chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Super tiny move error.rs (#14944)

Browse Source
This commit is contained in:
fzyzcjy
2025-12-12 14:25:30 +08:00
committed by GitHub
parent bd91f8824a
commit a7fa31ff17
29 changed files with 125 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
sgl-model-gateway/src/observability/metrics.rs
View File

@@ -318,8 +318,7 @@ impl RouterMetrics {
}
pub fn record_request_duration(duration: Duration) {
histogram!("sgl_router_request_duration_seconds")
.record(duration.as_secs_f64());
histogram!("sgl_router_request_duration_seconds").record(duration.as_secs_f64());
}
pub fn record_request_error(route: &str, error_type: &str) {

0
sgl-model-gateway/src/routers/grpc/error.rs → sgl-model-gateway/src/routers/error.rs
View File

5
sgl-model-gateway/src/routers/grpc/common/response_collection.rs
View File

@@ -5,8 +5,9 @@
use axum::response::Response;
use crate::routers::grpc::{
context::ExecutionResult, error, proto_wrapper::ProtoGenerateComplete, utils,
use crate::routers::{
error,
grpc::{context::ExecutionResult, proto_wrapper::ProtoGenerateComplete, utils},
};
/// Collect and merge responses from execution result

2
sgl-model-gateway/src/routers/grpc/common/responses/utils.rs
View File

@@ -18,7 +18,7 @@ use crate::{
responses::{ResponseTool, ResponseToolType, ResponsesRequest, ResponsesResponse},
},
routers::{
grpc::error,
error,
openai::{conversations::persist_conversation_items, mcp::ensure_request_mcp_client},
},
};

9
sgl-model-gateway/src/routers/grpc/common/stages/client_acquisition.rs
View File

@@ -5,9 +5,12 @@ use axum::response::Response;
use tracing::error;
use super::PipelineStage;
use crate::routers::grpc::{
context::{ClientSelection, RequestContext, WorkerSelection},
error, utils,
use crate::routers::{
error,
grpc::{
context::{ClientSelection, RequestContext, WorkerSelection},
utils,
},
};
/// Client acquisition stage: Get gRPC clients from selected workers

4
sgl-model-gateway/src/routers/grpc/common/stages/dispatch_metadata.rs
View File

@@ -7,9 +7,9 @@ use axum::response::Response;
use tracing::error;
use super::PipelineStage;
use crate::routers::grpc::{
context::{DispatchMetadata, RequestContext, RequestType, WorkerSelection},
use crate::routers::{
error,
grpc::context::{DispatchMetadata, RequestContext, RequestType, WorkerSelection},
};
/// Dispatch metadata stage: Prepare metadata for dispatch

8
sgl-model-gateway/src/routers/grpc/common/stages/request_execution.rs
View File

@@ -5,10 +5,12 @@ use axum::response::Response;
use tracing::{error, info_span, Instrument};
use super::PipelineStage;
use crate::routers::grpc::{
context::{ClientSelection, ExecutionResult, RequestContext},
use crate::routers::{
error,
proto_wrapper::{ProtoGenerateRequest, ProtoStream},
grpc::{
context::{ClientSelection, ExecutionResult, RequestContext},
proto_wrapper::{ProtoGenerateRequest, ProtoStream},
},
};
type StreamResult = Result<ProtoStream, Box<dyn std::error::Error + Send + Sync>>;

4
sgl-model-gateway/src/routers/grpc/common/stages/worker_selection.rs
View File

@@ -10,9 +10,9 @@ use super::PipelineStage;
use crate::{
core::{ConnectionMode, Worker, WorkerRegistry, WorkerType},
policies::PolicyRegistry,
routers::grpc::{
context::{RequestContext, WorkerSelection},
routers::{
error,
grpc::context::{RequestContext, WorkerSelection},
},
};

8
sgl-model-gateway/src/routers/grpc/harmony/processor.rs
View File

@@ -16,10 +16,12 @@ use crate::{
ResponseStatus, ResponseUsage, ResponsesRequest, ResponsesResponse, ResponsesUsage,
},
},
routers::grpc::{
common::{response_collection, response_formatting},
context::{DispatchMetadata, ExecutionResult},
routers::{
error,
grpc::{
common::{response_collection, response_formatting},
context::{DispatchMetadata, ExecutionResult},
},
},
};

18
sgl-model-gateway/src/routers/grpc/harmony/responses.rs
View File

@@ -54,15 +54,17 @@ use crate::{
ResponsesUsage, StringOrContentParts,
},
},
routers::grpc::{
common::responses::{
build_sse_response, ensure_mcp_connection, persist_response_if_needed,
streaming::{OutputItemType, ResponseStreamEventEmitter},
},
context::SharedComponents,
routers::{
error,
harmony::{processor::ResponsesIterationResult, streaming::HarmonyStreamingProcessor},
pipeline::RequestPipeline,
grpc::{
common::responses::{
build_sse_response, ensure_mcp_connection, persist_response_if_needed,
streaming::{OutputItemType, ResponseStreamEventEmitter},
},
context::SharedComponents,
harmony::{processor::ResponsesIterationResult, streaming::HarmonyStreamingProcessor},
pipeline::RequestPipeline,
},
},
};

