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

[router][grpc] Cleanup debug logs in grpc_server and grpc_router (#11340)

Browse Source
This commit is contained in:
Chang Su
2025-10-08 13:26:19 -07:00
committed by GitHub
parent 677aa0e25f
commit a65ca73911
5 changed files with 31 additions and 102 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
sgl-router/src/routers/grpc/pipeline.rs
View File

@@ -56,8 +56,6 @@ pub struct PreparationStage;
#[async_trait]
impl PipelineStage for PreparationStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Processing request", self.name());
// Clone Arc before match to avoid borrow checker issues
// (matching borrows ctx, but prepare_* methods need mutable borrow)
// Arc clone is cheap (8 bytes) - avoids full request clone (15KB-200KB)
@@ -109,7 +107,6 @@ impl PreparationStage {
};
let token_ids = encoding.token_ids().to_vec();
debug!("Tokenized {} tokens from input", token_ids.len());
// Step 4: Build tool constraints if needed
let tool_call_constraint = body_ref.tools.as_ref().and_then(|tools| {
@@ -157,8 +154,6 @@ impl PreparationStage {
}
};
debug!("Resolved input with {} tokens", token_ids.len());
// Create stop sequence decoder for generate requests
let params = request.sampling_params.as_ref();
let stop_decoder = utils::create_stop_decoder(
@@ -259,8 +254,6 @@ impl WorkerSelectionStage {
#[async_trait]
impl PipelineStage for WorkerSelectionStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Selecting workers", self.name());
let prep = ctx
.state
.preparation
@@ -414,8 +407,6 @@ pub struct ClientAcquisitionStage;
#[async_trait]
impl PipelineStage for ClientAcquisitionStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Acquiring gRPC clients", self.name());
let workers = ctx
.state
.workers
@@ -464,8 +455,6 @@ impl RequestBuildingStage {
#[async_trait]
impl PipelineStage for RequestBuildingStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Building proto request", self.name());
let prep = ctx
.state
.preparation
@@ -578,8 +567,6 @@ pub struct DispatchMetadataStage;
#[async_trait]
impl PipelineStage for DispatchMetadataStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Preparing dispatch metadata", self.name());
let proto_request = ctx
.state
.proto_request
@@ -656,8 +643,6 @@ impl RequestExecutionStage {
#[async_trait]
impl PipelineStage for RequestExecutionStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Executing gRPC request", self.name());
let proto_request = ctx
.state
.proto_request
@@ -713,8 +698,6 @@ impl RequestExecutionStage {
.dual_mut()
.ok_or_else(|| utils::internal_error_static("Expected dual clients but got single"))?;
debug!("Sending concurrent requests to prefill and decode workers");
let prefill_request = proto_request.clone();
let decode_request = proto_request;
@@ -780,8 +763,6 @@ impl ResponseProcessingStage {
#[async_trait]
impl PipelineStage for ResponseProcessingStage {
async fn execute(&self, ctx: &mut RequestContext) -> Result<Option<Response>, Response> {
debug!("Stage {}: Processing response", self.name());
// Delegate to request-type specific processing
match &ctx.input.request_type {
RequestType::Chat(_) => return self.process_chat_response(ctx).await,
@@ -1199,15 +1180,9 @@ impl ChatCompletionPipeline {
// Execute each stage in sequence
for (idx, stage) in self.stages.iter().enumerate() {
debug!("Executing stage {}: {}", idx + 1, stage.name());
match stage.execute(&mut ctx).await {
Ok(Some(response)) => {
// Stage completed successfully with a response (e.g., streaming)
debug!(
"Stage {} ({}) completed with response",
idx + 1,
stage.name()
);
return response;
}
Ok(None) => {
@@ -1249,15 +1224,9 @@ impl ChatCompletionPipeline {
// Execute each stage in sequence
for (idx, stage) in self.stages.iter().enumerate() {
debug!("Executing stage {}: {}", idx + 1, stage.name());
match stage.execute(&mut ctx).await {
Ok(Some(response)) => {
// Stage completed successfully with a response (e.g., streaming)
debug!(
"Stage {} ({}) completed with response",
idx + 1,
stage.name()
);
return response;
}
Ok(None) => {

13
sgl-router/src/routers/grpc/utils.rs
View File

@@ -21,7 +21,7 @@ use serde_json::{json, Map, Value};
use std::collections::HashMap;
use std::sync::Arc;
use tonic::codec::Streaming;
use tracing::{debug, error, warn};
use tracing::{error, warn};
use uuid::Uuid;
/// Get gRPC client from worker, returning appropriate error response on failure
@@ -602,10 +602,6 @@ pub async fn collect_stream_responses(
Ok(gen_response) => {
match gen_response.response {
Some(Complete(complete)) => {
debug!(
"{} completed: prompt_tokens={}, completion_tokens={}, finish_reason={}",
worker_name, complete.prompt_tokens, complete.completion_tokens, complete.finish_reason
);
all_responses.push(complete);
}
Some(Error(err)) => {
@@ -615,11 +611,11 @@ pub async fn collect_stream_responses(
worker_name, err.message
)));
}
Some(Chunk(chunk)) => {
debug!("{} chunk: {} tokens", worker_name, chunk.token_ids.len());
Some(Chunk(_chunk)) => {
// Streaming chunk - no action needed
}
None => {
debug!("{}: empty response", worker_name);
// Empty response - no action needed
}
}
}
@@ -633,7 +629,6 @@ pub async fn collect_stream_responses(
}
}
debug!("{} stream closed", worker_name);
Ok(all_responses)
}