SGLang Model Gateway

High-performance model routing control and data plane for large-scale LLM deployments. The gateway orchestrates fleets of workers, balances traffic across HTTP and gRPC backends, and exposes OpenAI-compatible APIs with pluggable history storage and tool integrations—while remaining deeply optimized for the SGLang serving runtime.

Overview

• Unified control plane for registering, monitoring, and orchestrating prefill, decode, and regular workers across heterogeneous model fleets.
• Data plane that routes requests across HTTP, PD (prefill/decode), gRPC, and OpenAI-compatible backends with shared reliability features.
• Industry-first gRPC pipeline with native Rust tokenization, reasoning, and tool-call execution for high-throughput OpenAI-compatible serving.
• Multi-model inference gateway mode (--enable-igw) that runs several routers at once and applies per-model policies.
• Conversation, response, and chat-history connectors that centralize state at the router, enabling compliant sharing across models/MCP loops with in-memory, no-op, or Oracle ATP storage options.
• Built-in reliability primitives: retries with exponential backoff, circuit breakers, token-bucket rate limiting, and queuing.
• First-class observability with structured logging, OpenTelemetry trace and Prometheus metrics.

Architecture at a Glance

Control Plane

• Worker Manager validates workers, discovers capabilities, and keeps the registry in sync.
• Job Queue serializes background operations (add/remove) and exposes status via /workers/{worker_id}.
• Background health checker and load monitor keep circuit breakers and policies informed.
• Optional Kubernetes service discovery keeps the registry aligned with pods.

Data Plane

• SGLang HTTP routers for regular and PD (prefill/decode) traffic with policy-aware selection.
• SGLang gRPC router and pipeline that stream tokenized requests through SRT gRPC workers with fully Rust tokenizer, reasoning parser, and tool parser implementations for maximal OpenAI API performance, supporting both single-stage and PD serving topologies.
• OpenAI router that proxies OpenAI-style requests, responses, and conversations to remote vendors (OpenAI, xAI, Gemini, and other OpenAI-compatible providers) while preserving streaming/SSE semantics.
• Router Manager coordinates multiple router implementations when IGW is enabled.
• Resilience layer delivers token-bucket rate limiting, request queuing, retry executor, and per-worker circuit breakers to keep traffic flowing through failures.
• Advanced load balancing with cache-aware request reuse, load-aware (power-of-two) selection, and per-model policy overrides.

Feature Highlights

• Multiple load balancing strategies (random, round_robin, cache_aware, power_of_two, bucket) with DP-aware scheduling.
• Multi-model HTTP serving and inference gateway routing with model-specific policies.
• Prefill/decode disaggregation, including bootstrap port handling and cache-aware merging.
• gRPC routing with fully Rust tokenizer loading, reasoning parser selection, and tool parser integration for OpenAI-compatible endpoints—supporting streaming and non-streaming modes across DeepSeek, Llama, Kimi K2, Qwen, GPT-OSS, Mistral, Step-3, GLM4, GLM4.7 and other reasoning-capable models.
• OpenAI-compatible /v1/chat/completions, /v1/responses, /v1/conversations, /v1/embeddings, /v1/rerank, /v1/classify endpoints.
• Tokenization APIs: HTTP endpoints for tokenize (/v1/tokenize) and detokenize (/v1/detokenize) with batch support; tokenizer management APIs for dynamic registration.
• Parser endpoints: Reasoning parser (/parse/reasoning) and function call parser (/parse/function_call) for separating reasoning content and extracting tool calls.
• Native MCP client integration supporting all MCP transport protocols (STDIO, HTTP, SSE, and Streamable) for tool execution loops.
• Pluggable history connectors: in-memory, disabled, Oracle ATP, or PostgreSQL (with pooling and credential support).
• Reliability controls: retry with jitter, worker-scoped circuit breakers, token bucket limiter with optional queue, and cache flush APIs.
• Service discovery for regular and PD workloads with independent selectors.
• Comprehensive observability: 40+ Prometheus metrics across HTTP, router, worker, circuit breaker, retry, discovery, MCP, and database layers; OpenTelemetry tracing with OTLP export; structured logging with request ID propagation.

Documentation

• User Guide: docs.sglang.io/advanced_features/sgl_model_gateway.html
• Additional guides, API references, and deployment patterns are continuously updated alongside SGLang releases.

Installation

Docker

Pre-built Docker images are available on Docker Hub with multi-architecture support (x86_64 and ARM64):

docker pull lmsysorg/sgl-model-gateway:latest

Prerequisites

• Rust and Cargo

  # Install rustup (Rust installer and version manager)
  curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
  
  # Reload shell environment
  source "$HOME/.cargo/env"
  
  # Verify installation
  rustc --version
  cargo --version
  

• Python with pip and virtualenv tooling available.

Rust Binary

# Build release binary
cargo build --release

Python Package

pip install maturin

# Fast development mode (debug build, no wheel, instant)
# Uses system OpenSSL (requires libssl-dev/openssl-devel)
cd bindings/python
maturin develop

# Production build (optimized, creates wheel)
# Uses vendored OpenSSL (cross-platform compatibility)
cd bindings/python
maturin build --release --out dist --features vendored-openssl
pip install --force-reinstall dist/*.whl

# Development build with system OpenSSL (faster)
# Requires: apt install libssl-dev pkg-config (Ubuntu/Debian)
#       or: yum install openssl-devel (RHEL/CentOS)
cd bindings/python
maturin build --release --out dist
pip install --force-reinstall dist/*.whl

Note: Python bindings are located in bindings/python/ with their own Cargo.toml. Use maturin develop for fast iteration during development (builds in debug mode and installs directly). Use maturin build --release --features vendored-openssl for production wheels with full optimizations (opt-level="z", lto="fat") and cross-platform compatibility. The package uses abi3 support for Python 3.8+ compatibility.

