Why does a NodeJS http server close socket on timeout without response? - node.js

Given a NodeJS http server with a timeout of 10s:
const httpServer = require('http').createServer(app);
httpServer.timeout = 10 * 1000;
On timeout, Postman shows this without any response code:
Error: socket hang up
Warning: This request did not get sent completely and might not have all the required system headers
If the NodeJS server is behind an nginx reverse proxy, nginx returns a 502 response (upstream prematurely closed connection while reading response header from upstream). But here it is just NodeJS/express running on localhost. Still one would expect a proper http response.
According to this answer, this is expected behavior, the socket is simply destroyed.
In an architecture with an nginx reverse proxy, is it usual that the server just destroys the socket without sending a timeout response to the proxy?

You're setting the socket timeout when you're setting the http server timeout. The socket timeout prevents abuse from clients that might want to hang on to your connection to DOS you. It has other benefits like ensuring a certain level of service (though these are often more important when you're a client).
The reason it uses a socket timeout instead of sending a 408 status code (Request Timeout) is because the status code might have already been sent for a successful message.
If you want to implement a response timeout on your backend and handle it gracefully, you can timeout the response yourself. Note, you should likely respond with a 408 instead. 502 is for gateways like http proxies (nginx) to indicate that a downstream connection failed.
Here's a simple strawman implementation of handling that.
const httpServer = require('http').createServer((req, res) => {
setTimeout(()=>{
res.statusCode = 200;
res.statusMessage = "Ok";
res.end("Done"); // I'm never called because the timeout will be called instead;
}, 10000)
});
httpServer.on('request', (req, res) => {
setTimeout(()=>{
res.statusCode = 408;
res.statusMessage = 'Request Timeout';
res.end();
}, 1000)
});
httpServer.listen(8080);

Related

Tweak nodejs socket.io behavior when receving a connection upgrade request

I am using socket.io with nodejs to start a nodejs server with websocket capabilities.
Roughly the code looks like this.
var fs = require('fs'),
app = require('http').createServer(handler),
io = require('socket.io')(app);
app.listen(8999);
function handler (req, res) {
console.log("http");
res.writeHead(101);
res.end();
}
io.on('connection', function (socket) {
socket.emit("A new connection is established");
});
One problem I have, is that when the client requests GET http://localhost:8999/ with header
Connection: Upgrade
Upgrade: websocket
I noticed that the server will not use the web server handler, and instead it does something else.
How do I ensure that when it receives any http request, the http handler is invoked? Due to the inflexibility of my client, I actually do need it to return a HTTP 101 Switching Protocols for that particular request - even with those special headers.

What happens if you don't end the connection in Express?

What the title says. I ran into a bug where the issue was an express endpoint not ending the request connection which caused it to seemingly hang. I am still confused how the request and response flow looks like.
The Express http server object has a configurable timeout and after that timeout with no response on the http connection, the server will close the socket.
Similarly, most http clients at the other end (such as browsers) have some sort of timeout that will likely close the TCP socket if they've been waiting for a response for too long.
The http server timeout is built into the underlying http server object that Express uses and you can see how to configure its timeout here: Express.js Response Timeout. By default, the nodejs http server timeout is set to 0 which means "no timeout" is enforced.
So, if you have no server timeout and no client timeout, then the connection will just sit there indefinitely.
You can configure your own Express timeout with:
// set server timeout to 2 minutes
server.timeout = 1000 * 60 * 2;
See doc here.
Where server is the http server object created by either http.createServer() or app.listen().

Request made using wrong protocol hangs indefinitely in express js

Consider the following.
let bluebird = require('bluebird');
let fs = bluebird.promisifyAll(require('fs'));
let express = require('express');
let https = require('https');
let app = express();
let eventualKeys = bluebird.all(['key', 'crt'].map(x => fs.readFileAsync("server." + x)));
let eventualCredentials = eventualKeys.then(([key, cert]) => {
return {key: key, cert: cert};
});
let eventualHttpsServer = eventualCredentials.then(credentials => https.createServer(credentials, app));
eventualHttpsServer.then(httpsServer => httpsServer.listen(4443));
If I make a request to the server using https, everything works fine.
However if I make a request using http, it hangs indefinitely.
Obviously as it is an https server, it can't be expected to handle http requests. But is there a cleaner way of handling this? For instance, nginx replies to attempts to query the https port using http with a much less confusing "The plain HTTP request was sent to HTTPS port" message.
Also is this behavior likely to cause a resource leak on the server side?
This hangs because the express server is not responding to the client (because it is not listening to the port). As the server is not handling the request it will not cause a resource leak on the server, and the client is waiting as long as it can to give the server a chance to respond.
You could set up another server listening on the http port (80) to respond with a failure or a redirect (301) if you want to handle this kind of response. However if you are using nginx or apache it is recommended that you handle any such refusals or redirects with them, as they are less resource intensive than starting up a new node http instance just to drop a connection.

