Zero-Code Modification in 5 minutes Enables Go Applications to Automatically Obtain End-to-End Observability

This article introduces the Loongsuite Go agent, a compile-time instrumentation tool that enables zero-code modification for end-to-end observability in Go applications.

💡Are you still worried about the observability transformation of Go applications?
💡Are you still performing manual tracking, modifying code, or importing SDKs?
💡Are you still worried about tracking points affecting performance? Today, we are bringing a solution with zero-code modification-Loongsuite Go agent, allowing your Go application to automatically obtain end-to-end observability capabilities at compile-time!🚀

😫Three Pain Points of Traditional Observability Solutions

In the microservices model, observability has become an essential capability for application O&M. However, traditional observability solutions often face three major pain points:

According to statistics, traditional tracking plans require developers to spend 20-30% of their time on monitoring code, and it is very error-prone.

1. High code intrusiveness

Traditional tracking solutions require developers to manually insert monitoring code into the business code:

// Traditional method: Manual tracking is required.  
func handleRequest(w http.ResponseWriter, r *http. Request) {  
// Manually create a span.  
ctx, span := tracer.Start(r.Context(), "handleRequest")  
defer span.End()  
  
// Business logic  
result := doSomething()  
  
// Manually record attributes  
span.SetAttributes(attribute.String("result", result))  
}

This method raises the following issues:

● Code pollution: Business code and monitoring code are mixed together.

● High maintenance costs: The monitoring code must be updated each time the business logic is modified.

● Easy to omit: Developers may forget to add tracking points to some critical paths.

2. Heavy modification workload

For an existing Go application, if you want to integrate observability, you usually need to:

● Import the OpenTelemetry SDK

● Modify each key function and add tracking code

● Configure the exporter and sampling policy.

● Test to verify that the tracking point is correct.

This process can take days or even weeks of workload.

3. Performance overhead concerns

Although runtime tracking is flexible, it incurs certain performance overhead:

● The tracking logic must be executed for each call.

● Serialization, network transmission, and other operations

● may affect application performance.

✨Solution: Automatic Compile-time Instrumentation

The Loongsuite Go agentuses compile-time instrumentation technology to automatically inject monitoring code during the compilation phase, achieving true zero-code modification..

This is an enterprise-level Go application observability solution open-sourced by Alibaba, which has been used on a large scale in the production environment.

Core Strengths

Zero-Code Modification

You only need to add the otel prefix before go build without modifying any business code:

# Traditional method  
go build -o app cmd/app  
  
# Use the Loongsuite Go agent  
otel go build -o app cmd/app

It is that simple! Your application automatically obtains end-to-end observability capabilities.

🚀Automatic Instrumentation

The tool automatically detects the frameworks and libraries you use and injects the corresponding monitoring code:

● HTTP frameworks: Gin, Echo, Fiber, FastHTTP, and Hertz

● RPC frameworks: gRPC, Dubbo-go, Kitex, and Kratos

● Databases: Database/SQL, GORM, MongoDB, and Elasticsearch

● Caches: go-redis and redigo

● Logstores: Zap, Logrus, Slog, and Zerolog

● AI frameworks: LangChain and Ollama

● More: Supports more than 50 mainstream Go frameworks and libraries.

⚡Performance-friendly

Compile-time instrumentation means:

● Low runtime overhead: Monitoring code is already optimized at compile time.

● No reflection overhead: Does not rely on runtime reflection mechanisms.

● Production-ready: Validated in large-scale production environments.

🎯Case Study: Automatic Instrumentation for the Official MCP SDK

Recently, we implemented automatic instrumentation support for the official Model Context Protocol (MCP) Go SDK. MCP is a protocol introduced by companies such as Google and Anthropic. It is used to integrate LLM applications with external data sources and tools, becoming increasingly important in AI application development.

Why choose MCP?

With the rapid development of AI applications, more and more developers are using the MCP protocol to build LLM applications. However, the observability of MCP applications has always been a challenge:

● Complex protocol: MCP supports multiple operations (such as tools/call, resources/read, and prompts/get).

