I checked the environment and it looks like os-collect-config is not running on all the overcloud nodes(which might be the result of previous failed upgrade attempts):

(undercloud) [stack@undercloud-0 ~]$ for node in $(nova list | awk  '{print $12}' | grep -oP '[0-9.]+'); do ssh heat-admin@$node 'sudo systemctl is-active os-collect-config';done
inactive
active
active
active
active
inactive

As a workaround I suggest starting the os-collect-config service on the overcloud nodes and re-try the upgrade(not using ansible since the inventory is only available after running ffwd upgrade prepare):

for node in $(nova list | awk  '{print $12}' | grep -oP '[0-9.]+'); do ssh heat-admin@$node 'sudo systemctl start os-collect-config';done

I am closing this bug as a duplicate of BZ#1594211 where we keep track of this issue: re-running upgrade may fail with time out after failed upgrade attempts and when os-collect-config service is not running on the overcloud nodes.

*** This bug has been marked as a duplicate of bug 1594211 ***

clearing needinfo.