Ceph

the backport cannot be a cherry pick from any of the commits at https://github.com/ceph/ceph/pull/5999/commits. I assume it is out of the question to change the systemd / udev chain in hammer to allow for a proper asynchronous run of ceph-disk activate, although it is key to stabilize ceph-disk.

Since we'll have to keep ceph-disk activate in hammer running from udev, I should first figure out exactly why and when udev interrupts an action. What is the timeout ? And also check if ceph-disk indirectly runs the exact same action or even worse, something that could remove the device (partprobe).

When this is clarified, we'll need to figure out a strategy to fix those in a hammer specific way.