Emond Papegaaij created WICKET-6861:
---------------------------------------

             Summary: Possible race condition in clearing and triggering of 
Wicket timers
                 Key: WICKET-6861
                 URL: https://issues.apache.org/jira/browse/WICKET-6861
             Project: Wicket
          Issue Type: Bug
          Components: wicket-core
    Affects Versions: 9.2.0
            Reporter: Emond Papegaaij


In our test suite we see some intermittent failures related to 
{{AbstractAjaxTimerBehavior}}. At a few places in our application, we use a 
background poller at a 1 sec interval that checks for an out-of-band submission 
of the form data. So the user either has to fill the form via the web interface 
or via some other route. Either route can complete the form, but as soon as one 
of the two is triggered, the other must stop. The race condition lies in the 
{{AbstractAjaxTimerBehavior}} triggering while the user has already submitted 
the form manually.

The naive implementation would stop {{AbstractAjaxTimerBehavior}} via 
{{Wicket.Timer.clear('timer0')}} in the Ajax response of the form submit. 
However, this results in a very large gap between the moment of submission and 
the actual moment the timer is stopped. To fix this, we've added the following 
code to the {{AjaxSubmitLink}}:

{code:java}
@Override
public void renderHead(IHeaderResponse response) {
    super.renderHead(response);
    response.render(OnDomReadyHeaderItem.forScript(
            "Wicket.Event.add('" + getMarkupId() + "', 'click', " +
            "function() {Wicket.Timer.clear('" + getTimerId() + "');})"));
}
{code}

This puts the call to {{Wicket.Timer.clear}} before the actual Ajax submit and 
should therefore prevent the timer from triggering after the Ajax submit is 
done. However, in very rare occurrences we still see the timer triggering. When 
it happens, the timer is always directly after the Ajax submit (often within 
10ms). This makes us believe the current implementation has a race condition in 
the following way:

* The user clicks the Ajax submit link
* The execution of the event handlers is started in the browser
* At this moment the {{setTimeout}} triggers, but it is suspended because JS is 
single threaded and the browser is already execution JS in response to UI 
interaction
* The first event handler now clears the timer
* The second event handler performs the Ajax call
* Now the JS executor is freed and the timer function is executed

Although I'm not entirely at home in the execution model of JS in a browser, 
this is the only explanation I could come up with. There is no way to reproduce 
it, other than run a complex test suite 1000ths of times. My proposed fix is to 
replace the timeout function in {{wicket-ajax-jquery.js}} in 
{{Wicket.Timer.set}} with this:

{code:javascript}
Wicket.TimerHandles[timerId] = setTimeout(function() {
    if (timerId in Wicket.TimerHandles) {
        Wicket.Timer.clear(timerId);
        f();
    }
}, delay);
{code}

This should prevent the function {{f()}} to be executed after the timer has 
been cleared (Wicket.Timer.clear deletes the {{timerId}} from 
{{Wicket.TimerHandles}}.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to