NodeJS - What does "socket hang up" actually mean?

I'm building a web scraper with Node and Cheerio, and for a certain website I'm getting the following error (it only happens on this one website, no others that I try to scrape.
It happens at a different location every time, so sometimes it's url x that throws the error, other times url x is fine and it's a different url entirely:
Error!: Error: socket hang up using [insert random URL, it's different every time]
Error: socket hang up
at createHangUpError (http.js:1445:15)
at Socket.socketOnEnd [as onend] (http.js:1541:23)
at Socket.g (events.js:175:14)
at Socket.EventEmitter.emit (events.js:117:20)
at _stream_readable.js:910:16
at process._tickCallback (node.js:415:13)
This is very tricky to debug, I don't really know where to start. To begin, what IS a socket hang up error? Is it a 404 error or similar? Or does it just mean that the server refused a connection?
I can't find an explanation of this anywhere!
EDIT: Here's a sample of code that is (sometimes) returning errors:
function scrapeNexts(url, oncomplete) {
request(url, function(err, resp, body) {
if (err) {
console.log("Uh-oh, ScrapeNexts Error!: " + err + " using " + url);
errors.nexts.push(url);
}
$ = cheerio.load(body);
// do stuff with the '$' cheerio content here
});
}
There is no direct call to close the connection, but I'm using Node Request which (as far as I can tell) uses http.get so this is not required, correct me if I'm wrong!
EDIT 2: Here's an actual, in-use bit of code that is causing errors. prodURL and other variables are mostly jquery selectors that are defined earlier. This uses the async library for Node.
function scrapeNexts(url, oncomplete) {
request(url, function (err, resp, body) {
if (err) {
console.log("Uh-oh, ScrapeNexts Error!: " + err + " using " + url);
errors.nexts.push(url);
}
async.series([
function (callback) {
$ = cheerio.load(body);
callback();
},
function (callback) {
$(prodURL).each(function () {
var theHref = $(this).attr('href');
urls.push(baseURL + theHref);
});
var next = $(next_select).first().attr('href');
oncomplete(next);
}
]);
});
}
There are two cases when socket hang up gets thrown:
When you are a client
When you, as a client, send a request to a remote server, and receive no timely response. Your socket is ended which throws this error. You should catch this error and decide how to handle it: whether retry the request, queue it for later, etc.
When you are a server/proxy
When you, as a server, perhaps a proxy server, receive a request from a client, then start acting upon it (or relay the request to the upstream server), and before you have prepared the response, the client decides to cancel/abort the request.
This stack trace shows what happens when a client cancels the request.
Trace: { [Error: socket hang up] code: 'ECONNRESET' }
at ClientRequest.proxyError (your_server_code_error_handler.js:137:15)
at ClientRequest.emit (events.js:117:20)
at Socket.socketCloseListener (http.js:1526:9)
at Socket.emit (events.js:95:17)
at TCP.close (net.js:465:12)
Line http.js:1526:9points to the same socketCloseListener mentioned by #Blender, particularly:
// This socket error fired before we started to
// receive a response. The error needs to
// fire on the request.
req.emit('error', createHangUpError());
...
function createHangUpError() {
var error = new Error('socket hang up');
error.code = 'ECONNRESET';
return error;
}
This is a typical case if the client is a user in the browser. The request to load some resource/page takes long, and users simply refresh the page. Such action causes the previous request to get aborted which on your server side throws this error.
Since this error is caused by the wish of a client, they don't expect to receive any error message. So, no need to consider this error as critical. Just ignore it. This is encouraged by the fact that on such error the res socket that your client listened to is, though still writable, destroyed.
console.log(res.socket.destroyed); //true
So, no point to send anything, except explicitly closing the response object:
res.end();
However, what you should do for sure if you are a proxy server which has already relayed the request to the upstream, is to abort your internal request to the upstream, indicating your lack of interest in the response, which in turn will tell the upstream server to, perhaps, stop an expensive operation.
Take a look at the source:
function socketCloseListener() {
var socket = this;
var parser = socket.parser;
var req = socket._httpMessage;
debug('HTTP socket close');
req.emit('close');
if (req.res && req.res.readable) {
// Socket closed before we emitted 'end' below.
req.res.emit('aborted');
var res = req.res;
res.on('end', function() {
res.emit('close');
});
res.push(null);
} else if (!req.res && !req._hadError) {
// This socket error fired before we started to
// receive a response. The error needs to
// fire on the request.
req.emit('error', createHangUpError());
req._hadError = true;
}
}
The message is emitted when the server never sends a response.
One case worth mentioning: when connecting from Node.js to Node.js using Express, I get "socket hang up" if I don't prefix the requested URL path with "/".
below is a simple example where I got the same error when I missed to add the commented code in below example. Uncommenting the code req.end() will resolve this issue.
var fs = require("fs");
var https = require("https");
var options = {
host: "en.wikipedia.org",
path: "/wiki/George_Washington",
port: 443,
method: "GET"
};
var req = https.request(options, function (res) {
console.log(res.statusCode);