Checking Version

After installation, verify the installation and check version information:

# Simple version (Rust binary)
./target/release/sgl-model-gateway --version
# or use aliases
./target/release/smg --version
./target/release/amg --version

# Full version info with build details
./target/release/sgl-model-gateway --version-verbose

# Python CLI
amg --version
amg --version-verbose
python3 -m sglang_router --version

The --version (or -V) flag displays the version string. Use --version-verbose for comprehensive build information including Git commit, build time, compiler versions, and platform details.

Quick Start

Regular HTTP Routing

• Rust binary

  ./target/release/sgl-model-gateway \
    --worker-urls http://worker1:8000 http://worker2:8000 \
    --policy cache_aware
  

  cargo run --release -- … provides the same behavior during development.
• Python launcher

  python3 -m sglang_router.launch_router \
    --worker-urls http://worker1:8000 http://worker2:8000 \
    --policy cache_aware
  

Prefill/Decode Disaggregation (PD)

• Rust binary

  ./target/release/sgl-model-gateway \
    --pd-disaggregation \
    --prefill http://prefill1:30001 9001 \
    --prefill http://prefill2:30002 \
    --decode http://decode1:30011 \
    --decode http://decode2:30012 \
    --policy cache_aware \
    --prefill-policy cache_aware \
    --decode-policy power_of_two
  

• Python launcher

  python3 -m sglang_router.launch_router \
    --pd-disaggregation \
    --prefill http://prefill1:30001 9001 \
    --prefill http://prefill2:30002 \
    --decode http://decode1:30011 \
    --decode http://decode2:30012 \
    --policy cache_aware
  

Prefill entries accept an optional bootstrap port. PD mode merges prefill metadata with decode outputs and streams results back to the client.

Multi-Model Inference Gateway

Enable IGW mode to route multiple models through a single router while applying per-model policies:

./target/release/sgl-model-gateway \
  --enable-igw \
  --policy cache_aware \
  --max-concurrent-requests 512

# Register workers dynamically
curl -X POST http://localhost:30000/workers \
  -H "Content-Type: application/json" \
  -d '{
        "url": "http://worker-a:8000",
        "model_id": "mistral",
        "priority": 10,
        "labels": {"tier": "gold"}
      }'

# Add another worker with a different model/policy hint
curl -X POST http://localhost:30000/workers \
  -H "Content-Type: application/json" \
  -d '{
        "url": "http://worker-b:8000",
        "model_id": "llama3",
        "priority": 20,
        "labels": {"policy": "power_of_two", "tier": "silver"}
      }'

# Inspect registered workers
curl http://localhost:30000/workers

Sample response (http workers):

{
  "workers": [
    {"id":"2f3a0c3e-3a7b-4c3f-8c70-1b7d4c3a6e1f","url":"http://0.0.0.0:31378","model_id":"mistral","priority":50,"cost":1.0,"worker_type":"regular","is_healthy":true,"load":0,"connection_mode":"Http"},
    {"id":"9b0f6c2a-1c4f-4c2a-9f4a-1f2a6c0b9d3e","url":"http://0.0.0.0:34881","model_id":"llama3","priority":50,"cost":1.0,"worker_type":"regular","is_healthy":true,"load":0,"connection_mode":"Http"}
  ],
  "total": 2,
  "stats": {
    "prefill_count": 0,
    "decode_count": 0,
    "regular_count": 2
  }
}

Add more workers with the same API; include optional labels (for per-model policies) or tokenizer_path / reasoning_parser / tool_parser fields as needed. /workers/{worker_id} exposes queued job status while background jobs finalize registration.

gRPC Routing

• Rust binary

  ./target/release/sgl-model-gateway \
    --worker-urls grpc://worker-grpc-0:31001 grpc://worker-grpc-1:31002 \
    --tokenizer-path /path/to/tokenizer.json \
    --reasoning-parser deepseek-r1 \
    --tool-call-parser json
  

• Python router

  python3 -m sglang_router.launch_router \
    --worker-urls grpc://127.0.0.1:20000 \
    --model-path meta-llama/Llama-3.1-8B-Instruct \
    --host 0.0.0.0 \
    --port 8080
  

The gRPC router tokenizes inputs locally, supports tool-call parsing, and streams responses. It supports both regular HTTP-equivalent serving and PD (prefill/decode) serving when the worker registry contains PD workers. Provide --model-path or --tokenizer-path (HuggingFace ID or local directory) whenever connection mode resolves to gRPC. Use --reasoning-parser to select built-in reasoning pipelines (DeepSeek-R1, Qwen3, Step-3, GLM4, GLM4.7, etc.) and --tool-call-parser for JSON/Pythonic/XML tool contracts in streaming or non-streaming modes.

