เทคนิคการติดตั้งเครื่องมืออัตโนมัติ

Welcome to Auto-Instrumentation Techniques! This lesson is all about adding observability to your applications without touching their code.

Think of it as a magic trick: your app starts reporting logs, metrics, and traces, but you didn't write a single line of extra code for it. This is incredibly powerful for existing or legacy systems.

So, how does this magic happen? Auto-instrumentation typically uses specialized agents or profilers that attach to your application at runtime.

• Bytecode Manipulation: For languages like Java, agents can modify the application's bytecode as it loads, injecting OpenTelemetry's tracing and metric collection logic.
• Library Wrapping: For other languages, it might involve dynamically wrapping common libraries (like HTTP clients or database drivers) to intercept their calls.

Auto-instrumentation offers significant advantages:

• Rapid Setup: Get basic observability running in minutes, not hours or days.
• No Code Changes: Crucial for legacy applications where modifying code is risky or impossible.
• Baseline Visibility: Provides immediate insights into common operations like HTTP requests, database queries, and method executions.
• Reduced Effort: Less developer time spent writing boilerplate instrumentation code.

A prominent example is the OpenTelemetry Java Agent. It's a single JAR file that you attach to your Java Virtual Machine (JVM) using a command-line argument.

Once attached, it automatically instruments many popular Java libraries and frameworks, generating traces, metrics, and logs without any code modification in your application.

Let's look at a very simple Java application. This program runs a main method and calls another method sayHello. Normally, you'd only see its print statements.

Try running it to see its normal output:

To auto-instrument the previous app, you would typically run it like this from your terminal (assuming you have the agent JAR):

The -javaagent flag tells the JVM to load the OpenTelemetry agent. The agent then automatically detects and creates spans for the main and sayHello method calls, and sends them to your configured OpenTelemetry Collector.

Auto-instrumentation is particularly effective for common I/O operations like HTTP requests. The agent automatically detects these calls made by standard libraries and creates spans showing their latency and success/failure.

Here's a simple Java program that simulates an HTTP call. If run with the OTel agent, this 'simulated' call would appear as a network span.

When using auto-instrumentation, you typically get:

• Traces: Spans for method calls, HTTP requests (incoming and outgoing), database queries, and message queue operations.
• Metrics: Basic metrics like request latency, error rates, and call counts for instrumented operations.
• Logs: Some agents can also capture logs and enrich them with trace and span IDs, helping to correlate logs with specific operations.

While powerful, auto-instrumentation has limitations:

• Lack of Business Context: It won't automatically know your application's specific business logic (e.g., "user signup" vs. just "HTTP POST").
• Custom Attributes: You can't easily add custom attributes specific to your domain without manual code changes.
• Limited Custom Metrics/Logs: It provides generic signals, but if you need very specific custom metrics or log enrichment, manual intervention is often required.

This is where manual instrumentation (our next lesson!) becomes essential to fill the gaps.

In this lesson, we explored OpenTelemetry's auto-instrumentation. You learned that it uses agents or profilers to inject observability logic into your application at runtime, without requiring source code modifications.

This technique provides rapid, baseline visibility into common operations like HTTP calls and method executions, making it ideal for quickly gaining insights into new or legacy applications. However, for deep business context and custom data, manual instrumentation is needed.