The plan, infrastructure, and tested runbook that bring a system back from a region-level loss, fire, flood, regional cloud outage.
Disaster recovery (DR) is the discipline of preparing for and recovering from disasters that take out an entire region or datacenter, rather than the routine outages high-availability handles. DR has two parameters: RTO (Recovery Time Objective, how fast you need to be back up) and RPO (Recovery Point Objective, how much data you can afford to lose). A 1-hour RTO and 15-minute RPO is dramatically more expensive to engineer than 24-hour and 1-hour. The plan must be tested, untested DR is a wishlist.
Most teams have a DR plan; few have a tested DR plan. The first time the plan runs is during a real disaster, where every untested assumption breaks at once: stale credentials, undersized standby capacity, DNS TTL too long, missing dependencies in the DR region. Quarterly DR drills catch those problems while there's still time to fix them.
See the part of the platform that handles disaster recovery (dr) in production.