node-memwatcher

Quickly watch real-time memory stats of your node app

Introduction

This library lets you quickly log and view graphs of the memory stats of your node application. Also it can e.g. create automatic heap diffs when a possible memory leak is detected. It is in essential a wrapper for node-memwatch, but it also provides a fake node-memwatch api which uses v8.getHeapStatistics and global.gc instead. This is because node-memwatch uses gyp bindings which might not be what you want in some cases, therefore node-memwatch is a peer dependency.

Log gc stats with automatich heap dumps on possible memory leaks

Setup

Install

npm install --save node-memwatcher
// or
yarn add node-memwatcher

Install the peer dependency (recommended)

npm install --save @airbnb/node-memwatch
// or
yarn add @airbnb/node-memwatch

Import node-memwatcher and start listening

import { start } from 'node-memwatcher'
await start()

See the example for a demo application. To run the example: clone this repo and run yarn install && yarn demo

FAQ

Please read the FAQ

Options

graph boolean (false)

If true then we print time-based graphs for the heap statistics (see demo), if false then we log the stats as normal text (and verbose: true, see below)

:information_source: Make sure your project doesnt log any other information to the console because that will (partially) overwrite the graph

verbose boolean (true)

If true then we listen for the stats event from node-memwatch and display real time gc statistics

gcMetrics boolean (false)

If true then the graph is updated when a stats event is received from node-memwatch. The graph is updated every 1 second, to match that interval we add the metrics by default also every second. As gc stats might not be available, we use v8.getHeapStatistics to retrieve the stats. This gives us a nice resolution, but this method returns actual heap statistics (as in, there might be memory which node could release but just hasnt yet). When you are hunting down a memory leak, the heap usage just after the gc has run gives you a better understanding of your app's memory usage (with a lower resolution as trade off). See the faq for a possible work-around to still get a high resolution

averages boolean (false)

If true then we calculate and log long standing averages (think eg of uptime with 1min, 5min averages but with number of stats). This option is ignored when graph: true

heapAverages [number] ([10, 50, 100])

If averages: true, graph: false then we will print averages for each of these number of stats events.

useMovingAverage number (0)

If set to a number larger then 0 we will calculate the used_heap_size by taking the moving average of this last number of stats events. See the faq for more information

leakGrowthCount number (5)

We define a memory leak as when this number of stats events have consecutively been growing the heap size

autoHeapDiff boolean (false)

If true then we will automatically create a heap diff when a memory leak is detected. The first heap dump is created at leakGrowthCount - 1.

:fire: Taking heap dumps can be very expensive, you probably shouldnt enable this in production. Also please check the faq

headerEveryLines number (25)

How often we print a header with column names when graph: false, verbose: true

gcOnInterrupt boolean (true)

If true then the gc is run when a user signal is sent to the running process

:information_source: the interrupt signal is SIGBREAK on Windows and SIGUSR2 on others

heapDiffOnInterrupt boolean (false)

If true then you can create a heap diff by sending an user signal to the running process. You will always need to sent two signals for both the start heap dump as the end heap dump. The heap diff will then be calculated and logged

:information_source: the interrupt signal is SIGBREAK on Windows and SIGUSR2 on others

graphSetup [function] (undefined)

A function or array of functions which receives the graph setup as the first argument. Use this to setup your own metrics

See the readme of turtle-race and zibar for more information

graphAddMetric [function] (undefined)

A function or array of functions which are called every time new metrics are added to the graph. It receives the turtle graph as first argument and the stats as second argument. Use this to add your own metrics

Exported Methods

object = setOptions(options = {})

This helper method sets the default options and warns/corrects confliciting options

object = getMemwatch()

Returns the node-memwatch instance if the peer dependency is installed, otherwise returns a fake memwatch api

HeapDiff = startHeapDiff()

A helper function which returns the heap diff instance created by new node-memwatch.HeapDiff

object/false = endHeapDiff(boolean)

A helper function which returns false if no heap diff was started and otherwise the heap diff created by node-memwatch. When it receives true as first argument the diff is pretty printed to the console

void clearHeapDiff()

A helper function to clear / cancel a started heap diff

object = getHeapStats(stats?)

A helper function which add extra stats like usage_trend and min / max heap usage. When the first argument is omitted, it uses v8.getHeapStatistics for stats. If you pass any other other variable then a stats object from v8.getHeapStatistiscs or node-memwatch it will probably result in an error

interval = startStatsInterval(callback)

Starts a timer which calls getHeapStats every second and passes the results to the provided callback. This is used internally to update the graph if you are not listening for gc stat events

void start(options = {})

When called it starts logging or graphing memory usage. Depending on options listens either for the stats event or cakks startStatsInterval to periodically graph the memory usage

Alternatives

• Run node with --trace-gc or --trace-gc-nvp for more detailed information

Related projects

• nuxt-memwatch For watching memory stats of your Nuxt.js application