Availability

Availability is the fraction of time a system is operational, expressed as a percentage SLA. The gap between 99.9% ("three nines") and 99.999% ("five nines") is enormous: three nines permits ~8.7 hours of downtime per year while five nines permits only ~5 minutes. Availability is achieved through redundancy (no single points of failure), failover automation, and health-check-driven traffic routing. Composite system availability multiplies individual component availabilities — three 99.9% components in series yields only 99.7% end-to-end.

Key Points

• Serial dependencies multiply unavailability: A=99.9%, B=99.9%, C=99.9% in series → 99.7% composite — always minimize critical-path depth.
• Active-active deployments across multiple AZs or regions eliminate single-AZ failures; active-passive cuts costs but adds failover delay.
• Health checks must test actual business logic, not just port liveness — shallow health checks mask degraded dependencies.
• Circuit breakers prevent cascading failures by fast-failing calls to degraded upstreams, shedding load before the caller times out.
• Bulkhead pattern isolates failure domains — separate thread pools or connection pools for critical vs. non-critical paths.
• Blue-green and canary deployments reduce deployment-caused downtime, which is the #1 source of avoidable outages.
• SLA vs SLO vs SLI: SLI is the metric (latency), SLO is your internal target (P99 < 200 ms), SLA is the external contractual commitment with penalties.
• Error budgets (100% − SLO) give teams a risk budget — burning 50% of the monthly error budget mid-month triggers a freeze on risky deploys.