In case of misunderstanding, for instance, I have called request(), say 1k times, but got 9997 responses. Does anybody have same problem?

Same thing with us on the latest release.

@derekseymour Hi, do you use proxy in your code? I'm not sure if it is a proxy issue.

I have also noticed this, I'm running 2.75.0

Server-side logs show that a 200 OK response is indeed being sent back, but nothing is logged in the client application. The first thing the client does is log the response, so I'm fairly confident this isn't an application fail on my end. These requests are typically quick (load balancer logs suggest it took 88ms to from recv req -> fully send response) so I don't think timeouts are being hit. The one noticeable point here is that I sent 2 requests at virtually the same time: so I'm wondering if request has a race condition which can cause the callback to never be invoked.

Sadly, this is a spectacular failure mode and has serious consequences for us, as we queue up requests and rely on requests either succeeding/failing/timing out in request. I can no longer trust request to do this now.

@kegsay I have this kind of problem only in the scenario of using proxy. There is no race condition for concurrent because in request all requests are sent and an Agent will handle the queue if maxSocket isn't infinite. In default http.Agent, maxSocket is infinite, so if you use the default one which means you do not specify the agent option or using proxy, all the requests will be sent.

What I wonder is the timeout feature so I read the source code about timeout and found nothing incorrect. Timeout event would have happened on socket regardless of HTTP error in theory. Do you have any idea? @mscdex @simov

@mike442144 I don't know anything about the proxy support.

@mscdex @simov @koichik Okay, I check the test cases, and in test-timeout.js, there is no timeout case about proxy or tunnel-agent. Should we add cases to cover it?

@mike442144 request won't fire timeouts if it thinks the request succeeded. I suspect that request got the response and then for some reason decided not to invoke the callback. I speculate it's a race condition due to the fact that I sent 2 and the responses according to load balancer logs were transmitted at the same millisecond.

@kegsay Maybe you are right, I'll continue checking.

any progress on this? I'm having the same issue with particular websites I'm making a get request to.

@dish615 Do you use proxy?

I've found the bug, it is in tunnel-agent. So if you are using proxy and your target website protocol is https, you would have the same problem. Below is the detail of the bug:

1. TCP connection established from client, which is your process, to the proxy server. TCP status is ESTABLISHED and the socket fireup a 'connect' event.
2. Client send a http request whose method is CONNECT to proxy server, but never got a response

According to the nodejs manual, a tcp socket connection will be timed out before established by the operating system, but a http request never.So when you call reuqest() in request.js with proxy, request.js got a socket only tunnel-agent have a http connection with proxy. See the code below:

// in tunnel-agent.js
connectReq.once('connect', onConnect)
//...
function onConnect(){
    pending.request.onSocket(socket) // pending.request is an instance of ClientRequest
}

// in _http_client.js
function onSocket(socket){
    req.emit('socket',socket); // request.js got the 'socket' event.
}

// in request.js
self.req.on('socket', function(socket) {
    var setReqTimeout = function() {
        self.req.setTimeout(timeout, function () {
        if (self.req) {
          self.abort()
          var e = new Error('ESOCKETTIMEDOUT')
          e.code = 'ESOCKETTIMEDOUT'
          e.connect = false
          self.emit('error', e)
        }
      })
    }
}

If tunnel-agent could not get a response for a 'connect' request, obviously request.js cannot get the socket event, which resulted in no timeout timer set to the underlying socket. That's the root cause, it is deep but solution is quite simple:

connectReq.setTimeout(options.timeout || 15000, function(){
    connectReq.abort();
});

That's all. I'll open a PR to the tunnel-agent.js.
@simov

@simov What are you going to do?

Same issue here! Thanks @mike442144 for your PR 👍

@lneveu You're welcome.

Hello,

Any news concerning this issue?
We have the same issue in production (socket not being closed), and I would prefer not to use a fork.
Is there anything we can do to help ?

@yodeyer pls refer to #2440

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

not resolved, let's keep it open. refer to #2440 for more details.

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.