After a host dies, the order of the recovery decides whether it works. The secrets come first, then the data and the database dump, DNS last, and one check that proves the service is actually back.
A backup job that exits zero is a copy job until you restore something from it. The checks worth scheduling, the drill that finds the real failures, and the recovery kit people forget.
The new stable treats the data folder's permissions as a hard requirement: exact permissions or it will not run, with three documented ways out for the setups that cannot comply.
