1pub mod model_manager;
2pub mod routes;
3pub mod state;
4
5#[cfg(test)]
6mod routes_test;
7
8use anyhow::Result;
9use mold_core::{Config, ModelPaths};
10use std::net::SocketAddr;
11use std::path::PathBuf;
12use tokio::net::TcpListener;
13use tower_http::cors::CorsLayer;
14use tower_http::trace::TraceLayer;
15use tracing::info;
16
17pub async fn run_server(bind: &str, port: u16, models_dir: PathBuf) -> Result<()> {
18 Config::install_runtime_models_dir_override(models_dir.clone());
19
20 let mut config = Config::load_or_default();
21 config.models_dir = models_dir.to_string_lossy().into_owned();
22 let model_name = config.default_model.clone();
23
24 let state = match ModelPaths::resolve(&model_name, &config) {
25 Some(paths) => {
26 info!(model = %model_name, "configured model");
27 info!(transformer = %paths.transformer.display());
28 info!(vae = %paths.vae.display());
29 if let Some(t5) = &paths.t5_encoder {
30 info!(t5 = %t5.display());
31 }
32 if let Some(clip) = &paths.clip_encoder {
33 info!(clip = %clip.display());
34 }
35 if let Some(t5_tok) = &paths.t5_tokenizer {
36 info!(t5_tok = %t5_tok.display());
37 }
38 if let Some(clip_tok) = &paths.clip_tokenizer {
39 info!(clip_tok = %clip_tok.display());
40 }
41 if let Some(clip2) = &paths.clip_encoder_2 {
42 info!(clip2 = %clip2.display());
43 }
44 if let Some(clip2_tok) = &paths.clip_tokenizer_2 {
45 info!(clip2_tok = %clip2_tok.display());
46 }
47 for (i, te) in paths.text_encoder_files.iter().enumerate() {
48 info!(text_encoder_shard = i, path = %te.display());
49 }
50 if let Some(text_tok) = &paths.text_tokenizer {
51 info!(text_tok = %text_tok.display());
52 }
53
54 let engine = mold_inference::create_engine(
55 model_name,
56 paths,
57 &config,
58 mold_inference::LoadStrategy::Eager,
59 )?;
60 state::AppState::new(engine, config)
61 }
62 None => {
63 info!("no default model configured — models will be pulled on first request");
64 state::AppState::empty(config)
65 }
66 };
67
68 let cors = build_cors_layer()?;
69
70 let app = routes::create_router(state)
71 .layer(TraceLayer::new_for_http())
72 .layer(cors);
73
74 let addr: SocketAddr = format!("{bind}:{port}").parse()?;
75 info!(%addr, "starting mold server");
76
77 let listener = TcpListener::bind(addr).await?;
78 axum::serve(listener, app).await?;
79
80 Ok(())
81}
82
83fn build_cors_layer() -> Result<CorsLayer> {
84 let cors = match std::env::var("MOLD_CORS_ORIGIN") {
85 Ok(origin) if !origin.is_empty() => {
86 let origin = origin
87 .parse::<axum::http::HeaderValue>()
88 .map_err(|_| anyhow::anyhow!("invalid MOLD_CORS_ORIGIN value: {origin}"))?;
89 CorsLayer::new()
90 .allow_origin(origin)
91 .allow_methods([
92 axum::http::Method::GET,
93 axum::http::Method::POST,
94 axum::http::Method::DELETE,
95 ])
96 .allow_headers(tower_http::cors::Any)
97 .expose_headers([axum::http::header::HeaderName::from_static(
98 "x-mold-seed-used",
99 )])
100 }
101 _ => CorsLayer::permissive(),
102 };
103 Ok(cors)
104}
105
106#[cfg(test)]
107mod tests {
108 use super::build_cors_layer;
109 use std::sync::Mutex;
110
111 static ENV_LOCK: Mutex<()> = Mutex::new(());
112
113 #[test]
114 fn invalid_cors_origin_returns_error() {
115 let _lock = ENV_LOCK.lock().unwrap();
116 std::env::set_var("MOLD_CORS_ORIGIN", "\nnot-a-header");
117 let result = build_cors_layer();
118 std::env::remove_var("MOLD_CORS_ORIGIN");
119 assert!(result.is_err());
120 }
121
122 #[test]
123 fn valid_cors_origin_builds_layer() {
124 let _lock = ENV_LOCK.lock().unwrap();
125 std::env::set_var("MOLD_CORS_ORIGIN", "https://example.com");
126 let result = build_cors_layer();
127 std::env::remove_var("MOLD_CORS_ORIGIN");
128 assert!(result.is_ok());
129 }
130}