1927158 - [Experiment] Users will be unenrolled from the “userrestricted” branch of the experiment after a browser restart

Status: VERIFIED FIXED

(Core :: Security: Process Sandboxing, defect)

Description

[Alexandru Bejenaru [:abejenaru], Ecosystem QA]

[Notes]:

• After enrolling in the “userrestricted” branch, the "security.sandbox.content.level" pref is set to "8" and it is marked as a user set pref. After the browser is restarted the experiment targeting is re-evaluated and see that the pref is now marked as a user pref set and unenroll the users with the "did not match due to targeting" reasoning since the targeting conditions are no longer satisfied.
• The issue is not reproducible with the “control” branch of the experiment, since the pref is not changed after enrollment.

[Affected versions]:

• Firefox RC 132.0, BUILD ID: 20241021175835.

[Affected Platforms]:

• Windows 10.

[Prerequisites]:

• User is enrolled in the “userrestricted” branch of the USER_RESTRICTED for content process experiment.

[Steps to reproduce]:

1. Open the browser with the profile from the prerequisites.
2. Restart/relaunch the browser.
3. Access the “about:studies” page and observe the behavior.

[Expected result]:

• The “USER_RESTRICTED for content process” experiment is displayed in the “Active studies” section.

[Actual result]:

• The “USER_RESTRICTED for content process” experiment is finalized and moved to the “Completed studies” section.

Comment 1

[Bob Owen (:bobowen)]

I think I’ve fixed the targeting. I think the issue was excluding the pref we were using in the targeting config.
I’ve used a different target that targets windows 10+, which is all windows users now anyway on current versions.

You can exclude people who have set the pref through Nimbus and it only checks if they are not already enrolled in the experiment, so I think that should work. It would be great if you could confirm before full testing.

Comment 2

[Alexandru Bejenaru [:abejenaru], Ecosystem QA]

Hello. We were unable to reproduce this issue anymore using Firefox RC 132.0, BUILD ID: 20241021175835 on Windows 10 x64. The fix was done by modifying the targeting of the experiment. We can confirm that the user is no longer unenrolled from the experiment in the given scenario and now works as expected. As a result I'm closing this ticket as Resolved > Verified Fixed, thanks!