If the database goes down, it will indeed be impossible to determine who should be master. Instead, both nodes will run as “provisional master”, meaning that both nodes will operate, but historical data will be cached. When the database is available again, they will determine who should be master based on the standard criteria as well as the amount of cache data. When the decision is made, the master will write its data cache to the database, and the backup will delete its cache.
Hope this answers your question,