});
// req.end();
I used require('http') to consume https service and it showed "socket hang up".
Then I changed require('http') to require('https') instead, and it is working.
Expanding on Blender's answer, this happens in a number of situations. The most common ones I run into are:
The server crashed.
The server refused your connection, most likely blocked by User-Agent.
socketCloseListener, as outlined in Blender's answer, is not the only place that hangup errors are created.
For example, found here:
function socketOnEnd() {
var socket = this;
var req = this._httpMessage;
var parser = this.parser;
if (!req.res) {
// If we don't have a response then we know that the socket
// ended prematurely and we need to emit an error on the request.
req.emit('error', createHangUpError());
req._hadError = true;
}
if (parser) {
parser.finish();
freeParser(parser, req);
}
socket.destroy();
}
You could try curl with the headers and such that are being sent out from Node and see if you get a response there. If you don't get a response with curl, but you do get a response in your browser, then your User-Agent header is most likely being blocked.
Another case worth mentioning (for Linux and OS X) is that if you use a library like https for performing the requests, or if you pass https://... as a URL of the locally served instance, you will be using port 443 which is a reserved private port and you might be ending up in Socket hang up or ECONNREFUSED errors.
Instead, use port 3000, f.e., and do an http request.
For request module users
Timeouts
There are two main types of timeouts: connection timeouts and read timeouts. A connect timeout occurs if the timeout is hit while your client is attempting to establish a connection to a remote machine (corresponding to the connect() call on the socket). A read timeout occurs any time the server is too slow to send back a part of the response.
Note that connection timeouts emit an ETIMEDOUT error, and read timeouts emit an ECONNRESET error.
This caused me issues, as I was doing everything listed here, but was still getting errors thrown. It turns out that calling req.abort() actually throws an error, with a code of ECONNRESET, so you actually have to catch that in your error handler.
req.on('error', function(err) {
if (err.code === "ECONNRESET") {
console.log("Timeout occurs");
return;
}
//handle normal errors
});
I had the same problem while using Nano library to connect to Couch DB. I tried to fine tune connection pooling with use of keepaliveagent library and it kept failing with socket hang up message.
var KeepAliveAgent = require('agentkeepalive');
var myagent = new KeepAliveAgent({
maxSockets: 10,
maxKeepAliveRequests: 0,
maxKeepAliveTime: 240000
});
nano = new Nano({
url : uri,
requestDefaults : {
agent : myagent
}
});
After some struggling I was able to nail the problem - as it came out it was very, very simple mistake. I was connecting to the database via HTTPS protocol, but I kept passing to my nano object a keepalive agent created as the examples for use of this library show (they rely on some defaults that use http).
One simple change to use HttpsAgent did the trick:
var KeepAliveAgent = require('agentkeepalive').HttpsAgent;
I think "socket hang up" is a fairly general error indicating that the connection has been terminated from the server end. In other words, the sockets being used to maintain the connection between the client and the server have been disconnected. (While I'm sure many of the points mentioned above are helpful to various people, I think this is the more general answer.)
In my case, I was sending a request with a payload in excess of 20K. This was rejected by the server. I verified this by removing text and retrying until the request succeeded. After determining the maximum acceptable length, I verified that adding a single character caused the error to manifest. I also confirmed that the client wasn't the issue by sending the same request from a Python app and from Postman. So anyway, I'm confident that, in my case, the length of the payload was my specific problem.
Once again, the source of the problem is anecdotal. The general problem is "Server Says No".
I had the same problem during request to some server. In my case, setting any value to User-Agent in headers in request options helped me.
const httpRequestOptions = {
hostname: 'site.address.com',
headers: {
'User-Agent': 'Chrome/59.0.3071.115'
}
};
It's not a general case and depends on server settings.
This error also can happen when working with http.request, probably your request is not finished yet.
Example:
const req = https.request(options, res => {})
And you always need to add this line: req.end()
With this function we will order to finish sending request.
As in documentation is said:
With http.request() one must always call req.end() to signify the end of the request - even if there is no data being written to the request body.
Also reason can be because of using app instance of express instead of server from const server = http.createServer(app) while creating server socket .
Wrong
const express = require('express');
const http = require('http');
const WebSocket = require('ws');
const app = express();
app.use(function (req, res) {
res.send({ msg: "hello" });
});
const wss = new WebSocket.Server({ server: app }); // will throw error while connecting from client socket
app.listen(8080, function listening() {
console.log('Listening on %d', server.address().port);
});
Correct
const express = require('express');
const http = require('http');
const WebSocket = require('ws');
const app = express();
