads {
args.push("--threads".to_string());
args.push(t.to_string());
}
if let Some(tb) = preset.global.threads_batch {
args.push("--threads-batch".to_string());
args.push(tb.to_string());
}
if foreground {
println!("[PROC] Starting llama.cpp Server (Direct Mode)...");
println!(" Binding Address : http://{}:{}", bind_host, server_port);
println!(" Profile Target : {:?}", preset_path);
println!(" Press Ctrl+C to stop the server.\n");
let mut child = AsyncCommand::new("llama-server")
.args(&args)
.stdout(Stdio::inherit())
.stderr(Stdio::inherit())
.spawn()?;
let status = child.wait().await?;
if !status.success() {
eprintln!("[WARN] Server exited with code: {}", status);
}
Ok(())
} else {
println!("[PROC] Starting llama.cpp Server in background (Direct Mode)...");
println!(" Binding Address : http://{}:{}", bind_host, server_port);
println!(" Profile Target : {:?}", preset_path);
println!(" Log file : {:?}", log_stdout);
println!(" Error log : {:?}", log_stderr);
println!();
let stdout_file = std::fs::OpenOptions::new()
.create(true)
.append(true)
.open(&log_stdout)?;
let stderr_file = std::fs::OpenOptions::new()
.create(true)
.append(true)
.open(&log_stderr)?;
let mut cmd = std::process::Command::new("llama-server");
cmd.args(&args).stdout(stdout_file).stderr(stderr_file);
unsafe {
cmd.pre_exec(|| {
if libc::setsid() == -1 {
return Err(std::io::Error::last_os_error());
}
Ok(())
});
}
match cmd.spawn() {
Ok(c) => {
let pid = c.id();
fs::write(&pid_file, pid.to_string()).await?;
println!("[ OK ] Server started (PID {})", pid);
println!(" Run 'muthr stop' to stop the server.");
Ok(())
}
Err(e) => Err(e.into()),
}
}
}
pub async fn stop() -> Result<(), color_eyre::Report> {
let home = std::env::var("HOME")?;
let cache_dir = PathBuf::from(&home).join(".cache/muthr");
let pid_file = cache_dir.join("llama-server.pid");
if !pid_file.exists() {
println!("[WARN] No PID file found — server may not be running.");
return Ok(());
}
let pid_bytes = fs::read_to_string(&pid_file).await?;
let pid = pid_bytes.trim().parse::<u32>()?;
if !is_llama_server_pid(pid).await {
println!(
"[WARN] PID {} is not llama-server (stale PID file). Cleaning up.",
pid
);
fs::remove_file(&pid_file).await.ok();
return Ok(());
}
println!(
"[PROC] Initiating graceful shutdown (SIGTERM) for PID {}...",
pid
);
let _ = AsyncCommand::new("kill")
.args(["-15", &pid.to_string()])
.output()
.await;
let mut died = false;
for _ in 0..10 {
tokio::time::sleep(std::time::Duration::from_millis(500)).await;
if !is_llama_server_pid(pid).await {
died = true;
break;
}
}
if died {
println!("[ OK ] Server stopped gracefully. VRAM released.");
} else {
eprintln!("[WARN] Server did not respond to SIGTERM. Escalating to SIGKILL.");
let _ = AsyncCommand::new("kill")
.args(["-9", &pid.to_string()])
.output()
.await;
println!("[ OK ] Server force-killed.");
}
fs::remove_file(&pid_file).await.ok();
Ok(())
}
pub async fn status() -> Result<(), color_eyre::Report> {
let home = std::env::var("HOME")?;
let active_path = PathBuf::from(&home).join(".cache/muthr/active-preset.ini");
if !active_path.exists() {
println!("[STATUS] No active profile configured.");
return Ok(());
}
let preset_path = preset::resolve_preset(
&PathBuf::from(&home)
.join(".cache/muthr/active-preset.ini")
.file_stem()
.and_then(|s| s.to_str())
.map(String::from)
.unwrap_or_default(),
);
let preset = if let Some(p) = preset_path {
preset::parse_preset(&p).ok()
} else {
None
};
let preset_basename = preset
.as_ref()
.map(|p| p.name.clone())
.unwrap_or_else(|| "unknown".to_string());
println!("[STATUS] Active profile: {}", preset_basename);
if let Some(p) = &preset {
println!(" Preset: {}", p.path.display());
} else {
println!(" Preset: (not found on disk)");
}
let is_running = AsyncCommand::new("pgrep")
.arg("-x")
.arg("llama-server")
.output()
.await?
.status
.success();
if is_running {
println!(" Server: running");
} else {
println!(" Server: stopped");
}
Ok(())
}
pub fn list() -> Result<(), color_eyre::Report> {
let presets = preset::list_presets()?;
if presets.is_empty() {
println!("[WARN] No presets found in ~/.config/muthr/presets/");
return Ok(());
}
let mut rows: Vec<Vec<String>> = Vec::new();
for p in &presets {
rows.push(vec![
p.name.clone(),
p.path.to_string_lossy().to_string(),
p.slots.len().to_string(),
]);
}
let headers = vec!["Preset", "Path", "Slots"];
match ui::select_table(&headers, rows) {
Some(idx) => println!("{}", presets[idx].name),
None => {
println!("Available presets:");
println!(
"==============================================================================="
);
for p in &presets {
println!(
" {:<30} {} [{}]",
p.name,
p.path.to_string_lossy(),
p.slots.len()
);
}
println!(
"==============================================================================="
);
}
}
Ok(())
}
async fn apply_vram_limits() {
let mem_bytes = sysctl_memsize().await;
let threshold: u64 = 32 * 1024 * 1024 * 1024;
if mem_bytes >= threshold {
let gb = mem_bytes / 1024 / 1024 / 1024;
println!(
"[INFO] High-memory host detected ({}GB). Adjusting wired Metal VRAM limits...",
gb
);
let _ = AsyncCommand::new("sudo")
.args(["sysctl", "-w", "iogpu.wired_limit_mb=43000"])
.output()
.await;
}
}
async fn sysctl_memsize() -> u64 {
let output = AsyncCommand::new("sysctl")
.arg("-n")
.arg("hw.memsize")
.output()
.await;
match output {
Ok(out) if out.status.success() => {
let raw = String::from_utf8_lossy(&out.stdout);
let digits: String = raw.chars().filter(|c| c.is_ascii_digit()).collect();
digits.parse::<u64>().unwrap_or(0)
}
_ => 0,
}
}