OpenAI Backend Mode

Route requests to OpenAI or OpenAI-compatible endpoints:

# Route to OpenAI API
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \

# Route to custom OpenAI-compatible endpoint (Gemini, xAI, etc.)
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls http://my-openai-compatible-service:8000 \

Notes

• OpenAI backend mode acts as a proxy to a single remote endpoint; load balancing is not applied.
• Provide exactly one --worker-urls entry per router instance.
• The Rust binary supports the same flags (./target/release/sgl-model-gateway --backend openai ...).

MCP Integration

The SGL Model Gateway provides native Model Context Protocol (MCP) client integration, enabling tool calling across STDIO, SSE, and Streamable transports. MCP servers are configured via a YAML configuration file and registered at startup through the workflow engine.

Basic Usage

# Rust binary
./target/release/sgl-model-gateway \
  --mcp-config-path /path/to/mcp-config.yaml \
  --worker-urls http://worker1:8000

# Python launcher
python3 -m sglang_router.launch_router \
  --mcp-config-path /path/to/mcp-config.yaml \
  --worker-urls http://worker1:8000

MCP Configuration File

Create an MCP configuration file to define servers, transports, and connection settings:

servers:
  - name: "filesystem"
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
    protocol: "stdio"
    required: false

  - name: "github"
    url: "https://api.github.com/mcp"
    token: "ghp_xxxxx"
    protocol: "sse"
    required: false

  - name: "custom-tools"
    url: "https://tools.example.com/mcp"
    protocol: "streamable"
    required: true

pool:
  max_connections: 100
  idle_timeout: 300  # seconds

proxy:
  http: "http://proxy.internal:8080"
  https: "https://proxy.internal:8443"
  no_proxy: "localhost,127.0.0.1,*.internal"

inventory:
  enable_refresh: true
  tool_ttl: 300  # seconds - how long tools are considered fresh
  refresh_interval: 300  # seconds - background refresh interval

Configuration Options

Server Configuration (servers array):

• name: Unique identifier for the MCP server
• command + args: For STDIO transport (local process execution)
• url: For SSE or Streamable transports (HTTP/HTTPS endpoints)
• token: Optional authentication token for HTTP-based transports
• protocol: Protocol type ("sse", "streamable", or "stdio")
• required: If true, router fails to start if server is unreachable (default: false)
• envs: Environment variables for STDIO processes (optional)
• proxy: Per-server proxy override (set to null to bypass global proxy)

Connection Pool (pool):

• max_connections: Maximum pooled connections for dynamic servers (default: 100)
• idle_timeout: Idle connection timeout in seconds before cleanup (default: 300)

Proxy Configuration (proxy):

• http/https: Proxy URLs for MCP server connections (not LLM traffic)
• no_proxy: Comma-separated hosts to exclude from proxying (supports wildcards)
• Note: Proxy settings are currently ignored for streamable transport. Use STDIO or SSE transports if proxy support is required.

Inventory Settings (inventory):

• enable_refresh: Enable automatic background refresh of tool inventory (default: true)
• tool_ttl: Tool cache TTL in seconds - how long tools are considered fresh (default: 300)
• refresh_interval: Background refresh interval in seconds - proactive inventory refresh (default: 300)

Transport Types

STDIO (Local Process):

name: "local-tools"
command: "python"
args: ["-m", "my_mcp_server"]
envs:
  API_KEY: "secret"
  DEBUG: "true"

SSE (Server-Sent Events):

name: "remote-sse"
url: "https://mcp.example.com/events"
token: "bearer-token"
protocol: "sse"

Streamable (Bidirectional Streaming):

name: "streaming-tools"
url: "https://mcp.example.com/stream"
protocol: "streamable"
required: true

Server Lifecycle

• MCP servers are registered via the workflow engine with retry logic (100 attempts, 2-hour timeout for STDIO servers)
• Discovery phase identifies tools, prompts, and resources
• Tool inventory is cached with configurable TTL and periodic refresh
• Failed optional servers log warnings; required servers halt startup
• Static servers (from config) are permanent; dynamic servers (per-request) use connection pooling

Check Prometheus metrics for MCP activity (mcp_* metrics) and workflow job status via the admin API.

Python Launcher (Router + Workers)

Launch router and SGLang worker processes together; launch_server spins up workers (HTTP or gRPC) and the router in one shot.

python3 -m sglang_router.launch_server --host 0.0.0.0

Add flags as needed for production deployments:

python3 -m sglang_router.launch_server \
  --host 0.0.0.0 \
  --port 8080 \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --tp-size 1 \
  --dp-size 8 \
  --grpc-mode

Omit --grpc-mode to start HTTP workers; the router automatically configures worker URLs and schedules them based on the provided DP size.

Mini Load Balancer (Debug)

python3 -m sglang_router.launch_router \
  --mini-lb \
  --pd-disaggregation \
  --prefill http://localhost:30001 \
  --decode http://localhost:30011

MiniLB forwards PD requests using simple random routing and is intended for local debugging only.

Running Worker Servers

Use upstream SGLang binaries to start dedicated worker processes.

• Prefill worker server (gRPC mode):

  python3 -m sglang.launch_server \
    --model meta-llama/Llama-3.1-8B-Instruct \
    --port 20000 \
    --tp-size 1 \
    --grpc-mode
  

  Remove --grpc-mode for HTTP workers. Combine with the router commands above to register the worker via CLI flags or the control-plane API.

