Diving Deep into Node.js-Buffer

Introduction

In Node.js Buffer class is the core library. It brings a way for Node.js to store raw data, allowing Node.js to handle binary data. The raw data is stored in the instance of the buffer class. A buffer is similar to an array of integers, but it corresponds to a piece of raw memory other than the v8 heap memory.

Way to create a new Buffer

At the start of the Node process Buffer is put into the global object, so doesn’t need explicit require.

> console.log(Buffer)
{ [Function: Buffer]
  poolSize: 8192,
  from: [Function],
  alloc: [Function],
  allocUnsafe: [Function],
  allocUnsafeSlow: [Function],
  isBuffer: [Function: isBuffer],
  compare: [Function: compare],
  isEncoding: [Function],
  concat: [Function],
  byteLength: [Function: byteLength],
  [Symbol(node.isEncoding)]: [Function] }
undefined
>

Starting from Node v6 we can create a new buffers from:

Buffer.from()
Buffer.alloc()
Buffer.allocUnsafe()

Buffer.from() and Buffer.alloc() get the original buffer, replace the data so they are safe.

Buffer.from

lib/buffer.js

/**
 * Functionally equivalent to Buffer(arg, encoding) but throws a TypeError
 * if value is a number.
 * Buffer.from(str[, encoding])
 * Buffer.from(array)
 * Buffer.from(buffer)
 * Buffer.from(arrayBuffer[, byteOffset[, length]])
 **/
Buffer.from = function from(value, encodingOrOffset, length) {
  if (typeof value === 'string')
    return fromString(value, encodingOrOffset);

  if (isAnyArrayBuffer(value))
    return fromArrayBuffer(value, encodingOrOffset, length);

  ......

  const valueOf = value.valueOf && value.valueOf();
  if (valueOf != null && valueOf !== value)
    return Buffer.from(valueOf, encodingOrOffset, length);

  var b = fromObject(value);
  if (b)
    return b;

  ....
};

ArrayBuffer: Create FasterBuffer directly using ArrayBuffer
String: Less than 4k use 8k pool, more than 4k call binding.createFromString()
Object: Less than 4k use 8k pool, more than 4k call createUnsafeBuffer()
Buffer.allocUnsafe(): Less than 4k use 8k pool, more than 4k call createUnsafeBuffer()

Important Read from offical documentaion Buffer.from(), Buffer.alloc(), and Buffer.allocUnsafe()

Buffer.alloc

lib/buffer.js

/**
 * Creates a new filled Buffer instance.
 * alloc(size[, fill[, encoding]])
 **/
Buffer.alloc = function alloc(size, fill, encoding) {
  assertSize(size);
  if (size > 0 && fill !== undefined) {
    // Since we are filling anyway, don't zero fill initially.
    // Only pay attention to encoding if it's a string. This
    // prevents accidentally sending in a number that would
    // be interpreted as a start offset.
    if (typeof encoding !== 'string')
      encoding = undefined;
    return createUnsafeBuffer(size).fill(fill, encoding);
  }
  return new FastBuffer(size);
};

Why we need buffer ?

Consider the following two case:

1. Using the large string to send as response

const http = require('http');
let hello = '';
for (let i = 0; i < 100000; i++) {
  hello += 'h';
}

http
  .createServer(function(req, res) {
    res.writeHead(200);
    res.end(hello);
  })
  .listen(8080);

Doing ab test on the above

1	ab -n 50000 -c 200 http://localhost:8080/

stringAbTest

2. Using the buffer transmission

const http = require('http');
let hello = '';
for (let i = 0; i < 100000; i++) {
  hello += 'h';
}

const buf = Buffer.from(hello);
http
  .createServer(function(req, res) {
    res.writeHead(200);
    res.end(buf);
  })
  .listen(8080);

ab test result
BufferABTest

Comparing the two result we can see that the binary data transmission rate is much faster, and less time consuming too.

Node.js internally uses Buffer at lots of places. For example

When we use the fs module to read the contents of the documents, it returns a buffer.

1
2
3

fs.readFile('filename', function(err, buf) {
  // buffer 2f 2a .....
});

In the case of net or http module, the data event parameters are a buffer.

var bufs = [];
conn.on('data', function(err, buff) {
  bufs.push(buff);
});
conn.on('end', function(err) {
  var buff = Buffer.concat(bufs);
})

Since the memory space occupied by the buffer object is not calculated in the node.js process memory, We often use Buffer to store data that requires a lot of memory as node.js process has a maximum memory limit as follow in general.

32bit - 512MB
64bit - 1.4GB

Buffer 8k Pool slab

lib/buffer.js the buffer slab size is Buffer.poolSize = 8 * 1024; 8k. When the user calls for new Buffer,

function allocate(size) {
  if (size <= 0) {
    return new FastBuffer();
  }
  if (size < (Buffer.poolSize >>> 1)) { //<4k
    if (size > (poolSize - poolOffset))
      createPool();
    var b = new FastBuffer(allocPool, poolOffset, size);
    poolOffset += size;
    alignPool();
    return b;
  } else {
    return createUnsafeBuffer(size);
  }

  function fromString(string, encoding) {
  var length;
  ...
  if (length >= (Buffer.poolSize >>> 1))
    return createFromString(string, encoding);

  if (length > (poolSize - poolOffset))
    createPool();
  var b = new FastBuffer(allocPool, poolOffset, length);
  const actual = b.write(string, encoding);
  if (actual !== length) {
    // byteLength() may overestimate. That's a rare case, though.
    b = new FastBuffer(allocPool, poolOffset, actual);
  }
  poolOffset += actual;
  alignPool();
  return b;
}
}

allocate() and fromString() check for two cases

Less than 4k - check 8k pool slab remaning capacity.
Greater than 4k - and if more than remaining capacity Directly create a new 8k pool.

function createPool() {
  poolSize = Buffer.poolSize;
  allocPool = createUnsafeArrayBuffer(poolSize);
  poolOffset = 0;
}
createPool();


function alignPool() {
  // Ensure aligned slices
  if (poolOffset & 0x7) {
    poolOffset |= 0x7;
    poolOffset++;
  }
}

8k Slab are reclaimbed by the v8 if all other references have become null. createPool() interanally calls createUnsafeArrayBuffer() to get the corresponing instance of ArrayBuffer and because ArrayBuffer is

fixed-length raw binary data buffer

it’s insecure, and there is danger of leakage of sensitive information in memory.

Side Notes: There is larger stream read-buffer in libuv unix/stream.c 64K with a slab Size of around 1MB