app.use(function (req, res) {
res.send({ msg: "hello" });
});
const server = http.createServer(app);
const wss = new WebSocket.Server({ server });
server.listen(8080, function listening() {
console.log('Listening on %d', server.address().port);
});
it's been a long time but another case is when performing requests which takes a long time on the server side (more then 2 minutes which is the default for express) and the timeout parameter was not configured in the server side. In my case I was doing client->server->server request (Node.js express) and I should set the timeout parameter on each request router on the server and on the client.
So in both servers I needed to set the request timeout by using
req.setTimeout([your needed timeout])
on the router.
I do both web (node) and Android development, and open Android Studio device simulator and docker together, both of them use port 8601, it complained socket hang up error, after close Android Studio device simulator and it works well in node side. Don’t use Android Studio device simulator and docker together.
There seems to be one additional case here, which is Electron not being a fan of the "localhost" domain name. In my case I needed to change this:
const backendApiHostUrl = "http://localhost:3000";
to this:
const backendApiHostUrl = "http://127.0.0.1:3000";
After that the problem just went away.
This means that DNS resolution (local or remote) might be causing some problems too.
I got a similar error when using CouchDB on OCP cluster.
const cloudantSessionStore = sessionStore.createSessionStore(
{
type: 'couchdb',
host: 'https://' + credentials['host'],
port: credentials['port'],
dbName: 'sessions',
options: {
auth: {
username: credentials['username'],
password: credentials['password']
},
cache: false
}
}
Which should be "http", not "https", to connect with my CouchDB instance. Hope it could be helpful for anyone who is faced with similar issue.
In my case, it was because a application/json response was badly formatted (contains a stack trace). The response was never send to the server.
That was very tricky to debug because, there were no log. This thread helps me a lot to understand what happens.
In case you're using node-http-proxy, please be aware to this issue, which will result a socket hang-up error : https://github.com/nodejitsu/node-http-proxy/issues/180.
For resolution, also in this link, simply move declaring the API route (for proxying) within express routes before express.bodyParser().
Ran into this issue yesterday running my web application and node.js server through IntelliJ IDEA 2016.3.6. All I had to do was clear my cookies and cache in my Chrome browser.
If you are experiencing this error over a https connection and it's happening instantly it could be a problem setting up the SSL connection.
For me it was this issue https://github.com/nodejs/node/issues/9845 but for you it could be something else. If it is a problem with the ssl then you should be able to reproduce it with the nodejs tls/ssl package just trying to connect to the domain
I think worth noting...
I was creating tests for Google APIs. I was intercepting the request with a makeshift server, then forwarding those to the real api. I was attempting to just pass along the headers in the request, but a few headers were causing a problem with express on the other end.
Namely, I had to delete connection, accept, and content-length headers before using the request module to forward along.
let headers = Object.assign({}, req.headers);
delete headers['connection']
delete headers['accept']
delete headers['content-length']
res.end() // We don't need the incoming connection anymore
request({
method: 'post',
body: req.body,
headers: headers,
json: true,
url: `http://myapi/${req.url}`
}, (err, _res, body)=>{
if(err) return done(err);
// Test my api response here as if Google sent it.
})
I my case it's was not an error, but expected behavior for chrome browser. Chrome keeps tls connection alive (for speed i think), but node.js server stop it after 2 min and you get an error.
If you try GET request using edge browser, there will be no error at all.
If you will close chrome window - you will get error right away.
So what to do?
1)You can filter this errors, because they are not really errors.
2)Maybe there is a better solution :)
After a long debug into node js code, mongodb connection string, checking CORS etc, For me just switching to a different port number server.listen(port); made it work, into postman, try that too. No changes to proxy settings just the defaults.
I was using nano, and it took me a long time to figure out this error. My problem was I was using the wrong port. I had port 5948 instead of 5984.
var nano = require('nano')('http://localhost:5984');
var db = nano.use('address');
var app = express();
Might be your server or Socket connection crashes unexpectedly.
I had this error when running two applications on the same port by mistake.
I had a next.js app and another one in nest.js, running both on port 8080, when I looked at the .env files I realized that they had the same port, so I changed the one from nest.js to 3000 and everything worked.
I'm not saying that this is the reason for the error but it's a possibility.
Your problem might also come from an attempt to connect to an HTTP URL while your service is only published on HTTPS...
Definitely a time-consuming mistake!
Got "[GET] localhost:4200, Socket hang up" during Azure Static Web App (SWA) Emulator for Angular app.
Solution is to remove this from angular.json:
"headers": {"cross-origin-opener-policy": "same-origin-allow-popups"}