● Middleware mechanism: The official SDK provides middleware, but users may not actively use it.

● Time measurement: It is necessary to accurately measure the complete time of requests and responses.

Our Solution

We adopted the strategy of automatic injection during initialization. Monitoring middleware is automatically injected when NewServer and NewClient are created, ensuring 100% coverage.

Technical Challenges

The official MCP SDK provides a comprehensive middleware mechanism, but how to automatically inject monitoring middleware without modifying user code is a technical challenge.

Solution

We adopted the strategy of automatic injection during initialization:

// Automatically inject monitoring middleware when NewServer is created.
func afterNewServer(call api.CallContext, s *mcp.Server) {
    if s == nil {
        return
    }
    // Automatically inject monitoring middleware.
    monitoringMiddleware := createServerMonitoringMiddleware()
    s.AddReceivingMiddleware(monitoringMiddleware)
}

// Automatically inject monitoring middleware when NewClient is created.
func afterNewClient(call api.CallContext, c *mcp.Client) {
    if c == nil {
        return
    }
    // Automatically inject monitoring middleware.
    monitoringMiddleware := createClientMonitoringMiddleware()
    c.AddReceivingMiddleware(monitoringMiddleware)
}

Implementation Effect

In this way, we achieved:

100% coverage: The monitoring middleware is automatically injected regardless of whether the user manually invokes AddReceivingMiddleware.
Accurate time measurement: The middleware can be executed before and after request processing, allowing accurate measurement of the complete request-response time.
Automatically record key information:

MCP method names (initialize, tools/call, and resources/read)
Tool name, resource URI, and prompt name
Request parameters and response results
Error messages and duration statistics

Examples

User code does not need to be modified at all:

// User code: Create an MCP server.
server := mcp.NewServer(&mcp.Implementation{
    Name: "my-server",
    Version: "1.0.0",
}, nil)

// Add a tool for normal use.
mcp.AddTool(server, &mcp.Tool{
    Name: "greet",
    Description: "Say hi",
}, handler)

// Run the server.
server.Run(ctx, transport)

After compilation using otel go build, all MCP requests are automatically monitored, including:

● Client invoking tools (tools/call)

● Read resources (resources/read)

● Retrieving prompts (prompts/get)

● Initializing connections (initialize)

Technical Principle: Compile-time Instrumentation

Workflow

The Loongsuite Go agent adds two key phases during compile-time:

Traditional Go compilation flow:
Source code parsing → Type checking → Semantic analysis → Code optimization → Code generation → Linking

Use the Loongsuite Go agent:
Preprocessing → Instrumentation → Source code parsing → Type checking → Semantic analysis → Code optimization → Code generation → Linking

Preprocessing: Analyze dependencies and select applicable instrumentation rules.
Instrumentation: Generate code based on rules and inject the code into the source code.

Core Technologies

● go:linkname: Linking the instrumentation function to the namespace of the target package.

● AST operation: Modify the abstract syntax tree to inject monitoring code.

● Rule-driven: Define instrumentation behavior via JSON rule files.

Instrumentation Methods

Based on framework attributes, we support multiple instrumentation methods:

Intermediary injection (such as MCP and gRPC): Inject the middleware during initialization.
Hook mechanism (such as Redis and Kafka): Utilize the Hook API of the framework.
Direct function peg (such as OpenAI SDK): Instrument directly on key functions.
Struct field injection (such as database and SQL): Inject fields to store metadata.

