views:

2134

answers:

9

I created a web page that makes an Ajax call every second. In Internet Explorer 7, it leaks memory badly (20 MB in about 15 minutes).

The program is very simple. It just runs a JavaScript function that makes an Ajax call. The server returns an empty string, and the JavaScript code does nothing with it. I use setTimeout to run the function every second, and I'm using Drip to watch the thing.

Here is the source:

<html>
  <head>
    <script type="text/javascript" src="http://www.google.com/jsapi"&gt;&lt;/script&gt;
    <script type="text/javascript">
      google.load('jquery', '1.4.2');
      google.load('jqueryui', '1.7.2');
    </script>
    <script type="text/javascript">
      setTimeout('testJunk()',1000);
      function testJunk() {
        $.ajax({ url: 'http://xxxxxxxxxxxxxx/test', // The url returns an empty string
                 dataType: 'html',
                 success: function(data){}
               });
        setTimeout('testJunk()',1000)
      }
    </script>
  </head>
  <body>
    Why is memory usage going up?
  </body>
</html>

How to plug this leak? I have a real application that updates a large table this way, but left unattended it will eat up gigabytes of memory.

Edit: okay, so after some good suggestions, I modified the code to:

<html>
  <head>
    <script type="text/javascript" src="http://www.google.com/jsapi"&gt;&lt;/script&gt;
    <script type="text/javascript">
      google.load('jquery', '1.4.2');
      google.load('jqueryui', '1.7.2');
    </script>
    <script type="text/javascript">
      setTimeout(testJunk,1000);
      function testJunk() {
        $.ajax({ url: 'http://xxxxxxxxxxxxxx/test', // The url returns an empty string
                 dataType: 'html',
                 success: function(data){setTimeout(testJunk,1000)}
               });
      }
    </script>
  </head>
  <body>
    Why is memory usage going up?
  </body>
</html>

It didn't seem to make any difference, though. I'm not doing anything with the DOM, and if I comment out the Ajax call, the memory leak stops. So it looks like the leak is entirely in the Ajax call. Does jQuery Ajax inherently create some sort of circular reference, and if so, how can I free it? By the way, it doesn't leak in Firefox.

Someone suggested running the test in another VM and see if the results are the same. Rather than setting up another VM, I found a laptop that was running XP Home with Internet Explorer 8. It exhibits the same problem.

I tried some older versions of jQuery and got better results, but the problem didn't go away entirely until I abandoned Ajax in jQuery and went with more traditional (and ugly) Ajax.

+6  A: 

eval() will eat up memory for sure (eval happens when passing a string to setTimeout to evaluate), don't use it in testing:

setTimeout('testJunk()',1000);

should be:

setTimeout(testJunk, 1000);

Also a better use overall would be setInterval() for a repeated operation like you want, try this:

setInterval(testJunk, 1000);
Nick Craver
Thanks for the suggestion. I tried it. If it made a difference, it was small. Memory is still going up.
Thomas Lane
@Thomas - Is that the entire page in your question, or just an abbreviated version and more is happening in the full version?
Nick Craver
I'm running the program as listed above, except that I've mangled the url for the listing.
Thomas Lane
@Thomas - I've left your sample running for the past half hour in 2 different IE7 VMs...memory usage climbs a few megs then looks like garbage collection kicks in and knocks it back down. I would test on another machine, best to get a third sample.
Nick Craver
Nick - Interesting. In my case it doesn't appear to be a lack of garbage collection. I added a call to "CollectGarbage" in my testJunk function and it didn't help. I ran it on a different VM running IE6 and have no memory leaks there. I'm going to try it on a system with IE8 and see what happens.
Thomas Lane
A: 

One problem with your code is that if your ajax request start to take a while you will start flooding the browser and server with ajax request you really should wait till the browser gets a return from the server before kicking off the next one.

function testJunk() {
    $.ajax({ url: 'http://xxxxxxxxxxxxxx/test', // The url returns an empty string
        dataType: 'html',
        complete: function(data){
            setTimeout(testJunk,1000);
        }
    });
}
testJunk();
PetersenDidIt
@petersendidit - Best to read the comments on the question :)
Nick Craver
Thanks. I made the change, but I didn't notice a difference.
Thomas Lane
A: 

I don't have IE here (running Linux) but what happen if you change dataType: 'html' to dataType: 'text'? It might be something in the parsing?

MyGGaN
Thanks. I made the change, but it didn't help.
Thomas Lane
+6  A: 

The problem appears to be with jQuery 1.4 in Internet Explorer, and to a lesser extent, versions 1.2 and 1.3.

