Pipe/stream performance regression after v2.5.0 #3477
Description
(extracting from #3429)
Not sure if this is something that's well known and/or being worked on, but just in case it's not flagged somewhere: certain streams (pipes?) still appear (as of v4.2.1) to be at only 70% of the performance they were in v2.5.0 and prior.
Here's how the performance looks for different node.js versions reading from the stdout of a spawned cat process on a 2GB file (the results are similar for reading from stdin):
Generated using bench.sh (after npm install -g nave):
#!/bin/bash
VERSIONS="4.2.1 0.12.7 0.10.40 2.5.0 3.0.0 3.1.0"
fallocate -l 2000000000 file.dat
for i in 0 1 2 3 4 5; do
for v in $VERSIONS; do
nave use $v node spawn.js
done
donespawn.js:
var spawn = require('child_process').spawn
var FILE = process.argv[2] || 'file.dat'
var SIZE_MB = 2000
spawnPipe(function(err, time) {
if (err) return console.error(err.stack || err)
console.log([process.version, time].join(','))
})
function spawnPipe(cb) {
var bytesRead = 0
var time = process.hrtime()
var cat = spawn('cat', [FILE], {stdio: ['ignore', 'pipe', 'ignore']})
cat.stdout.on('data', function(data) {
bytesRead += data.length
})
cat.on('error', cb)
cat.on('close', function(code) {
if (code !== 0) return cb(new Error('cat exited with: ' + code))
if (bytesRead != SIZE_MB * 1000 * 1000) return cb(new Error('Incorrect bytes read: ' + bytesRead))
var diff = process.hrtime(time)
cb(null, SIZE_MB / (diff[0] + diff[1] / 1e9))
})
}