🚀Get Started in 5 Minutes

Step 1: Install the tool (1 minute)

# Linux/MacOS (Recommended)
sudo curl -fsSL https://cdn.jsdelivr.net/gh/alibaba/loongsuite-go-agent@main/install.sh | sudo bash

# Or download manually
wget https://github.com/alibaba/loongsuite-go-agent/releases/latest/download/otel-linux-amd64
chmod +x otel-linux-amd64
sudo mv otel-linux-amd64 /usr/local/bin/otel

Step 2: Compile the application (1 minute)

# Just prefix the go build with otel.
otel go build -o app cmd/app

Step 3: Configure the export (1 minute)

# Export to Jaeger (development environment)
export OTEL_EXPORTER_JAEGER_ENDPOINT=http://localhost:14268/api/traces

# Or export to OTLP (production environment)
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317

Step 4: Run the application (1 minute)

./app

It's that simple! Your application is now equipped with end-to-end observability capabilities.🎉

Demonstration

After use, you can see the following on Jaeger, Zipkin, or other observability platforms that support OpenTelemetry:

● ✅Complete invocation chain: From HTTP requests to database queries, everything is clear at a glance.

● ✅Detailed performance metrics: duration and error rate of each operation

● ✅Rich contextual information: request parameters, response results, and error messages

Supported export methods

The tool supports multiple export methods. You only need to configure environment variables:

Export method	Environment variable	Scenario
Jaeger	OTEL_EXPORTER_JAEGER_ENDPOINT	Developer/staging environment
OTLP	OTEL_EXPORTER_OTLP_ENDPOINT	Production environment (recommended)
Console	OTEL_EXPORTER_CONSOLE=true	Debugging/Demo
Prometheus	OTEL_EXPORTER_PROMETHEUS_PORT	Metric monitoring

For more information about the configuration options, see Official documentation.

Supported frameworks

The Loongsuite Go agent supports 50+ mainstream Go frameworks and libraries, including:

Category	Framework example
HTTP framework	Gin, Echo, Fiber, FastHTTP, Hertz, Iris, Mux
RPC framework	gRPC, Dubbo-go, Kitex, Kratos, TRPC-go
Database	database/sql, GORM, MongoDB, Elasticsearch, Cassandra
Cache	go-redis, redigo
Log	Zap, Logrus, Slog, Zerolog, go-kit/log
AI framework	LangChain, Ollama, OpenAI SDK, MCP
Message queue	Kafka (Sarama, Segmentio)
Service governance	Sentinel, Nacos

For more information, see GitHub repository.

⚡Production-grade Performance

Performance advantages brought by compile-time instrumentation:

Metric	Traditional runtime instrumentation	Loongsuite Go Agent
Runtime overhead	5-10%	< 3%
Compile time increase	0%	~20%
Code intrusive	High	Zero
Maintenance cost	High	Low

Benefits:

● ✅Low runtime overhead: Monitoring code is optimized at compile-time, and no runtime reflection is required.

● ✅Production verification: It has been verified in large-scale production environments of companies such as Alibaba.

● ✅Performance-friendly: According to benchmarks, the application performance overhead after instrumentation is usually less than 3%.

💡Note: Although the compile time increases, this only occurs during the developer/build phase and does not affect runtime performance.

Community and Support

Open-source address

● GitHub: https://github.com/alibaba/loongsuite-go-agent

● Document: https://alibaba.github.io/loongsuite-go-agent/

Join the community

● DingTalk group: 102565007776

● GitHub issues: Feedback on questions and suggestions

● Contribution code: Welcome to submit pull requests.

Comparison summary

Attribute	Traditional manual tracking	Loongsuite Go Agent
Code modification	Requires extensive modifications	✅ Zero-code modification
Development time	Days to weeks	✅ 5 minutes
Maintenance cost	High (synchronous update required)	✅ Low (automatic maintenance)
Coverage	Easy to miss	✅ 100% coverage
Performance overhead	5-10%	✅ < 3%
Framework support	Requires individual implementation	✅ 50+ frameworks automatically supported

📚 Related Resources

● 🌟GitHub: https://github.com/alibaba/loongsuite-go-agent

● 📖Document: https://alibaba.github.io/loongsuite-go-agent/

● 💼Commercial edition: https://www.alibabacloud.com/help/arms/application-monitoring/user-guide/monitoring-the-golang-applications/

● 💬DingTalk group: 102565007776

If you find it useful, welcome to star⭐ and share!

References: