The same app, three regions, one domain

Users in Singapore feel every one of the 180 milliseconds to your Frankfurt box. The usual fix is a re-architecture with a load balancer product at the front.

Here it's three rentals and a snapshot. Copy the release snapshot to a US box and a Singapore box; each region's container gets its own proxy alias; your DNS provider's GeoDNS answers api.yourapp.com with the nearest alias. Deploys stay boring: the same snapshot lands on all three, so the regions cannot drift. A region having a bad day is a DNS record you repoint.

The global footprint costs three flat-rate boxes and one DNS feature you probably already pay for.

Latency became a DNS answer, not an architecture.