All Chrome tests failing with "Could not open connection: An unknown server-side error"

The error with not being able to switch to the "mediaStyleSelector" frame is not a timing issue.

For some reason, the browser/driver is not able to click the img tag in the WYSIWYG to select it. Because the image is not selected, when the browser/driver clicks the Add Media button in the WYSIWYG, what opens up is not the style selector but the media browser. The iframes have different names. I discovered this by changing the line in the test to try to open the "mediaBrowser" iframe, and that step passed.

Screenshot of original test (with a delay added):

Screenshot of altered test:

Working on this at the DrupalCon sprint with @dsnopek. Updating to a later ubuntu version on Travis might cut down on some of the errors.

I believe part of the reason this is failing is that the files path in behat.template.yml was not updated when the paths were renamed recently. I'll try updating the above patch to include this change.

@dsnopek, in order to get this far I think you had to update Travis to the latest trusty build and make some changes to the MySQL setup. Can you post that patch here as well?

Patch to change files path attached.

Actually the Travis files path is set differently. The change in the updated patch is still necessary for local testing, but shouldn't affect Travis.

Review of the screenshots generated by the test run for patch 4 indicate that the image is not being successfully uploaded, or not being rendered in either the WYSIWYG or in the node when you save it.

After doing some manual intervention in local testing (pausing the test and deleting the image entity after it's been inserted in the WYSIWYG the first time) I am able to cause the test to fail locally, even though it passes locally when run without interference.

Going through the test step by step, it appears that the browser/driver can select the broken image tag in the WYSIWYG. I'm working on a local child distribution that adds a class to images in the WYSIWYG editor when they are selected, to visually highlight them, and this is happening in the test even after the image entity has been deleted and the broken image is replaced in the WYSIWYG by the browser's default broken-image icon. So the custom step that clicks the img tag in the WYSIWYG is working the same way whether or not the image is loaded or rendered successfully. Doesn't seem like any change to this custom step would solve the problem, as it's not what's broken.

So it seems like the problem lies in the handoff from the selected image tag to the Media browser. If the selected image isn't recognized, the media browser opens instead of the media style selector.

We know from #2428097: Image styles won't generate on Travis-CI (breaking tests that need valid images) that image style generation doesn't work on Travis. My guess is that something changed recently in the browser or driver so that info about the broken-image icon is being passed to Media, instead of info about the expected image. If this can't be fixed in Media and/or the WYSIWYG editor, we'll have to rewrite the test to avoid the problem, or run the build on a webserver that can handle image style generation.