1.4.0, 1.4.1, and 1.4.2 all exhibited the severe memory leak.

1.2.3, 1.2.6, 1.3.0, 1.3.1, and 1.3.2 all exhibited a much smaller leak (about 100 KB after 10 minutes).

I also tried a version of my program that calls Ajax in a more traditional way:

<html>
  <head>
    <script language="javascript" type="text/javascript">
      function getHTTPObject() {
        var xmlhttp;
        /*@cc_on
        @if (@_jscript_version >= 5)
          try {
            xmlhttp = new ActiveXObject("Msxml2.XMLHTTP");
          } catch (e) {
            try {
              xmlhttp = new ActiveXObject("Microsoft.XMLHTTP");
            } catch (E) {
              xmlhttp = false;
            }
          }
        @else
        xmlhttp = false;
        @end @*/
        if (!xmlhttp && typeof XMLHttpRequest != 'undefined') {
          try {
            xmlhttp = new XMLHttpRequest();
            if (xmlhttp.overrideMimeType) {
              xmlhttp.overrideMimeType("text/xml"); 
            }
          } catch (e) {
            xmlhttp = false;
          }
        }
        return xmlhttp;
      }
      var ajaxObject = getHTTPObject();
      setTimeout(testJunk,1000);
      function testJunk() {
        ajaxObject.open('POST', 'http://XXXXXXXXXXXXXXX/delme2', true);
        ajaxObject.onreadystatechange = handleAjaxResponse;
        ajaxObject.send(null);
      }
      function handleAjaxResponse() {
        if (ajaxObject.readyState==4) {
          setTimeout(testJunk,1000);
        }
      }
    </script>
  </head>
  <body>
    <div id="test">Why is memory usage going up?</div>
  </body>
</html>

This got rid of the leak entirely.

So it looks like I'll have to do my repeating Ajax calls the ugly old way until the jQuery folks iron out this problem.

Thomas Lane
Yeah, if you want to keep doing things the jQuery way, use the patch in my answer... Actually having to manage your XHR stuff directly sucks!
Ryley
+3  A: 

I encountered the same issue and had been stumped all morning ... until a few moments ago. The problem is a circular reference that is created when you set the onreadystatechange handler, that IE isn't smart enough to break. The solution, therefore, is to break it explicitly. However, obviously you can't do it from within the handler itself (though it would be convenient if you could!).

The magic statement:

delete request['onreadystatechange'];

You need to keep a record of each XMLHttpRequest object for which you set onreadystatechange. Then, at some point after readyState goes to 4, do your magic on the object. If you are already performing a repeated AJAX poll, the logical place to check for requests to clean up would be in the same polling loop. I defined a simple RequestTracker object to manage my requests.

This worked for me; I verified that it solved the leak. Here's one link in particular that led the way (I would post more, but StackOverflow isn't letting me):

Haw-Bin
+1 Great research, this is sure to be useful.
Plynx
+6  A: 

Here's a link to the bug over on jQuery, along with this as a suggested fix for jQuery 1.4.2:

--- jquery-1.4.2.js     2010-04-08 12:10:20.000000000 -0700
+++ jquery-1.4.2.js.fixed       2010-04-08 12:10:38.000000000 -0700
@@ -5219,7 +5219,7 @@

                            // Stop memory leaks
                            if ( s.async ) {
-                                       xhr = null;
+                                       xhr.onreadystatechange = null; xhr.abort = null; xhr = null;
                            }
                    }
            };
Ryley
A: 

Ryley's right, that fix on the bug ticket link works like a charm.

When I applied the change and reran the code, the memory leak is gone.

Before fix:

3000 JSON calls (Internet Explorer 7, Windows XP) goes from 18 MB to 105 MB in 15 minutes.

After fix:

3000 calls 23.4 MB to 23.2 MB in 15 minutes.

Josh Mangum
A: 

@Ryley: it fixed our issue too. Ajax in jQuery is much simpler to write. Making that line change in jquery.1.4.2.js to the new XMLHttpRequest (XHR). It worked like a charm. I report no more memory leaks as the same as @Josh.

Spencer
A: 

I've seen this, and I don't believe it's a memory leak per-se. It's just that the Ajax request returns with no data because of caching.

Add deliberate cache control, as in:

$.ajax({ url: 'http://xxxxxxxxxxxxxx/test', // The url returns an empty string
         dataType: 'html',
         cache: false,
         success: function(data){}
    });
    setTimeout('testJunk()',1000)

It's one of those things where stuff falls down the cracks, that is, does a specific thing with caching and XMLHttpRequest and jQuery doesn't use cache off by default.

nic ferrier