11
sgl-model-gateway/src/routers/grpc/harmony/stages/preparation.rs
View File

@@ -12,10 +12,13 @@ use crate::{
common::{Tool, ToolChoice, ToolChoiceValue},
responses::ResponsesRequest,
},
routers::grpc::{
common::{responses::utils::extract_tools_from_response_tools, stages::PipelineStage},
context::{PreparationOutput, RequestContext, RequestType},
error, utils,
routers::{
error,
grpc::{
common::{responses::utils::extract_tools_from_response_tools, stages::PipelineStage},
context::{PreparationOutput, RequestContext, RequestType},
utils,
},
},
};

10
sgl-model-gateway/src/routers/grpc/harmony/stages/request_building.rs
View File

@@ -5,11 +5,13 @@ use axum::response::Response;
use tracing::{debug, error};
use uuid::Uuid;
use crate::routers::grpc::{
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, RequestType, WorkerSelection},
use crate::routers::{
error,
proto_wrapper::ProtoGenerateRequest,
grpc::{
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, RequestType, WorkerSelection},
proto_wrapper::ProtoGenerateRequest,
},
};
/// Harmony Request Building stage: Convert Harmony tokens to gRPC request

8
sgl-model-gateway/src/routers/grpc/harmony/stages/response_processing.rs
View File

@@ -7,10 +7,12 @@ use axum::response::Response;
use tracing::error;
use super::super::{HarmonyResponseProcessor, HarmonyStreamingProcessor};
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext, RequestType},
use crate::routers::{
error,
grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext, RequestType},
},
};
/// Harmony Response Processing stage: Parse and format Harmony responses

1
sgl-model-gateway/src/routers/grpc/mod.rs
View File

@@ -5,7 +5,6 @@ use crate::{grpc_client::sglang_proto::MultimodalInputs, protocols::common::Stri
pub mod client;
pub mod common;
pub mod context;
pub mod error;
pub mod harmony;
pub mod pd_router;
pub mod pipeline;

3
sgl-model-gateway/src/routers/grpc/pipeline.rs
View File

@@ -11,7 +11,7 @@ use tracing::error;
use super::{
common::stages::*,
context::*,
error, harmony,
harmony,
regular::{processor, stages::*, streaming},
};
use crate::{
@@ -22,6 +22,7 @@ use crate::{
generate::GenerateRequest,
},
reasoning_parser::ParserFactory as ReasoningParserFactory,
routers::error,
tokenizer::traits::Tokenizer,
tool_parser::ParserFactory as ToolParserFactory,
};

12
sgl-model-gateway/src/routers/grpc/regular/processor.rs
View File

@@ -16,12 +16,14 @@ use crate::{
generate::{GenerateMetaInfo, GenerateRequest, GenerateResponse},
},
reasoning_parser::ParserFactory as ReasoningParserFactory,
routers::grpc::{
common::{response_collection, response_formatting},
context::{DispatchMetadata, ExecutionResult},
routers::{
error,
proto_wrapper::ProtoGenerateComplete,
utils,
grpc::{
common::{response_collection, response_formatting},
context::{DispatchMetadata, ExecutionResult},
proto_wrapper::ProtoGenerateComplete,
utils,
},
},
tokenizer::{
stop::{SequenceDecoderOutput, StopSequenceDecoder},

6
sgl-model-gateway/src/routers/grpc/regular/responses/handlers.rs
View File

@@ -64,12 +64,12 @@ use crate::{
ResponsesUsage,
},
},
routers::grpc::{
common::responses::{
routers::{
error,
grpc::common::responses::{
build_sse_response, ensure_mcp_connection, persist_response_if_needed,
streaming::ResponseStreamEventEmitter,
},
error,
},
};

4
sgl-model-gateway/src/routers/grpc/regular/responses/tool_loop.rs
View File

@@ -34,9 +34,9 @@ use crate::{
ResponsesResponse,
},
},
routers::grpc::{
common::responses::streaming::{OutputItemType, ResponseStreamEventEmitter},
routers::{
error,
grpc::common::responses::streaming::{OutputItemType, ResponseStreamEventEmitter},
},
};

11
sgl-model-gateway/src/routers/grpc/regular/stages/chat/preparation.rs
View File

@@ -8,10 +8,13 @@ use tracing::error;
use crate::{
protocols::chat::ChatCompletionRequest,
routers::grpc::{
common::stages::PipelineStage,
context::{PreparationOutput, RequestContext},
error, utils,
routers::{
error,
grpc::{
common::stages::PipelineStage,
context::{PreparationOutput, RequestContext},
utils,
},
},
};

