@datadog/pprof

What is @datadog/pprof?

The @datadog/pprof package is a Node.js module provided by Datadog for profiling Node.js applications. It allows developers to generate and analyze CPU and heap profiles to understand the performance characteristics of their applications and identify bottlenecks or memory leaks.

What are @datadog/pprof's main functionalities?

CPU Profiling

This feature enables CPU profiling, allowing developers to understand where their application spends most of its execution time. The code sample demonstrates how to start CPU profiling with a specified duration and sampling rate.

const profiler = require('@datadog/pprof');
profiler.start({
  service: 'my-service',
  // 10 minutes profiling duration
  duration: 600,
  // Profile CPU every 1000 milliseconds
  sampleRate: 1000
});

Heap Profiling

Heap profiling helps in identifying memory leaks and understanding memory allocation patterns. The code sample shows how to start heap profiling with a specified interval between snapshots.

const profiler = require('@datadog/pprof');
profiler.heap.start({
  service: 'my-service',
  // Take a snapshot every 600 seconds
  snapshotInterval: 600
});

Other packages similar to @datadog/pprof

v8-profiler-next

v8-profiler-next is a Node.js package for profiling the V8 JavaScript engine. It provides similar functionalities for CPU and heap profiling. Compared to @datadog/pprof, it focuses more on the V8 engine specifically and might offer more detailed insights for applications heavily relying on V8's features.

node-memwatch

node-memwatch is another Node.js package designed for memory leak detection and heap diffing. While it provides valuable insights into memory usage and leaks, it does not offer CPU profiling, making it less comprehensive than @datadog/pprof for overall performance analysis.

README

pprof support for Node.js

pprof support for Node.js.

Prerequisites

1. Your application will need to be using Node.js 10.4.1 or greater, or any version of Node 12 or 14.

The profiler should not be enabled when using earlier versions of Node 10, since versions of Node.js 10 prior to 10.4.1 are impacted by this issue, which can cause garbage collection to take several minutes when heap profiling is enabled.

1. The pprof module has a native component that is used to collect profiles with v8's CPU and Heap profilers. You may need to install additional dependencies to build this module.

   • For Linux: pprof has prebuilt binaries available for Linux and Alpine Linux for Node 10, 12 and 14. No additional dependencies are required.
   • For other environments: when using @google-cloud/profiler on environments that pprof does not have prebuilt binaries for, the module node-gyp will be used to build binaries. See node-gyp's documentation for information on dependencies required to build binaries with node-gyp.

2. The pprof CLI can be used to view profiles collected with this module. Instructions for installing the pprof CLI can be found here.

Basic Set-up

Install pprof with npm or add to your package.json.

# Install through npm while saving to the local 'package.json'
npm install --save pprof

Using the Profiler

Collect a Wall Time Profile

In code:

1. Update code to collect and save a profile:

   const profile = await pprof.time.profile({
     durationMillis: 10000,    // time in milliseconds for which to 
                               // collect profile.
   });
   const buf = await pprof.encode(profile);
   fs.writeFile('wall.pb.gz', buf, (err) => {
     if (err) throw err;
   });
   

2. View the profile with command line pprof:

   pprof -http=: wall.pb.gz
   

Requiring from the command line

1. Start program from the command line:

   node --require pprof app.js
   

2. A wall time profile for the job will be saved in pprof-profile-${process.pid}.pb.gz. View the profile with command line pprof:

   pprof -http=: pprof-profile-${process.pid}.pb.gz
   

Collect a Heap Profile

1. Enable heap profiling at the start of the application:

   // The average number of bytes between samples.
   const intervalBytes = 512 * 1024;
   
   // The maximum stack depth for samples collected.
   const stackDepth = 64;
   
   heap.start(intervalBytes, stackDepth); 
   

2. Collect heap profiles:

   • Collecting and saving a profile in profile.proto format:

     const profile = await pprof.heap.profile();
     const buf = await pprof.encode(profile);
     fs.writeFile('heap.pb.gz', buf, (err) => {
       if (err) throw err;
     })
     

   • View the profile with command line pprof.

     pprof -http=: heap.pb.gz
     

   • Collecting a heap profile with V8 allocation profile format:

       const profile = await pprof.heap.v8Profile();