How do I implement basic “Long Polling”?

Question

I can find lots of information on how Long Polling works (For example, this, and this), but no simple examples of how to implement this in code.

All I can find is cometd, which relies on the Dojo JS framework, and a fairly complex server system..

Basically, how would I use Apache to serve the requests, and how would I write a simple script (say, in PHP) which would "long-poll" the server for new messages?

The example doesn't have to be scaleable, secure or complete, it just needs to work!

Answer 1

Below is a long polling solution I have developed for Inform8 Web. Basically you override the class and implement the loadData method. When the loadData returns a value or the operation times out it will print the result and return.

If the processing of your script may take longer than 30 seconds you may need to alter the set_time_limit() call to something longer.

Apache 2.0 license. Latest version on github https://github.com/ryanhend/Inform8/blob/master/Inform8-web/src/config/lib/Inform8/longpoll/LongPoller.php

Ryan

abstract class LongPoller {

  protected $sleepTime = 5;
  protected $timeoutTime = 30;

  function __construct() {
  }


  function setTimeout($timeout) {
    $this->timeoutTime = $timeout;
  }

  function setSleep($sleep) {
    $this->sleepTime = $sleepTime;
  }


  public function run() {
    $data = NULL;
    $timeout = 0;

    set_time_limit($this->timeoutTime + $this->sleepTime + 15);

    //Query database for data
    while($data == NULL && $timeout < $this->timeoutTime) {
      $data = $this->loadData();
      if($data == NULL){

        //No new orders, flush to notify php still alive
        flush();

        //Wait for new Messages
        sleep($this->sleepTime);
        $timeout += $this->sleepTime;
      }else{
        echo $data;
        flush();
      }
    }

  }


  protected abstract function loadData();

}

Answer 2

This is one of the scenarios that PHP is a very bad choice for. As previously mentioned, you can tie up all of your Apache workers very quickly doing something like this. PHP is built for start, execute, stop. It's not built for start, wait...execute, stop. You'll bog down your server very quickly and find that you have incredible scaling problems.

That said, you can still do this with PHP and have it not kill your server using the nginx HttpPushStreamModule: http://wiki.nginx.org/HttpPushStreamModule

You setup nginx in front of Apache (or whatever else) and it will take care of holding open the concurrent connections. You just respond with payload by sending data to an internal address which you could do with a background job or just have the messages fired off to people that were waiting whenever the new requests come in. This keeps PHP processes from sitting open during long polling.

This is not exclusive to PHP and can be done using nginx with any backend language. The concurrent open connections load is equal to Node.js so the biggest perk is that it gets you out of NEEDING Node for something like this.

You see a lot of other people mentioning other language libraries for accomplishing long polling and that's with good reason. PHP is just not well built for this type of behavior naturally.

Answer 3

Take a look at this blog post which has code for a simple chat app in Python/Django/gevent.

Answer 4

Simplest NodeJS

const http = require('http');

const server = http.createServer((req, res) => {
  SomeVeryLongAction(res);
});

server.on('clientError', (err, socket) => {
  socket.end('HTTP/1.1 400 Bad Request\r\n\r\n');
});

server.listen(8000);

// the long running task - simplified to setTimeout here
// but can be async, wait from websocket service - whatever really
function SomeVeryLongAction(response) {
  setTimeout(response.end, 10000);
}

Production wise scenario in Express for exmaple you would get response in the middleware. Do you what you need to do, can scope out all of the long polled methods to Map or something (that is visible to other flows), and invoke <Response> response.end() whenever you are ready. There is nothing special about long polled connections. Rest is just how you normally structure your application.

If you dont know what i mean by scoping out, this should give you idea

const http = require('http');
var responsesArray = [];

const server = http.createServer((req, res) => {
  // not dealing with connection
  // put it on stack (array in this case)
  responsesArray.push(res);
  // end this is where normal api flow ends
});

server.on('clientError', (err, socket) => {
  socket.end('HTTP/1.1 400 Bad Request\r\n\r\n');
});

// and eventually when we are ready to resolve
// that if is there just to ensure you actually 
// called endpoint before the timeout kicks in
function SomeVeryLongAction() {
  if ( responsesArray.length ) {
    let localResponse = responsesArray.shift();
    localResponse.end();
  }
}

// simulate some action out of endpoint flow
setTimeout(SomeVeryLongAction, 10000);
server.listen(8000);

As you see, you could really respond to all connections, one, do whatever you want. There is id for every request so you should be able to use map and access specific out of api call.

Answer 5

I used this to get to grips with Comet, I have also set up Comet using the Java Glassfish server and found lots of other examples by subscribing to cometdaily.com

Answer 6

You can try icomet(https://github.com/ideawu/icomet), a C1000K C++ comet server built with libevent. icomet also provides a JavaScript library, it is easy to use as simple as

var comet = new iComet({
    sign_url: 'http://' + app_host + '/sign?obj=' + obj,
    sub_url: 'http://' + icomet_host + '/sub',
    callback: function(msg){
        // on server push
        alert(msg.content);
    }
});

icomet supports a wide range of Browsers and OSes, including Safari(iOS, Mac), IEs(Windows), Firefox, Chrome, etc.