As from the SELinux bug, we need to do some fixes to the GMR code to make it more robust & debuggable - If an exception occurs while collecting information we must log that so we can more easily diagnose similar problems in future - If a single data collector (like the process listing) fails, we should still be able to carry on and report the GMR with all other info, just record an error from the process listing in the GMR - There is absolutely no reason to be walking entire of /proc to get all processes. Given the process ID, you can access /proc/$PID/task/NNNN to get all threads that are children. We should see about improving psutil to address this.
Can you post the entire log when running in permissive? Also this bug is for 8.0 correct? I'm assuming that's why it's cloned here.
There is agreement that bug exists, however I am closing this bug due to following reasons: 1. The bug is only visible during debugging operation for nova services 2. There is known work-around which allows debugging to work, see comment below: https://bugzilla.redhat.com/show_bug.cgi?id=1292787#c10 3. Proper fix should be in both nova and python-psutil, however for debug only operation known work-around is sufficient. 4. Original bug is from 2015 with almost none following requests, which is low priority.