Control Plane

Worker Lifecycle & Job Queue

• JobQueue handles asynchronous add/remove operations to avoid blocking clients.
• WorkerManager inspects worker metadata (/get_server_info, /get_model_info), tracks load, and exposes flush_cache and get_loads.
• Per-worker circuit breakers and health probes keep the registry healthy; load monitor feeds metrics to cache-aware and power-of-two policies.

Administrative & Worker APIs

Method	Path	Description
POST	/workers	Queue worker registration (prefill/decode/regular). Body matches WorkerConfigRequest. Returns 202 Accepted while the job queue processes the request.
GET	/workers	List workers with health, load, policy metadata, and queued job status.
GET	/workers/{worker_id}	Inspect a specific worker or job queue entry (UUID).
PUT	/workers/{worker_id}	Queue worker update by UUID.
DELETE	/workers/{worker_id}	Queue worker removal by UUID.
POST	/flush_cache	Trigger cache flush across HTTP workers with success/failure breakdown.
GET	/get_loads	Sample current load reported by each worker.

All administrative routes inherit router API-key protection when --api-key is supplied. Job status includes pending, processing, and failed phases with timestamps.

Service Discovery

Enable Kubernetes discovery to reconcile workers automatically:

./target/release/sgl-model-gateway \
  --service-discovery \
  --selector app=sglang-worker role=inference \
  --service-discovery-namespace sglang-system \
  --service-discovery-port 8000

PD mode accepts dedicated selectors:

--pd-disaggregation \
--prefill-selector app=sglang component=prefill \
--decode-selector app=sglang component=decode \
--service-discovery

Prefill pods can expose bootstrap ports via the sglang.ai/bootstrap-port annotation. RBAC must allow get, list, and watch on pods.

Data Plane

Router Capabilities (HTTP & gRPC)

Both router stacks:

• Share load-balancing policies (random, round-robin, cache-aware, power-of-two) with DP-aware scheduling, retries, circuit breakers, and rate limiting.
• Record metrics per request, track running load, and integrate with the router-wide policy registry.

The HTTP router exposes the full OpenAI-compatible surface area (/generate, /v1/chat/completions, /v1/completions, /v1/embeddings, /v1/responses, /v1/rerank, etc.). The gRPC router delivers blazing-fast /generate and /v1/chat/completions today, with the remaining endpoints returning 501 Not Implemented until their pipelines are finalised.

HTTP Router specifics

• Regular router handles classic single-stage workers with per-model policy overrides.
• Prefill/Decode router coordinates disaggregated prefill and decode workers, merges metadata, and manages streaming fan-in.

gRPC Router specifics

• Industry-first fully Rust implementation of an OpenAI-compatible gRPC inference gateway, including tokenizer, reasoning parser, and tool parser execution in-process for maximum throughput.
• Supports both single-stage and PD (prefill/decode) worker topologies; the router automatically selects the appropriate pipeline per model.
• Provides the same /v1/* APIs as the HTTP router while streaming tokenized requests/responses directly to SRT gRPC workers.
• Built-in reasoning parsers for DeepSeek, Qwen, Llama, Mistral, GPT-OSS, Step-3, GLM4, GLM4.7, Kimi K2, and other structured-thought models.
• Tool-call parsers for JSON, Pythonic, XML, and custom schemas with streaming and non-streaming execution loops.
• Tokenizer factory supporting HuggingFace models, local tokenizer.json files, and chat template overrides (see src/tokenizer).
• Explore the code paths in src/reasoning_parser, src/tool_parser, and src/tokenizer for the end-to-end Rust implementations that power gRPC mode.

OpenAI Router

• Proxies OpenAI-compatible chat completions and responses APIs, preserving headers and SSE streams end-to-end.
• Supports /v1/responses background jobs with cancellation, deletion, and listing input items—enabling agentic, multi-turn orchestration without persisting data at remote vendor endpoints.
• Conversation APIs (/v1/conversations and /v1/conversations/{id}/items) interact with the configured conversation storage backend for compliant chat-history management. Conversation state lives at the router tier, so the same history can drive different models or MCP loops without leaking data to upstream vendors.
• Chat history, agentic multi-turn /v1/responses, and the native MCP client (STDIO/HTTP/SSE/Streamable transports) are designed to satisfy enterprise data-privacy requirements by keeping sensitive state within the router.

Request Endpoints

Endpoint	Notes
POST /generate	SGLang generate API.
POST /v1/chat/completions	OpenAI-compatible chat. Supports streaming and tool calls.
POST /v1/completions	OpenAI-compatible text completions.
POST /v1/responses	Create background responses, returns response IDs.
GET /v1/responses/{id}	Retrieve stored responses.
Conversation endpoints (/v1/conversations, /v1/conversations/{id}, /v1/conversations/{id}/items)	Manage chat history.
POST /v1/embeddings	Forward embedding requests (HTTP and gRPC).
POST /v1/rerank, POST /rerank	Ranking APIs.
POST /v1/classify	Text classification endpoint.

Classification API

The /v1/classify endpoint provides text classification using sequence classification models (e.g., Qwen2ForSequenceClassification, BertForSequenceClassification).

