It seems like the right thing to do is write 'nobootwait' or 'nofail'
based on distro.uses_systemd()
** Description changed:
Snapshotted instances may fail to boot due to a change in udev in 15.10.
Starting with Ubuntu 15.10, "nobootwait" != "nofail".
+
+ For reference, Ubuntu's systemd
** Changed in: cloud-init (Ubuntu Xenial)
Assignee: (unassigned) => Ben Howard (utlemming)
** Changed in: cloud-init (Ubuntu Xenial)
Status: New => Confirmed
** Changed in: cloud-init (Ubuntu Wily)
Status: New => In Progress
** Changed in: cloud-init (Ubuntu Xenial)
Marking as "critical" owing to Azure instance death. Because Azure
instances consider ephemeral devices to be ephemeral per-boot, a stop,
resized, or fabric restarted instance may not come back if the instance
is presented with a new ephemeral device.
Linked branch has been tested.
** Changed
