Our functional automation indicates failure on this platform only, Ubuntu 13.0.4 32/64bit. A sequence of tests causes the modal dialog to open in the background, and for the overall performance of Firefox to degrade. Our tests eventually fail. This is a regression from 17.0.8. It does not occur on Ubuntu 12.0.4 either. The suite of tests is here: http://hg.mozilla.org/qa/mozmill-tests/file/mozilla-esr17/tests/functional/restartTests Adding Henrik for more detail as needed. Please note that a simple XPI install seems to be fine. The cause of this particular issue is currently not known.

Mea culpa - not a regression. I didn't do an apples-to-apples comparison, only compared OS config. When I try 17.0.8 on this exact config, I get the same results as 17.0.9. So, likely a machine- and/or OS-specific issue. Based on the facts that a) not new behavior and b) only on Ubuntu 13.04, I'm going to guess it's more likely an issue with our test infrastructure. Or at the very least, we need to debug further on our side.

Andrei, someone from your team should take this. It is reproducible on the 13.04 machines and we have to get this investigated ASAP. It could be a regression in Firefox, given that it also appeared in nightly tests now.

Cosmin, please look into this issue.

I tried to reproduce this in the morning but I was unsuccessfully, and later machines were busy or offline, so I will look into this again later.

I connected again today on machine mm-ub-1304-32-2, and still couldn't reproduce the issue http://mozmill-crowd.blargon7.com/#/functional/reports?branch=17.0&platform=Linux&from=2013-09-22&to=2013-09-25 Can you please give more details about how you reproduced this failure, or what was the error message ?

I'm totally not sure why you cannot reproduce it. I triggered an example ondemand testrun for 17.0.9esr via Jenkins on mm-ub-1304-32-3 now and it failed immediately with this problem. I would really love to get more feedback while you are trying to reproduce an issue, and it doesn't show up. Please use those steps to nail-down the problem. There is no error, but the addon installation dialog opens up with Mozmill wants to close the browser for a restart. Given that this is a modal dialog and Mozmill cannot close it, it hangs around forever and blocks this machine from further testing.

See: http://mm-ci-master.qa.scl3.mozilla.com:8080/job/ondemand_functional/1784/console

This one is failing only when running ondemand tests, not reproducible in a regular testrun or with mozmill-restart, and I don't thing is a regression since in reproduces with 17.0.7esr, 17.0.8esr, 17.0.9esr, I'm currently looking into this, I hope I can find the problem today.

I tried to add the System Variables under ~/.bashrc, and refresh the profile with source ~/.bashrc, but it did't reproduced, then I added the variables under the mozmill-env/bin/activate and still did't reproduce the failure in regular testruns, today I upgraded my machine to Ubuntu-13.04 and I will keep investigating this failure on my machine, as is quite dangerous to work on system variables.

I worked with Andreea and Mario on this, and we set the system variables under jenkins-env/bin/activate and run ondemand test to Ubuntu-13.04 slaves and still did't reproduce the issue. What I would like to try next is to add my Ubuntu-13.04 machine to master-ci and run ondemand tests, so I will have a clue if the problem is on Ubuntu-13.04 configuration or on jenkins (master-ci).

We managed to add the variables correctly, in the ondemand job we pass the env.properties file at the "Prepare an environment for the run" step. We checked those after each build finished running at the Environments variables (in the job). We have the same variables as in mozmill-ci, except: no_proxy, HTTPS_PROXY, ftp_proxy, http_proxy. With Ubuntu 13.04 64 bit and esr17.0.9 it's not reproducing locally. Cosmin is looking now at a remote 64 bit machine as I don't recall seeing this on 64bit machines, just to be sure. Maybe it's only on 32 bit.

I could't find the root of this issue, and as I can't think to anything else to look for I unassign from this bug. The investigation I did so far: I took one of the affected nodes and set the environment variables as on master-ci, I added them on: ~/.bashrc, and then I runned "source ~/.bashrc" mozmill-env/bin/activate , and then I runned "mozmill-env/bin/activate" Then I set an local mozmill-ci and set the env variables on node , then on jenkins-env/bin/activate, then on start.sh, and finally on job as Andreea described on comment 11.

After the upgrade to Mozmill 2.0 I noticed that nothing has been changed here. But what's great to see is that we do not run into a 60 minutes build timeout anymore. Mozmill now successfully kills the hang process after 60s of inactivity.

This will be fixed by the proxy changes on bug 916724.

This is fixed now via bug 916724.