Availability

Availability measures whether users can complete critical flows despite failures. Improve it with redundancy, health-checked failover, multi-AZ deployment, circuit breakers, and graceful degradation — serving cached or partial results when dependencies fail.

Express targets as nines (99.9%, 99.99%) tied to allowed downtime per month. Verify with synthetic probes and error budgets, not assumptions. Isolate dependencies so one slow payment provider does not hang checkout.

On interviews: translate 99.9% to downtime minutes, describe degradation for a recommendation service, and list redundancy in your design.

Common pitfalls: availability counted only at load balancer while app errors spike; no dependency timeouts; cascading failures without bulkheads.

The trade-off is flexibility versus complexity—know when the simpler path is enough.

Checklist:

• State availability SLO with user impact.
• Add redundancy and health-checked failover.
• Degrade non-critical features under stress.
• Isolate dependencies with timeouts and breakers.