Request:

curl http://localhost:30000/v1/classify \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jason9693/Qwen2.5-1.5B-apeach",
    "input": "I love this product!"
  }'

Response:

{
  "id": "classify-a1b2c3d4-5678-90ab-cdef-1234567890ab",
  "object": "list",
  "created": 1767034308,
  "model": "jason9693/Qwen2.5-1.5B-apeach",
  "data": [
    {
      "index": 0,
      "label": "positive",
      "probs": [0.12, 0.88],
      "num_classes": 2
    }
  ],
  "usage": {
    "prompt_tokens": 6,
    "completion_tokens": 0,
    "total_tokens": 6
  }
}

Fields:

• label: Predicted class label (from model's id2label config, or LABEL_N fallback)
• probs: Probability distribution over all classes (softmax of logits)
• num_classes: Number of classification classes

Notes:

• Classification reuses the embedding backend—the scheduler returns logits which are converted to probabilities via softmax
• Labels come from the model's HuggingFace config (id2label field); models without this mapping use generic labels (LABEL_0, LABEL_1, etc.)
• Both HTTP and gRPC routers support classification

Public health endpoints (/liveness, /readiness, /health, /health_generate) reflect registry state; readiness ensures PD workers are paired and IGW has at least one healthy route.

Tokenization Endpoints

The gateway provides HTTP endpoints for text tokenization, designed to mirror the SGLang Python tokenization API with support for batch operations.

Endpoint	Method	Description
POST /v1/tokenize	POST	Tokenize text to token IDs (single or batch).
POST /v1/detokenize	POST	Convert token IDs back to text (single or batch).
POST /v1/tokenizers	POST	Register a new tokenizer (async, returns job status).
GET /v1/tokenizers	GET	List all registered tokenizers.
GET /v1/tokenizers/{id}	GET	Get tokenizer info by UUID.
GET /v1/tokenizers/{id}/status	GET	Check async tokenizer loading status.
DELETE /v1/tokenizers/{id}	DELETE	Remove a tokenizer from the registry.

Tokenize Request:

{
  "model": "meta-llama/Llama-3.1-8B-Instruct",
  "prompt": "Hello, world!"
}

Batch Tokenize Request:

{
  "model": "meta-llama/Llama-3.1-8B-Instruct",
  "prompt": ["Hello", "World", "How are you?"]
}

Tokenize Response:

{
  "tokens": [15339, 11, 1917, 0],
  "count": 4,
  "char_count": 13
}

Detokenize Request:

{
  "model": "meta-llama/Llama-3.1-8B-Instruct",
  "tokens": [15339, 11, 1917, 0],
  "skip_special_tokens": true
}

Add Tokenizer (async registration):

# Register from HuggingFace
curl -X POST http://localhost:30000/v1/tokenizers \
  -H "Content-Type: application/json" \
  -d '{"name": "llama3", "source": "meta-llama/Llama-3.1-8B-Instruct"}'

# Check status
curl http://localhost:30000/v1/tokenizers/{tokenizer_id}/status

Parser Endpoints

The gateway provides admin endpoints for parsing reasoning content and function calls from LLM outputs.

Endpoint	Method	Description
POST /parse/reasoning	POST	Separate reasoning (<think>) from normal text.
POST /parse/function_call	POST	Parse function/tool calls from text.

Separate Reasoning Request:

{
  "text": "<think>Let me analyze this step by step...</think>The answer is 42.",
  "parser": "deepseek-r1"
}

Response:

{
  "normal_text": "The answer is 42.",
  "reasoning_text": "Let me analyze this step by step..."
}

Supported Reasoning Parsers:

• deepseek-r1 - DeepSeek-R1 (initial reasoning mode)
• qwen3 - Qwen-3 models
• qwen3-thinking / qwen-thinking - Qwen thinking variant
• kimi - Kimi K2 with Unicode tokens
• glm45 / glm47 - GLM-4.5/4.6/4.7 models
• step3 - Step-3 models
• minimax - MiniMax models

Function Call Parsing:

{
  "text": "{\"name\": \"get_weather\", \"arguments\": {\"city\": \"NYC\"}}",
  "parser": "json"
}

Supported tool parsers: json, python, xml.

Conversations, Responses, and Data Connectors

• --history-backend memory (default) stores responses and conversations in-process.
• --history-backend none disables persistence while keeping APIs.
• --history-backend oracle uses Oracle Autonomous Database; provide credentials via flags or environment variables.
• --history-backend postgres uses PostgreSQL Database.
• --history-backend redis uses Redis.
• Conversation item storage mirrors the history backend (Oracle or memory). The same storage powers OpenAI /responses and conversation APIs.

History Backend (OpenAI Router Mode)

Store conversation and response data for tracking, debugging, or analytics.

Note: History backends are currently supported only when running with --backend openai. gRPC mode support for the /v1/responses API is planned.

Available storage options

• Memory (default): In-memory storage, fast but ephemeral.
• None: No storage, minimal overhead.
• Oracle: Persistent storage backed by Oracle Autonomous Database.
• Postgres: Persistent storage backed by PostgreSQL Database.
• Redis: Persistent storage backed by Redis.

