Java application performance analysis should first locate bottlenecks and then choose the appropriate method. 1. Use JDK-owned tools such as jstat to view GC situation, jstack to troubleshoot thread problems, and jcmd for simple analysis; 2. Enable JFR to record runtime events, suitable for overall behavior observation; 3. Use visual utilities such as VisualVM to intuitively view call stack and hotspot methods; 4. Add monitoring buried points to the code for long-term observation of specific operations. Each method is suitable for different scenarios, and it is recommended to gradually and in-depth analysis from simple to traditional.

Performance analysis of Java applications is actually not mysterious, and it does not require complex tool chains from the beginning. The key is to find the bottleneck, such as CPU, memory, IO or threading issues. This article talks about some practical methods to help you get started quickly.

1. Use the command line tool that comes with JDK

Before you start, don't rush to install a bunch of third-party software. JDK itself provides some very practical gadgets that can help you quickly get the running status:

• jstat : used to view the usage and GC frequency of each area of ??the JVM heap memory. For example, jstat -gc <pid></pid> can see the use of young generations, old generations, and meta-spaces.
• jstack : If you suspect thread blocking or deadlocking, use jstack <pid></pid> to export thread snapshots to see which threads are in BLOCKED or WAITING state.
• jcmd : This is a multi-function command that can trigger Full GC, view JVM parameters, and even do simple CPU analysis (such as jcmd <pid> Thread.print</pid> ).

The benefits of these tools are lightweight, fast, and do not require additional dependencies, which are suitable for troubleshooting basic problems in the online environment.

2. Enable JFR for detailed analysis (Java Flight Recorder)

If your application runs in the HotSpot JVM (including OpenJDK 11), you can directly enable JFR to record detailed runtime events:

• CPU hotspot function
• Thread state changes
• GC Events
• File/Network IO Operation

Add parameters when starting:

 -XX: FlightRecorder -XX:StartFlightRecording=duration=60s,filename=myapp.jfr

In this way, a .jfr file will be automatically generated after the program runs for 60 seconds. You can open the analysis with JDK Mission Control. This method is especially suitable for situations where you want to see the overall behavior over a period of time and has a small impact on performance.

3. Use visual analysis tools (such as VisualVM, YourKit, JProfiler)

When you need to see the call stack, hotspot methods, and memory allocation paths more intuitively, graphical tools come in handy. Common ones are:

• VisualVM : Free, open source, full functions. It can be connected remotely or run locally, and supports plug-in extensions.
• YourKit/JProfiler : Commercial product, but with a more friendly interface and more powerful functions, especially suitable for complex scenarios.

This type of tool generally collects data by attaching it to running Java processes. You can see the execution time proportion of each method, object creation hotspots and other information. It is recommended to use in test environments because they may bring some performance overhead.

4. Add monitoring points to the code (suitable for long-term observation)

Sometimes, if you want to know the specific time-consuming of a module, or if you want to count the number of specific operations, you can add a log in the code or use libraries such as Micrometer and Dropwizard Metrics to collect indicators.

For example, use System.nanoTime() to record the execution time of the method and then print it out:

 long start = System.nanoTime();
// Execute business logic long duration = System.nanoTime() - start;
log.info("Execution time {} ns", duration);

Of course, this method is suitable for local debugging and is not recommended to add it everywhere. If you want to monitor for a long time, it is best to cooperate with Prometheus Grafana for visual display.

Basically these common methods are. Each method has applicable scenarios, and it is better to use it in combination. The key is not to start with the simplest ones and gradually deepen.

The above is the detailed content of How to profile a Java application for performance?. For more information, please follow other related articles on the PHP Chinese website!