12
sgl-model-gateway/src/routers/grpc/regular/stages/chat/request_building.rs
View File

@@ -5,12 +5,14 @@ use axum::response::Response;
use tracing::error;
use uuid::Uuid;
use crate::routers::grpc::{
client::GrpcClient,
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, WorkerSelection},
use crate::routers::{
error,
proto_wrapper::ProtoGenerateRequest,
grpc::{
client::GrpcClient,
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, WorkerSelection},
proto_wrapper::ProtoGenerateRequest,
},
};
/// Chat request building stage

10
sgl-model-gateway/src/routers/grpc/regular/stages/chat/response_processing.rs
View File

@@ -9,11 +9,13 @@ use async_trait::async_trait;
use axum::response::Response;
use tracing::error;
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext},
use crate::routers::{
error,
regular::{processor, streaming},
grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext},
regular::{processor, streaming},
},
};
/// Chat response processing stage

11
sgl-model-gateway/src/routers/grpc/regular/stages/generate/preparation.rs
View File

@@ -8,10 +8,13 @@ use tracing::error;
use crate::{
protocols::{common::InputIds, generate::GenerateRequest},
routers::grpc::{
common::stages::PipelineStage,
context::{PreparationOutput, RequestContext},
error, utils,
routers::{
error,
grpc::{
common::stages::PipelineStage,
context::{PreparationOutput, RequestContext},
utils,
},
},
tokenizer::traits::Tokenizer,
};

12
sgl-model-gateway/src/routers/grpc/regular/stages/generate/request_building.rs
View File

@@ -5,12 +5,14 @@ use axum::response::Response;
use tracing::error;
use uuid::Uuid;
use crate::routers::grpc::{
client::GrpcClient,
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, WorkerSelection},
use crate::routers::{
error,
proto_wrapper::ProtoGenerateRequest,
grpc::{
client::GrpcClient,
common::stages::{helpers, PipelineStage},
context::{ClientSelection, RequestContext, WorkerSelection},
proto_wrapper::ProtoGenerateRequest,
},
};
/// Generate request building stage

10
sgl-model-gateway/src/routers/grpc/regular/stages/generate/response_processing.rs
View File

@@ -6,11 +6,13 @@ use async_trait::async_trait;
use axum::response::Response;
use tracing::error;
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext},
use crate::routers::{
error,
regular::{processor, streaming},
grpc::{
common::stages::PipelineStage,
context::{FinalResponse, RequestContext},
regular::{processor, streaming},
},
};
/// Generate response processing stage

8
sgl-model-gateway/src/routers/grpc/regular/stages/preparation.rs
View File

@@ -8,10 +8,12 @@ use axum::response::Response;
use tracing::error;
use super::{chat::ChatPreparationStage, generate::GeneratePreparationStage};
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
use crate::routers::{
error as grpc_error,
grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
},
};
/// Preparation stage (delegates to endpoint-specific implementations)

8
sgl-model-gateway/src/routers/grpc/regular/stages/request_building.rs
View File

@@ -5,10 +5,12 @@ use axum::response::Response;
use tracing::error;
use super::{chat::ChatRequestBuildingStage, generate::GenerateRequestBuildingStage};
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
use crate::routers::{
error as grpc_error,
grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
},
};
/// Request building stage (delegates to endpoint-specific implementations)

10
sgl-model-gateway/src/routers/grpc/regular/stages/response_processing.rs
View File

@@ -7,11 +7,13 @@ use axum::response::Response;
use tracing::error;
use super::{chat::ChatResponseProcessingStage, generate::GenerateResponseProcessingStage};
use crate::routers::grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
use crate::routers::{
error,
regular::{processor, streaming},
grpc::{
common::stages::PipelineStage,
context::{RequestContext, RequestType},
regular::{processor, streaming},
},
};
/// Response processing stage (delegates to endpoint-specific implementations)

3
sgl-model-gateway/src/routers/grpc/utils.rs
View File

@@ -9,7 +9,6 @@ use uuid::Uuid;
use super::{
client::GrpcClient,
error,
proto_wrapper::{ProtoGenerateComplete, ProtoStream},
ProcessedMessages,
};
@@ -28,7 +27,7 @@ use crate::{
ParserFactory as ReasoningParserFactory, PooledParser as ReasoningPooledParser,
ReasoningParser,
},
routers::grpc::proto_wrapper::ProtoResponseVariant,
routers::{error, grpc::proto_wrapper::ProtoResponseVariant},
tokenizer::{
cache::CachedTokenizer,
chat_template::{ChatTemplateContentFormat, ChatTemplateParams},

1
sgl-model-gateway/src/routers/mod.rs
View File

@@ -21,6 +21,7 @@ use crate::protocols::{
};
pub mod conversations;
pub mod error;
pub mod factory;
pub mod grpc;
pub mod header_utils;