Any way to to turn console into fully synchronous?

[litmit]

With Node.js possible to write nice small an reliable fast synchronous console applications using Javascript.

But these applications may consume all available memory and crash with
message FATAL ERROR: CALL_AND_RETRY_LAST Allocation failed - JavaScript heap out of memory.
After investigation was founded cause of crash - console.log()
See also #3524 for details.

With Node 4.x console.log() worked in synchronous mode when stdout redirected to file, therefore this problem has workaround, but not with Node 6.x (and this not described in doc https://nodejs.org/api/console.html#console_asynchronous_vs_synchronous_consoles)

Please, add feature which allow to turn console into fully synchronous mode!
This may be command line option (--console-sync) or API call like console.mode('sync') or something else.

Without this it very hard to debug a synchronous application using console.log() or console.error() when an application generate a tons of output.

[Fishrock123]

There is more info here: https://github.com/nodejs/node/blob/master/doc/api/process.md#a-note-on-process-io

It will be up on the live docs in the next 7.x release.

The console (really process.{stdout|stderr}) is mostly synchronous but is sometimes not and there are good reasons for it. (mostly, maybe except windows pipes)

Please read the docs linked above before doing this.
You can force it into sync via:

if (process.stdout._handle) process.stdout._handle.setBlocking(true)

But, I don't recommend it.

Please let me know if you have more questions.

[litmit]

@Fishrock123

Writes may be synchronous depending on the what the stream is connected to and whether the system is Windows or Unix:

Files: synchronous on Windows and Linux

If console is mostly synchronous why console.log() eat all memory and crash as this described in #3524?
(on Node 6.x and 7.x)

Moreover @bnoordhuis wrote:

It's expected behavior: console.log is asynchronous,

I wrote a small package console-sync which replace console methods to be true synchronous, like this:

console.log = function log()
{
   fs.writeSync(this._stdout.fd, util.format.apply(null,arguments) + "\n");
} 

It some kind of hack but fixes problem with memory leaks and crashes. Moreover it works noticeably faster than the default asynchronous implementation!

console.time('console.log');
for(var i=0; i<1000000 ;++i)
{
   console.log("Value of i=%d",i);
}
console.timeEnd('console.log');

console.log: 10449.842ms

require('console-sync')
console.time('console.log sync');
for(var i=0; i<1000000 ;++i)
{
   console.log("Value of i=%d",i);
}
console.timeEnd('console.log sync');

console.log sync: 7735.458ms

[Fishrock123]

It's expected behavior: console.log is asynchronous,

It some kind of hack but fixes problem with memory leaks and crashes. Moreover it works noticeably faster than the default asynchronous implementation!

It is not asynchronous unless you are in one of the conditions as documented above. (I made this change.)

In some older versions the behaviour was different.

It still goes through libuv blocking or not so it's possible that has some overhead.

[litmit]

@Fishrock123 thanks for clarifications.

I'm thinking that i found a cause of leaks in the console.
It is PR: #9744
Without this PR the infinite console loop does not consume a memory at all and does not crash Node.js.

bug.js:

for(var i=0; ;++i)
{
   console.log("Value of i=%d",i);
}

To ensure, that console is synchronous run this script with redirection to file:

node bug.js > out.txt

console.js from commit 2ebd445 working well.
console.js from commit f18e08d cause memory leaks.

Task manager screenshot: how works console.js from commit f18e08d

[litmit]

But meanwhile...
It is theoretically possible to improve write function in console.js?
Some pseudocode:

function write(......) {
   if ( stream is synchronous )  { 
      fs.writeSync(stream.fd,.....)
   } else {
       //execute current code
   }
}

[Fishrock123]

cc @addaleax, @mcollina, @jasnell

[addaleax]

Hmm… so I assume the problem is that the callbacks passed to .write() are kept alive until the next tick?

[mcollina]

A couple of things: console.log is a debugging tool. It is not supposed to be the way to write constantly to stdout. Node.js is an asynchronous platform, and using console.log to write synchronously blocks the event loop.

I can reproduce the memory problem on all the versions of Node I checked (including v0.10) when writing to standard out on Mac OS X. This is normal and expected because we are not respecting backpressure. However, at some point we were writing synchronously when it was file, causing the process to block and slow down. IMHO we are doing the correct behavior atm, e.g. Node.js is expected to ran out of memory when using it in the way described in #11568 (comment). Probably we should explain this better.

[mcollina]

@litmit a stream cannot be synchronous.

[litmit]

A couple of things: console.log is a debugging tool.

YES!
And just because of this, the only one available embedded debugging tool should just work anytime and anywhere! And help me to debug apps but not crash them.

It is not supposed to be the way to write constantly to stdout

Simple call of console.log in a deep loop can generate a lot of output. But this really needed for find a bug.

using console.log to write synchronously blocks the event loop.

How it can block more when as described by @Fishrock123 stdout already "is mostly synchronous"?

a stream cannot be synchronous.

You completely confusing me because this does not match with https://github.com/nodejs/node/blob/master/doc/api/process.md#a-note-on-process-io

Suggested solution:
Detect if console attached to a streams like stdout and/or stderr and this streams have synchronous behaviour. Use synchronous API (fs.writeSync) for output in this case.

In this case we can turn console to be sync/async just using process.stdout._handle.setBlocking(true/false).

[Fishrock123]

a stream cannot be synchronous.

@mcollina stdio is weird and "violates" this expectation.