How to set the HTTP Keep-Alive timeout in a nodejs server

I'm actually doing some load testing against an ExpressJS server, and I noticed that the response send by the server includes a "Connection: Keep-Alive" header. As far as I understand it, the connection will remain opened until the server or the client sends a "Connection: Close" header.
In some implementations, the "Connection: Keep-Alive" header comes up with a "Keep-Alive" header setting the connection timeout and the maximum number of consecutive requests send via this connection.
For example : "Keep-Alive: timeout=15, max=100"
Is there a way (and is it relevant) to set these parameters on an Express server ?
If not, do you know how ExpressJS handles this ?
Edit:
After some investigations, I found out that the default timeout is set in the node standard http library:
socket.setTimeout(2 * 60 * 1000); // 2 minute timeout
In order to change this:
var http = require('http');
http.createServer(function (req, res) {
res.writeHead(200, {'Content-Type': 'text/plain'});
res.end("Hello World");
}).on('connection', function(socket) {
socket.setTimeout(10000);
}).listen(3000);
Anyway it still looks a little bit weird to me that the server doesn't send any hint to the client concerning its timeout.
Edit2:
Thanks to josh3736 for his comment.
setSocketKeepAlive is not related to HTTP keep-alive. It is a TCP-level option that allows you to detect that the other end of the connection has disappeared.
For Express 3:
var express = require('express');
var app = express();
var server = app.listen(5001);
server.on('connection', function(socket) {
console.log("A new connection was made by a client.");
socket.setTimeout(30 * 1000);
// 30 second timeout. Change this as you see fit.
});
To set keepAliveTimeout on the express server do:
var express = require('express');
var app = express();
var server = app.listen(5001);
server.keepAliveTimeout = 30000;
For Node.js 10.15.2 and newer with express, only server.keepAliveTimeout was not enough. We also need to configure server.headersTimeout longer than server.keepAliveTimeout.
server.keepAliveTimeout = 30000;
// Ensure all inactive connections are terminated by the ALB, by setting this a few seconds higher than the ALB idle timeout
server.headersTimeout = 31000;
// Ensure the headersTimeout is set higher than the keepAliveTimeout due to this nodejs regression bug: https://github.com/nodejs/node/issues/27363
Update
Since this issue Regression issue with keep alive connections has been closed. We could just set keepAliveTimeout on the latest node.js version.
One more thing, If your node.js server is deployed under AWS ELB and encounters 502 error code occasionally.
Clients -> AWS ELB -> Node Server
AWS ELB has 60 seconds of connection idle timeout by default, and per doc
We also recommend that you configure the idle timeout of your application to be larger than the idle timeout configured for the load balancer
Config the value of keepAliveTimeout to greater than 60 seconds could be one option to eliminate this issue.

Resources