Attempts affect pre failures. This means we can increase stability of
jobs by rejecting nodes that fail pre without failing runs at the same
time (unless we are really unlucky and hit b0rken nodes 5 times in a
row).
This change is adapted from I17b7f878c742fa8db66f738526855a02ab9f1905 in
kolla-ansible.
Change-Id: Ied5d2cb8ffa47b90833ee3cf241797601906f9b2
98fb7dc8ff