Intern Concurrency Problem

By  on  

Update: I've found the root issue and have detailed it at the end of this post.

Over the past year I've created and implemented Selenium testing on the Mozilla Developer Network using the Intern interface created by SitePen.  Intern's been awesome;  sure there's a learning curve with async JavaScript coding but it's simple when you get the hang of it.

One problem I encountered with functional testing via service like BrowserStack and Sauce Labs is that we get failures we generally don't get  when testing locally.  When I tested with one browser everything went well but testing multiple browsers sent our tests into a spiral of transient failures.  When I tweaked one setting, however, everything went to plan:

// Maximum number of simultaneous integration tests that should be executed on the remote WebDriver service
maxConcurrency: 1,

Setting the maxConcurrency value down to 1 was all we needed to do.  Instead of all browsers spawning at once, each test runs within one browser and then within another.  Bingo!

As to what was causing the ultimate issue, I'm not quite sure.  The tests included authorization and login testing so it's possible there were overlaps in signing in and out, causing confusion on the server side.  Regardless, if you need to get things moving quickly, limit the maxConcurrency setting and you may start seeing loads more test passes.

Update: Firefox + Focus + Selenium Bug

After loads of testing and digging, I found the root issue for my problems with tests passing when one browser is run and not when run concurrently: when Firefox is not the focused/"on top" browser, focus events don't get passed up. And in my specific case, I was testing CSS animations, and those don't occur when a browser isn't focused. Hopefully this bug is fixed in Selenium too!

Recent Features

  • By
    CSS vs. JS Animation: Which is Faster?

    How is it possible that JavaScript-based animation has secretly always been as fast — or faster — than CSS transitions? And, how is it possible that Adobe and Google consistently release media-rich mobile sites that rival the performance of native apps? This article serves as a point-by-point...

  • By
    fetch API

    One of the worst kept secrets about AJAX on the web is that the underlying API for it, XMLHttpRequest, wasn't really made for what we've been using it for.  We've done well to create elegant APIs around XHR but we know we can do better.  Our effort to...

Incredible Demos

  • By
    JavaScript Speech Recognition

    Speech recognition software is becoming more and more important; it started (for me) with Siri on iOS, then Amazon's Echo, then my new Apple TV, and so on.  Speech recognition is so useful for not just us tech superstars but for people who either want to work "hands...

  • By
    jQuery Wookmark

    The first thing that hits you when you visit Pinterest is "Whoa, the columns are the same width and the photos are cut to fit just the way they should."  Basic web users probably think nothing of it but as a developer, I can appreciate the...

Discussion

  1. Marco

    Yes, but your test suite will eventually take forever to finish.

    You could try improving the isolation of the different scenario’s. Have you tried using seperate users for each scenario?

    Best case you have a webservice for creating test users which you call at the beginning of every scenario.

    • I do use a web service to get test credentials for one test but I don’t want to flood our environments with test-only users.

  2. Marco

    Have you tried using separate users for each scenario?

    Sacrificing ui test concurrency is usually a bad idea, something you will definitely regret later.

    • Separate users would be a great idea and something I was hoping I wouldn’t have to do. Going to give that a shot today! :)

    • Marco

      Oops I accidentally double posted!

      Create a web-service to clean up the test-only users and use a new user for every scenario.
      That’s the safest approach

  3. Jeff

    Um… are you sure that your test failures aren’t a symptom of real race conditions in your code?

    • I think the race condition is login-related since testing the browsers individually works fine.

Wrap your code in <pre class="{language}"></pre> tags, link to a GitHub gist, JSFiddle fiddle, or CodePen pen to embed!