# Memory backend (default)
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \
  --history-backend memory

# No storage for maximum performance
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \
  --history-backend none

# Oracle ATP backend (see configuration below)
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \
  --history-backend oracle

# PostgreSQL backend
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \
  --history-backend postgres

# Redis backend
python3 -m sglang_router.launch_router \
  --backend openai \
  --worker-urls https://api.openai.com \
  --history-backend redis

Oracle configuration

Install the Oracle Instant Client and set LD_LIBRARY_PATH accordingly. Choose one connection method:

# Option 1: Full connection descriptor
export ATP_DSN="(description=(address=(protocol=tcps)(port=1522)(host=adb.region.oraclecloud.com))(connect_data=(service_name=service_name)))"

# Option 2: TNS alias (requires wallet)
export ATP_TNS_ALIAS="sglroutertestatp_high"
export ATP_WALLET_PATH="/path/to/wallet"

Provide database credentials and optional pool sizing:

export ATP_USER="admin"
export ATP_PASSWORD="YourPassword123"
export ATP_POOL_MIN=4
export ATP_POOL_MAX=32

Router flags map to these values:

• --oracle-dsn (env: ATP_DSN) or --oracle-tns-alias with --oracle-wallet-path.
• --oracle-user / --oracle-password (ATP_USER / ATP_PASSWORD).
• --oracle-wallet-path (ATP_WALLET_PATH) when using TNS alias.
• --oracle-pool-min, --oracle-pool-max, --oracle-pool-timeout-secs.

Only one of --oracle-dsn or --oracle-tns-alias should be supplied.

Redis configuration

Provide Redis connection URL and optional pool sizing:

export REDIS_URL="redis://localhost:6379"
export REDIS_POOL_MAX=16
export REDIS_RETENTION_DAYS=30

Router flags map to these values:

• --redis-url (env: REDIS_URL)
• --redis-pool-max (env: REDIS_POOL_MAX)
• --redis-retention-days (env: REDIS_RETENTION_DAYS). Set to -1 for persistent storage (default: 30 days).

Reliability & Flow Control

• Retries: Default max retries = 5 with exponential backoff (--retry-max-retries, --retry-initial-backoff-ms, --retry-max-backoff-ms, --retry-backoff-multiplier, --retry-jitter-factor). Retries trigger on 408/429/500/502/503/504.
• Circuit Breakers: Per worker thresholds (--cb-failure-threshold, --cb-success-threshold, --cb-timeout-duration-secs, --cb-window-duration-secs). Disable via --disable-circuit-breaker.
• Rate Limiting: Token bucket driven by --max-concurrent-requests. Set --rate-limit-tokens-per-second to override refill rate. Configure request queue via --queue-size and --queue-timeout-secs; queued requests observe FIFO order and respect cancellation.
• Health Checks: Runtime probes via --health-check-interval-secs, --health-check-timeout-secs, failure/success thresholds, and --health-check-endpoint. Use --disable-health-check to skip health checks entirely.
• Cache Management: /flush_cache ensures LRU eviction when redeploying PD workers.

Load Balancing Policies

• random: uniform random worker selection.
• round_robin: sequential rotation with atomic counters.
• cache_aware: maintains a prefix tree of prompts to route repeat traffic and evens load with configurable thresholds (--cache-threshold, --balance-abs-threshold, --balance-rel-threshold, --eviction-interval, --max-tree-size).
• power_of_two: chooses the lighter worker among two random candidates; integrates with LoadMonitor. Per-model overrides are available in PD mode (--prefill-policy, --decode-policy) and IGW mode via the worker registry.

Observability

Logging

Structured tracing through tracing with optional file sink (--log-dir) and --log-level (debug, info, warn, error).

Prometheus Metrics

Enable with --prometheus-host/--prometheus-port (defaults to 0.0.0.0:29000).

Metric Categories (40+ metrics):

Layer	Metric Prefix	Description
HTTP	smg_http_*	Request counts, duration, active connections, rate limiting
Router	smg_router_*	Requests by model/endpoint, latency, errors, upstream responses
Inference	smg_router_ttft/tpot/tokens_*	Time to first token, time per output token, token counts (gRPC)
Worker	smg_worker_*	Pool size, active connections, health checks, selection events
Circuit Breaker	smg_worker_cb_*	State (closed/open/half-open), transitions, outcomes
Retry	smg_worker_retries_*	Retry attempts, exhausted retries, backoff duration
Discovery	smg_discovery_*	K8s registrations, sync duration, workers discovered
MCP	smg_mcp_*	Tool calls, duration, active servers, iterations
Database	smg_db_*	Operations, duration, connections, items stored

Key Metrics:

• smg_router_ttft_seconds - Time to first token histogram (gRPC mode)
• smg_router_tpot_seconds - Time per output token histogram (gRPC mode)
• smg_router_tokens_total - Total input/output tokens by model
• smg_router_generation_duration_seconds - End-to-end generation time
• smg_worker_cb_state - Circuit breaker state gauge (0=closed, 1=open, 2=half-open)

Duration Buckets: 1ms, 5ms, 10ms, 25ms, 50ms, 100ms, 250ms, 500ms, 1s, 2.5s, 5s, 10s, 15s, 30s, 45s, 60s, 90s, 120s, 180s, 240s

OpenTelemetry Tracing

Enable distributed tracing with OTLP export:

python -m sglang_router.launch_router \
  --worker-urls http://worker1:8000 \
  --enable-trace \
  --otlp-traces-endpoint localhost:4317

Features:

• OTLP/gRPC exporter (default port 4317)
• W3C Trace Context propagation for HTTP and gRPC
• Batch span processing (500ms delay, 64 span batch size)
• Custom filtering to reduce noise (only exports relevant spans)
• Trace context injection into upstream worker requests

Configuration:

• --enable-trace - Enable OpenTelemetry tracing
• --otlp-traces-endpoint <host:port> - OTLP collector endpoint

Request ID Propagation

Configure headers for request ID extraction:

--request-id-headers x-request-id x-trace-id x-correlation-id

Responses include x-request-id header for correlation.

