child process output truncated

[ben-ng]

Create these two files:

// reproduce.js
var cp = require('child_process').spawn('node', ['output.js'])
  , buff = []

cp.stdout.on('data', function (d) {
  buff.push(d)
})

cp.stdout.on('end', function () {
  console.log(buff.join('').split('\n').length)
})

// output.js
for (var i=0, ii=20000; i<ii; ++i) {
  process.stdout.write(i + ' lorem ipsum dolor sit amet\n')
}

process.exit(0)

Run

node output.js

Note that you see 19999 as the last row in your console.

Run

node reproduce.js

Note that this script is supposed to output 20001. It doesn't, because the stdout from child_process ends before all the output is actually returned. It sends a nondeterministic amount of data before end is fired.

Removing the process.exit(0) from output.js resolves the problem. This is the workaround I'm using in my app. This is a regression; the code worked just fine in node 0.12

Also interesting: the bug reproduces on OSX, but it doesn't seem to happen on Heroku, which I think uses a 64 bit linux.

[mscdex]

This is a duplicate of #2049 although there is no public fix for this yet...

[ben-ng]

Hm, are you sure its precisely the same bug? Whats peculiar to me is that when output.js is run directly, i always get all the output back, even when I make the loop run to 100000. Its only when I run it as a child process does the output get truncated.

[cjihrig]

It does seem to be the same issue, and adding process.stdout._handle.setBlocking(true) to the child process seems to fix it, as recommended in #2049.

[ben-ng]

@cjihrig great! should this issue be closed then? i worded the issue differently from the duplicates, so i wonder if leaving it open will help others find this instead of spending time debugging themselves.

[cjihrig]

Yea, let's close the duplicate. People will be able to find this now.

[mohitsud]

Adding this option fixed it:

stdio: 'inherit'