memwatch-ng: Leak Detection and Heap Diffing for Node.JS

memwatch-ng is here to help you detect and find memory leaks in Node.JS code. memwatch-ng is based on the original node-memwatch project but adds some additional features to give you more control over how a leak is detected.

It provides:

• A leak event, emitted when it appears your code is leaking memory.

• A stats event, emitted occasionally, giving you data describing your heap usage and trends over time.

• A HeapDiff class that lets you compare the state of your heap between two points in time, telling you what has been allocated, and what has been released.

memwatch-ng has been tested on Node.js versions: 0.10, 0.12, 1.x, 2.x, 3.x, 4.x, 5.x, 6.x and 7.x.

Installation

• npm install memwatch-ng

or

• git clone git://github.com/srowatt/memwatch-ng.git

Description

There are a growing number of tools for debugging and profiling memory usage in Node.JS applications, but there is still a need for a platform-independent native module that requires no special instrumentation. This module attempts to satisfy that need.

To get started, import memwatch-ng like so:

var memwatch = require('memwatch-ng');

Leak Detection

You can then subscribe to leak events. A leak event will be emitted when your heap usage has increased for five consecutive garbage collections:

memwatch.on('leak', function(info) { ... });

The info object will look something like:

{ start: Fri, 29 Jun 2012 14:12:13 GMT,
  end: Fri, 29 Jun 2012 14:12:33 GMT,
  growth: 67984,
  reason: 'heap growth over 5 consecutive GCs (20s) - 11.67 mb/hr' }

Leak Detection Settings

By default, leaks are defined as 5 consecutive calls to the garbage collector that result in an increase of memory. To change the number of consecutive increases of memory limit, call

memwatch.set_consecutive_growth_limit(10);

To retrieve the current growth limit setting, call:

var growthLimit = memwatch.get_consecutive_growth_limit();

Two different algorithms are used to determine leak detections. One is used for initial startup when memory is being allocated for loading modules and the other is for when Node.js is in a standard running state where memory allocations are less frequent. The set_recent_period method is used to set how many garbage compactions are triggered on startup before switching from the first leak detection algorithm to the second.

memwatch.set_recent_period(10);

To retrieve the current recent period setting, call:

var recentPeriod = memwatch.get_recent_period();

The algorithm to detect leaks uses a moving window over a period of garbage compactions with a decaying weight applied to older garbage compactions. To set the moving window period, call the set_ancient_period method to set the number of garbage compactions used in the moving window:

memwatch.set_ancient_period(120);

To retrieve the current ancient period setting, call:

var ancientPeriod = memwatch.get_ancient_period();

Heap Usage

The best way to evaluate your memory footprint is to look at heap usage right aver V8 performs garbage collection. memwatch does exactly this - it checks heap usage only after GC to give you a stable baseline of your actual memory usage.

When V8 performs a garbage collection (technically, we're talking about a full GC with heap compaction), memwatch will emit a stats event.

memwatch.on('stats', function(stats) { ... });

The stats data will look something like this:

{
  "num_full_gc": 17,
  "num_inc_gc": 8,
  "heap_compactions": 8,
  "estimated_base": 2592568,
  "current_base": 2592568,
  "min": 2499912,
  "max": 2592568,
  "usage_trend": 0
}

estimated_base and usage_trend are tracked over time. If usage trend is consistently positive, it indicates that your base heap size is continuously growing and you might have a leak.

V8 has its own idea of when it's best to perform a GC, and under a heavy load, it may defer this action for some time. To aid in speedier debugging, memwatch provides a gc() method to force V8 to do a full GC and heap compaction.

The min and max stats report the most extreme values we've seen for base heap size and are only recorded after the first 3 GC's, by default. This can be changed by calling:

memwatch.set_extremes_period(10);

Heap Diffing

So far we have seen how memwatch can aid in leak detection. For leak isolation, it provides a HeapDiff class that takes two snapshots and computes a diff between them. For example:

// Take first snapshot
var hd = new memwatch.HeapDiff();

// do some things ...

// Take the second snapshot and compute the diff
var diff = hd.end();

The contents of diff will look something like:

{
  "before": { "nodes": 11625, "size_bytes": 1869904, "size": "1.78 mb" },
  "after":  { "nodes": 21435, "size_bytes": 2119136, "size": "2.02 mb" },
  "change": { "size_bytes": 249232, "size": "243.39 kb", "freed_nodes": 197,
    "allocated_nodes": 10007,
    "details": [
      { "what": "String",
        "size_bytes": -2120,  "size": "-2.07 kb",  "+": 3,    "-": 62
      },
      { "what": "Array",
        "size_bytes": 66687,  "size": "65.13 kb",  "+": 4,    "-": 78
      },
      { "what": "LeakingClass",
        "size_bytes": 239952, "size": "234.33 kb", "+": 9998, "-": 0
      }
    ]
  }
}

The diff shows that during the sample period, the total number of allocated String and Array classes decreased, but Leaking Class grew by 9998 allocations. Hmmm.

You can use HeapDiff in your on('stats') callback; even though it takes a memory snapshot, which triggers a V8 GC, it will not trigger the stats event itself. Because that would be silly.

Future Work

Please see the Issues to share suggestions and contribute!

License

http://wtfpl.net