CORS

Set --cors-allowed-origins for browser access.

Security

Router and Worker API Keys

• Router API key (--api-key) protects client access to router endpoints; all protected routes expect Authorization: Bearer <key>.
• Workers listed in --worker-urls inherit the router API key automatically.
• When adding workers dynamically, provide explicit API keys via payload or query string; they do not inherit automatically.

# Router and initial workers share the same key
python3 -m sglang_router.launch_router \
  --api-key "shared-api-key" \
  --worker-urls http://worker1:8000 http://worker2:8000

# Adding a worker without key while router has one triggers a warning and leaves the worker unprotected
curl -X POST http://localhost:8080/add_worker?url=http://worker3:8000

# Add worker with explicit key
curl -X POST "http://localhost:8080/add_worker?url=http://worker3:8000&api_key=worker3-specific-key"

Security Configurations

1. No Authentication (default): Router and workers accept requests without keys—use only in trusted environments.
2. Router-only Authentication: Provide --api-key; clients must present the key, router accesses workers without credentials.
3. Worker-only Authentication: Router open to clients; each worker requires its own key. Supply keys when calling /workers or /add_worker.
4. Full Authentication: Set router API key and provide per-worker keys. Example:

   python3 -m sglang_router.launch_router --api-key "router-key"
   curl -H "Authorization: Bearer router-key" \
     -X POST http://localhost:8080/add_worker?url=http://worker:8000&api_key=worker-key
   

Important Notes

• Initial workers declared via CLI inherit the router key; dynamic workers must supply keys explicitly.
• Router logs a warning when a worker is registered without a key while the router expects authentication.
• When router and workers share the same key, still include the key when invoking dynamic registration APIs.

TLS (HTTPS) for Gateway Server

Enable TLS to serve the gateway over HTTPS:

python3 -m sglang_router.launch_router \
  --worker-urls http://worker1:8000 \
  --tls-cert-path /path/to/server.crt \
  --tls-key-path /path/to/server.key

Parameter	Description
--tls-cert-path	Path to server certificate (PEM format)
--tls-key-path	Path to server private key (PEM format)

Both parameters must be provided together. The gateway uses rustls with the ring crypto provider for TLS termination. If TLS is not configured, the gateway falls back to plain HTTP.

mTLS for Worker Communication

Enable mutual TLS (mTLS) for secure communication with workers in HTTP mode:

python3 -m sglang_router.launch_router \
  --worker-urls https://worker1:8443 https://worker2:8443 \
  --client-cert-path /path/to/client.crt \
  --client-key-path /path/to/client.key \
  --ca-cert-path /path/to/ca.crt

Parameter	Description
--client-cert-path	Path to client certificate for mTLS (PEM format)
--client-key-path	Path to client private key for mTLS (PEM format)
--ca-cert-path	Path to CA certificate for verifying worker TLS (PEM format)

Key Points:

• Client certificate and key must be provided together
• Multiple CA certificates can be added with multiple --ca-cert-path flags
• Uses rustls backend when TLS is configured
• Single HTTP client is created for all workers (assumes single security domain)
• TCP keepalive (30 seconds) is enabled for long-lived connections

Full TLS Example (Gateway HTTPS + Worker mTLS):

python3 -m sglang_router.launch_router \
  --worker-urls https://worker1:8443 https://worker2:8443 \
  --tls-cert-path /etc/certs/server.crt \
  --tls-key-path /etc/certs/server.key \
  --client-cert-path /etc/certs/client.crt \
  --client-key-path /etc/certs/client.key \
  --ca-cert-path /etc/certs/ca.crt \
  --api-key "secure-api-key"

Control Plane Authentication

The gateway supports role-based access control (RBAC) for control plane APIs (worker management, tokenizer registration, cache operations). Two authentication methods are available:

Authentication Methods

Method	Use Case	Configuration
API Keys	Service accounts, internal services	--control-plane-api-keys
JWT/OIDC	User authentication via Identity Provider	--jwt-issuer, --jwt-audience

Both methods can be used together. Requests are authenticated in order: API key → JWT token.

Roles

Role	Access
admin	Full access to all control plane APIs (workers, tokenizers, cache, etc.)
user	Inference/data plane APIs only (chat completions, embeddings, etc.)

API Key Authentication

Static API keys for service accounts and automation:

python3 -m sglang_router.launch_router \
  --worker-urls http://worker1:8000 \
  --control-plane-api-keys 'svc1:CI Pipeline:admin:secret-key-123' \
                           'svc2:Monitoring:user:readonly-key-456' \
  --control-plane-audit-enabled

Format: id:name:role:key

• id - Unique identifier for the key
• name - Human-readable description
• role - Either admin or user
• key - The secret key (stored as SHA-256 hash internally)

Usage:

curl -H "Authorization: Bearer secret-key-123" \
  http://localhost:30000/workers

JWT/OIDC Authentication

Authenticate users via an external Identity Provider (Azure AD, Okta, Auth0, Keycloak, etc.):

python3 -m sglang_router.launch_router \
  --worker-urls http://worker1:8000 \
  --jwt-issuer "https://login.microsoftonline.com/{tenant-id}/v2.0" \
  --jwt-audience "api://my-gateway-client-id" \
  --jwt-jwks-uri "https://login.microsoftonline.com/{tenant-id}/discovery/v2.0/keys" \
  --jwt-role-mapping 'Gateway.Admins=admin' 'Gateway.Users=user' \
  --control-plane-audit-enabled

Parameter	Description
--jwt-issuer	OIDC issuer URL. Used to validate the iss claim and discover JWKS endpoint via .well-known/openid-configuration.
--jwt-audience	Expected audience (aud claim). Typically your application's client ID or API identifier (e.g., api://client-id).
--jwt-jwks-uri	(Optional) Explicit JWKS URI. If omitted, discovered automatically from the issuer's OIDC configuration.
--jwt-role-mapping	Map IDP group/role names to gateway roles. Format: idp_role=gateway_role.

