How do you build a highly available system in the cloud?

High availability is about continuing to serve traffic through component failures. It requires redundancy, detection and automatic failover. The design should identify every single point of failure, including the ones outside compute, such as DNS, NAT, a shared load balancer or a single database instance.

Patterns

• Deploy stateless tiers across at least two AZs behind a load balancer
• Use health checks and automatic replacement or failover
• Replicate databases with managed multi-AZ failover or synchronous replicas
• Externalise sessions and files so any instance can serve any request
• Add graceful degradation, retries with jitter and circuit breakers
• Remove cross-zone dependencies such as a single NAT gateway

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: topology.kubernetes.io/zone
    whenUnsatisfiable: ScheduleAnyway
    labelSelector:
      matchLabels:
        app: api

Failover must be tested. A standby that has never received traffic may have stale configuration, missing secrets or an untested DNS change. Game days and controlled failovers expose those gaps before a real incident does.