EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[router] Implement HTTP Dependency Injection Pattern for Router System (#8714)

Browse Source
This commit is contained in:
Simo Lin
2025-08-02 19:16:47 -07:00
committed by GitHub
parent 8ada1ab6c7
commit 828a4fe944
12 changed files with 197 additions and 186 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
sgl-router/src/routers/mod.rs
View File

@@ -7,7 +7,6 @@ use axum::{
http::{HeaderMap, StatusCode},
response::{IntoResponse, Response},
};
use reqwest::Client;
use std::fmt::Debug;
use crate::openai_api_types::{ChatCompletionRequest, CompletionRequest, GenerateRequest};
@@ -46,32 +45,27 @@ pub trait RouterTrait: Send + Sync + Debug + WorkerManagement {
fn as_any(&self) -> &dyn std::any::Any;
/// Route a health check request
async fn health(&self, client: &Client, req: Request<Body>) -> Response;
async fn health(&self, req: Request<Body>) -> Response;
/// Route a health generate request
async fn health_generate(&self, client: &Client, req: Request<Body>) -> Response;
async fn health_generate(&self, req: Request<Body>) -> Response;
/// Get server information
async fn get_server_info(&self, client: &Client, req: Request<Body>) -> Response;
async fn get_server_info(&self, req: Request<Body>) -> Response;
/// Get available models
async fn get_models(&self, client: &Client, req: Request<Body>) -> Response;
async fn get_models(&self, req: Request<Body>) -> Response;
/// Get model information
async fn get_model_info(&self, client: &Client, req: Request<Body>) -> Response;
async fn get_model_info(&self, req: Request<Body>) -> Response;
/// Route a generate request
async fn route_generate(
&self,
client: &Client,
headers: Option<&HeaderMap>,
body: &GenerateRequest,
) -> Response;
async fn route_generate(&self, headers: Option<&HeaderMap>, body: &GenerateRequest)
-> Response;
/// Route a chat completion request
async fn route_chat(
&self,
client: &Client,
headers: Option<&HeaderMap>,
body: &ChatCompletionRequest,
) -> Response;
@@ -79,16 +73,15 @@ pub trait RouterTrait: Send + Sync + Debug + WorkerManagement {
/// Route a completion request
async fn route_completion(
&self,
client: &Client,
headers: Option<&HeaderMap>,
body: &CompletionRequest,
) -> Response;
/// Flush cache on all workers
async fn flush_cache(&self, client: &Client) -> Response;
async fn flush_cache(&self) -> Response;
/// Get worker loads (for monitoring)
async fn get_worker_loads(&self, client: &Client) -> Response;
async fn get_worker_loads(&self) -> Response;
/// Get router type name
fn router_type(&self) -> &'static str;