Linux AMI test agents - firefox works, chrome only works once.

Hi all,

After some more digging into my issue with my webpagetest setup in amazon ec2, I’ve just discovered that it depends on the browser to be used on the test as to whether the server will give the agent subsequent tests, or just the first one.

If the agent has never run a test (new agent instance) the server will give it a test, not dependant on the browser.

If the agent has run a test, it seems the server will only give it the next test to run if that test is for Firefox or Firefox nightly, as so soon as the next test to run is for Chrome/Chrome Beta or Chrome Canary, the server never hands out the work, and so the test queue doesn’t get processed.

This happens until the server terminates the instance, then a new server gets created, can run one chrome test, then the server stops giving it work again.

So, for the moment I can run tests time after time after time with Firefox, but, as we all know, chrome is the dominant browser, and it would be really great if anyone could help in tracking down this issue :slight_smile:

-John.

Does it depend on the URL being tested? I don’t think it matters but how recent of a release is the server running?

Just wondering if there are some test options that were changed on the server for the new agent and not having them (for Chrome) is causing the agents to hang.

If you ssh into the agent as the “ubuntu” user (if you spin up an agent with ssh keys) and connect to the screen session with “screen -r” you should be able to see what the agent is doing.

All of the Akamai-sponsored EC2 locations on the public WebPageTest are running the Linux AMI’s so I am fairly certain they work and it’s going to be a compatibility issue somewhere.

Hi Patrick!

Its not that the agent hangs, it still hits the getwork.php url on the server, but the server never gives out any more work.

Now strangely, I’ve just discovered that if I instigate the test via the API (Using sitespeed.io) it works with Chrome time after time, so it seems to be only when I’m running tests via the UI that the server doesn’t want to hand out more work to the agent.

I’ll dig into the agent more tonight and let you know…

[Edit] Scratch that, it seems to be URL specific, I’ll PM you with the details[/Edit]

-John.

Hi,

Even, i have faced this issue, but due to lack of time for debugging, i ignored it. Earlier versions worked fine, but recently i have launched New WPT Server from AMI and thats where i started facing these issues.

Please let me know if you find a fix for this.

Thanks,
Vikram