High availability
Abbreviation: HA
High availability (HA) means designing a system to keep serving users when individual components fail, usually by running redundant copies across separate failure domains. It is measured in nines: 99.9% availability allows roughly 8.8 hours of downtime a year, 99.99% allows under an hour.
High availability (HA) means designing a system to keep serving users when individual components fail, usually by running redundant copies across separate failure domains. It is measured in “nines”: 99.9% availability allows roughly 8.8 hours of downtime a year; 99.99% allows about 53 minutes.
In plain terms
It’s two engines on the plane. The standard cloud pattern: two or more app servers spread across different availability zones, a load balancer routing traffic only to healthy ones, and a database with a standby replica that takes over automatically. AWS calls the database version Multi-AZ; Azure uses availability zones and zone-redundant services; Google Cloud offers regional instances and managed instance groups. Lose a server, or an entire zone, and users never notice.
Why it matters when you migrate
- The cloud makes HA affordable for the first time. On-premise redundancy meant a second server room; in the cloud, a second zone is a dropdown and roughly double the instance cost for the redundant tier. Most teams arrive with less resilience than they think and can leave with more.
- It roughly doubles the bill for what it covers, so choose per workload. The customer-facing API earns multi-zone; the internal reporting tool likely does not.
- HA is not backup. Redundancy survives hardware failure; it happily replicates a bad deploy or a deleted table to every copy. You still need backup and disaster recovery.