Mind to post the screenshots saved in jenkins? I gave #3 a try before but I can't reproduce it locally at all with chromedriver-helper (nor poltergeist). It always removed the override right away, I don't know why would it take 30s+ on Jenkins. Are there any memory stats for jobs in Jenkins?

Number 3 is synchronous Javascript (override_class_param), so I don't think that one failure caused by concurrency. Aside from that, Caoybara docs say that .find waits for Javascript :/

Notice new intermittent errors are showing up:

HostIntegrationTest::create new host page.test_0001_tabs are present
    HostIntegrationTest::edit page.test_0007_choosing a hostgroup does not override other host attributes

http://ci.theforeman.org/job/test_develop_pr_core/8119/

Found a reproducer - slow down the server thread by half the Capybara.default_max_time by changing ApplicationContrller to include

after_filter -> { sleep(15) }

If you set the default_max_time to 10, and slow down the requests by 5 seconds it's also easy to reproduce.

Some interesting findings:

When I slow down requests on ApplicationController with after_filter -> { sleep 5 } , Capybara.default_max_wait_time = 10, I get interesting results for 'correctly override global params':

click_button('Submit')                                                                                                                 
assert page.has_link?("Edit")

Fails , unexpectedly, as has_link? should work for 10 seconds, but it doesn't - the request returns in 5.1 seconds.

click_button('Submit')                                                                                                                 
wait_for_ajax
assert page.has_link?("Edit")

Fails, with Timeout error, unexpectedly. wait_for_ajax should work for 10 seconds, in that time the request has returned already.

click_button('Submit')                                                                                                                 
sleep 6
assert page.has_link?("Edit")

Succeeds - the request is definitely back by the time has_link? runs.

In short, if has_link has to wait for the server thread to return, it won't wait for Capybara.default_max_time, I don't think it waits at all. If has_link does not have to wait because the request is back (as in the example with sleep), then it succeeds.

I'm going to break this one up in a few others, as not all tests fail for the exact same reason.

More spotty test results:

http://ci.theforeman.org/job/test_develop_pr_core/8392/database=postgresql,label=fast,ruby=1.9.3/testReport/junit/%28root%29/HostIntegrationTest__clone%20page/test_0001_clones_lookup_values/

clones lookup values - "unable to find field login_login"

http://ci.theforeman.org/job/test_develop_pr_core/8392/database=postgresql,label=fast,ruby=2.2/testReport/junit/%28root%29/HostIntegrationTest/test_0002_show_page/

show page - "looked for h1 tag, not found"

http://ci.theforeman.org/job/test_develop_pr_core/8392/database=postgresql,label=fast,ruby=1.9.3/testReport/junit/%28root%29/HostIntegrationTest__edit%20page/test_0008_class_parameters_and_overrides_are_displayed_correctly_for_booleans/

class parameters and overrides are displayed correctly for booleans - Host::Base#name= delegated to primary_interface.name=, but primary_interface is nil

http://ci.theforeman.org/job/test_develop_pr_core/8392/database=postgresql,label=fast,ruby=1.9.3/testReport/junit/%28root%29/HostIntegrationTest__edit%20page/test_0001_class_parameters_and_overrides_are_displayed_correctly_for_strings/

class parameters and overrides are displayed correctly for strings - "Capybara::ElementNotFound: Unable to find css "textarea:enabled"..."

Daniel Lobato Garcia wrote:

More spotty test results:

http://ci.theforeman.org/job/test_develop_pr_core/8392/database=postgresql,label=fast,ruby=1.9.3/testReport/junit/%28root%29/HostIntegrationTest__clone%20page/test_0001_clones_lookup_values/

clones lookup values - "unable to find field login_login"

This should be fixed by a combination of #14097, #14067 and #14291.