How it works:

1. User authenticates with Identity Provider (OAuth2/OIDC flow)
2. IDP issues a JWT token
3. User sends token to gateway: Authorization: Bearer <jwt-token>
4. Gateway validates the JWT:
   • Verifies signature against JWKS
   • Checks iss matches --jwt-issuer
   • Checks aud matches --jwt-audience
   • Validates expiration and other standard claims
   • Extracts role from roles claim (or groups as fallback)
   • Maps IDP role to gateway role via --jwt-role-mapping

Example Azure AD Configuration:

# Azure AD issues tokens with:
#   iss: https://login.microsoftonline.com/{tenant}/v2.0
#   aud: api://your-client-id (or the client ID itself)
#   roles: ["Gateway.Admins"] or groups: ["group-id"]

python3 -m sglang_router.launch_router \
  --jwt-issuer "https://login.microsoftonline.com/your-tenant-id/v2.0" \
  --jwt-audience "api://your-client-id" \
  --jwt-role-mapping 'Gateway.Admins=admin' 'Gateway.Users=user'

Audit Logging

Enable --control-plane-audit-enabled to log all control plane operations with:

• Timestamp
• Principal (API key ID or JWT subject)
• Role
• Action performed
• Success/failure status

Combined Authentication Example

Use both API keys and JWT for different use cases:

python3 -m sglang_router.launch_router \
  --worker-urls http://worker1:8000 \
  # API keys for service accounts
  --control-plane-api-keys 'ci:CI/CD Pipeline:admin:ci-secret' \
  # JWT for human users via Azure AD
  --jwt-issuer "https://login.microsoftonline.com/{tenant}/v2.0" \
  --jwt-audience "api://gateway" \
  --jwt-role-mapping 'Platform.Admins=admin' 'Platform.Users=user' \
  # Enable audit logging
  --control-plane-audit-enabled

Development & Testing

# Build Rust components (debug mode, fast)
cargo build

# Run Rust tests
cargo test

# Fast Python development (rebuilds and installs in debug mode)
cd bindings/python && maturin develop

# Run Python tests
cd ../..  # Back to sgl-model-gateway root
pytest e2e_test/

For production builds, use maturin build --release --out dist from the bindings/python/ directory to create optimized wheels. During development, maturin develop rebuilds and installs instantly without creating wheel files. Use python -m sglang_router.launch_server to co-launch router and SGLang workers in small clusters for local validation.

Build Caching

Local development uses incremental compilation by default (configured in .cargo/config.toml), which is optimal for the edit-compile-test cycle.

For release builds or CI, you can optionally use sccache to cache compilation artifacts:

# Install sccache
cargo install sccache

# Option 1: Set environment variable (per-session)
export RUSTC_WRAPPER=sccache
cargo build --release

# Option 2: Add to your global cargo config (~/.cargo/config.toml)
# [build]
# rustc-wrapper = "sccache"

Note: sccache and incremental compilation are mutually exclusive—sccache cannot cache incrementally compiled crates. The project defaults to incremental compilation for faster local iteration. Use sccache for clean/release builds where caching across builds matters more. CI workflows use sccache with GitHub Actions cache backend for cross-job compilation caching.

---

Release Management

Creating Gateway Releases

Create releases for the Gateway/Router component with filtered commits:

# Using make
make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0

# Save to file
make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0 OUTPUT=RELEASE_NOTES.md

# Create draft release (requires gh CLI, DEFAULT behavior)
make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0 CREATE_RELEASE=1

# Publish release immediately (requires gh CLI)
make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0 CREATE_RELEASE=1 DRAFT=0

Tag Naming: Use gateway-* or router-* prefixes to avoid triggering unrelated CI workflows.

Release Workflow

1. Create and push tag:

   git tag -a gateway-v1.0.0 <commit-hash> -m "Gateway release v1.0.0"
   git push origin gateway-v1.0.0
   

2. Generate release notes (automatically filters gateway-related commits):

   make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0
   

3. Create GitHub release:

   # Create draft (DEFAULT - review before publishing)
   make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0 CREATE_RELEASE=1
   
   # Or publish immediately (skip draft)
   make release-notes PREV=gateway-v0.2.2 CURR=gateway-v1.0.0 CREATE_RELEASE=1 DRAFT=0
   

Filtered Paths

Release notes only include commits touching:

• sgl-model-gateway/ - Router codebase
• python/sglang/srt/grpc/ - gRPC protocol
• python/sglang/srt/entrypoints/grpc_server.py - gRPC server

The script automatically extracts author attribution, PR links, and identifies new contributors.

---

SGLang Model Gateway continues to evolve alongside the core SGLang runtime. Contributions should keep CLI flags, documentation, and Python bindings in sync with the Rust implementation.