Unknown system errno 110 on afterConnect

[jhartikainen]

On node v0.6.3, I'm encountering the following:

node.js:201
        throw e; // process.nextTick error, or 'error' event on first tick
          ^
Error: connect Unknown system errno 110
    at errnoException (net.js:614:11)
    at Object.afterConnect [as oncomplete] (net.js:605:18)

As far as I can tell, the issue happens somewhat randomly when some servers my application is connecting to are not responding for some reason or another.

This issue was supposedly fixed, see bnoordhuis' comment here:
#2121

Ubuntu 10.04 LTE, node 0.6.3

[bnoordhuis]

Thanks for the report. Question: did you install Node from source or a PPA? If it's from source, can you do a make distclean and rebuild? Sometimes the bundled libuv doesn't get recompiled after a git pull / git checkout.

[jhartikainen]

It was from source. I will do a distclean and rebuild, however it may take some time before this issue reappears.

[jhartikainen]

Just ran into it again. Seems distclean/rebuild had no effect.

[bnoordhuis]

Do you have a test case I can try? (I realize that may be difficult for 'once in a million' type of errors.)

[jhartikainen]

I was able to reproduce this using:

var net = require('net');

var c = net.createConnection(6667, 'underworld1.no.quakenet.org');
c.on('connect', function() {
    console.log('Connected');
});

It seems that server is having some connection issues, but I don't know how long it's going to last (they will probably fix the server at some point, after which I doubt this specific case will trigger it anymore).

In any case, it seems like if the connection is disconnected before it's opened fully, the error is thrown, because I didn't get the "connected" message when it was failing.

[jameshartig]

This is easy to reproduce:
https://gist.github.com/1425471

This doesn't seem to be a bug though. The system is throwing the error 110 on timeout. The same thing happens when the keep-alive probes cause the system to determine that the socket is dead. Maybe we should detect this better and possibly emit 'timeout', but I'm against doing that as this is an error in terms of node.

If you run watch on netstat with --timers, then you see that it sends a SYN_SENT but never gets a reply and the system has a timeout that kills it.

[jhartikainen]

The issue is that the error handler is not invoked for this, making it impossible to catch the error except by using a uncaughtException handler. That is at least what is happening on my end.

edit: I ran the code in your gist, and the error handler catches it, but when it happens in my app, it just skips the handler. I wish I knew why...

[jameshartig]

If you look at my gist, the error handler is being called for me with v0.6.4 and its dumping:
{ [Error: connect Unknown system errno 110]
code: 'Unknown system errno 110',
errno: 'Unknown system errno 110',
syscall: 'connect' }

[bnoordhuis]

Confirmed, I can reproduce it with @fastest963's test case. Will fix.

[bnoordhuis]

Fixed in joyent/libuv@248ca5d and a0fdd5f.

[ssssssssssss]

Can reproduce this issue in 0.6.5, so it will be included in next release?

[bnoordhuis]

Can reproduce this issue in 0.6.5, so it will be included in next release?

Yes.