jQuery Ajax Memory Leak in IE8

Denny Abraham says:

It’s been a while, but I think I’ve seen this before. IE doesn’t seem to always (or ever?) correctly break circular references between dom nodes and javascript objects. In this case, the xhr seems to be treated similarly so you have to break the loop manually.

Robby says:

Just curious here… but have you tried your sample in a non-recursive way? Perhaps in a for loop rather than chaining the requests like you are.

Matt Kauffman says:

Was able to patch this without modifying jQuery itself. We actually wrap the jQuery ajax so it made it was pretty easy to handle and I’d recommend doing the same if you’re not already. Here is what it ended up looking like for us:

options.complete = function () {
var xhr = arguments[0];
// fix memory leaks in IEs
if (options.async === true || typeof async == ‘undefined’) {
if (typeof xhr != ‘undefined’) {

if (typeof xhr.onreadystatechange != ‘unknown’ ) {
xhr.onreadystatechange = null;
}
if (typeof xhr.abort != ‘unknown’ ) {
xhr.abort = null;
}
xhr = null;
}
}
if (typeof complete == ‘function’) {
complete.apply(this, arguments)
}

}

That snippet lives in our chunk of code which wrap $.ajax and just wraps the complete callback in code to null out the leaky stuff in the xhr object and then calls the original one if it exists.

Haso Keric says:

jQuery 1.4.3 has been released.

Patrick Bacon says:

Philip: No, the fix is not in jQuery 1.4.3.

When I upgraded I went with a variation of the code in Matt Kaufman’s comment – basically wrapping the jQuery $.ajax call with my own function that sets the xhr properties to null on the “complete” handler.

Patrick Bacon says:

Philip: Here is an example wrapper function:

var myAjax = function(settings) {
  $.ajax($.extend({}, settings, {
    complete: function (xhr, status) {
      if (settings.complete) {
        settings.complete(xhr, status);
      }

      // Avoid IE memory leak
      xhr.onreadystatechange = null;
      xhr.abort = null;
    }
  }));
};

So instead of calling $.ajax() you would call myAjax(), passing it any of the normal settings. This will work if you are always calling $.ajax in your code. If you call any of the convenience functions (e.g. $.get()) you will want to replace $.ajax with your wrapper.

I also just noticed a .ajaxComplete event handler. I have not tried this out, but it might be possible to do something like:

$('body').ajaxComplete(function(e, xhr, settings) {
  // Avoid IE memory leak
  xhr.onreadystatechange = null;
  xhr.abort = null;
});

Jeremy says:

I have been struggling with this leak for a while. I have used the wrapper suggestion above but this still seems to leak for some reason. This all fires on Any ideas?

$(document).ready(function() {

var myAjax = function(settings) {
$.ajax($.extend({}, settings, {
complete: function (xhr, status) {
if (settings.complete) {
settings.complete(xhr, status);
}

// Avoid IE memory leak
xhr.onreadystatechange = null;
xhr.abort = null;
}
}));
};

function GetMyDataGrid() {

myAjax({
url: “includes/AjaxDataCheck/CheckForDataRefresh.cfm”,
cache: false,
dataType: ‘html’,
timeout: 5000, // 5 seconds,
data: “datacheck=DataGrid”,
success: function(result){
if (result.indexOf(“true”) >= 0) {
myAjax({
url: “includes/myinclude.asp”,
cache: false,
dataType: ‘html’,
timeout: 5000, // 5 seconds
data: “datacheck=MyVMs”,
success: function(data){
$(“#tab1response”).html(data);
}
});
}
}
});
};
setInterval(GetMyDataGrid,2000);
GetMyDataGrid();
});

Jeremy says:

Patrick, thanks for the reply. There were a few things messed up when I pasted the code in. I used your reworked version and it seems at first glance to be working. I won’t know for sure until I let it sit for a while. I am keeping my fingers crossed. The weird thing is that I could have sworn I used xhr.onreadystatechange = null and was getting the the Type mismatch error from IE, but I don’t seem to be getting that now for some reason.

I was not aware that jQuery 1.5 had been released. I will go ahead and use that version as well.

I sure will be glad to get this resolved as many users like to leave this particular page open virtually all the time.

Thanks again for the response.

Patrick Bacon says:

Michael: You are welcome. And thanks for confirming that 1.5.2 fixes the problem. That’s good to know going forward.