If you've only ever tested failover in a staging environment, here's the gap between the demo and production reality: picking an RTO number is really picking an architecture, and the architecture is really picking a cost.
There are four patterns worth knowing cold. Backup-and-restore is cross-region snapshots plus a runbook — cheap (+10–20% over single-region spend), slow (4–24hr RTO). Pilot-light keeps a synced database and pre-wired networking in region two with compute scaled to near-zero — a solid middle ground (+20–40%, 1–4hr RTO). Warm-standby runs a genuinely live, scaled-down stack in region two continuously — this is where most production systems should land (+40–70%, 15–60min RTO). Active-active runs full capacity in both regions simultaneously behind a load balancer, with real-time bidirectional replication — expensive (+80–150%+) but near-zero RTO. Treat those percentages as planning ranges, not quotes: your real premium depends on how much of the stack you replicate, how you size the standby, and how chatty your data layer is.
Three things that bite engineers who haven't built this before: cross-region egress isn't free (₹1.5–2.5/GB between India regions adds up fast on a chatty database), reserved-capacity discounts (25–70% off on-demand) materially change which tier is affordable, and nobody budgets for the quarterly failover drill that actually proves the DR plan works instead of just existing on a diagram.
If you're building on India infrastructure specifically: AWS's Mumbai (ap-south-1) + Hyderabad (ap-south-2) pair, Azure's Central India + South India pair, and GCP's asia-south1 + asia-south2 pair are the natural defaults — not just for latency, but because RBI's payment-data localisation rule means DR copies of payment data can't legally live offshore anyway.
We wrote up the full RTO/RPO cost breakdown, including an INR pricing table by business size, here: Multi-Region Cloud Architecture & Disaster Recovery Cost for Indian Businesses in 2026.