Geo-partitioned ride-hailing database

Design a geo-partitioned database for a ride-hailing platform where most reads and writes for a trip stay within one city, but cross-city operations (a rider traveling, fleet analytics) exist. Scale is 500k active trips, 40k writes/sec concentrated in dense metros, with a 50ms in-region write SLA. Some metros are far hotter than others. How do you partition data by geography, place leaders, and handle a metro that suddenly doubles in traffic?