zerodowntime

Zero-downtime, or zerodowntime, is the design goal of performing software deployments, configuration changes, or maintenance without user-visible interruptions. It aims to maintain continuous availability and consistent service level during changes that would traditionally require a restart or downtime. In practice, zero-downtime is achieved through architectural choices, deployment pipelines, and operational processes designed to avoid service disruption.

Key techniques include redundancy and fault tolerance, such as active-active or active-passive deployments, load balancing, and health checks. Rolling updates deploy changes to a subset of instances while routing traffic away from updated ones, gradually replacing instances. Blue-green deployments run two production environments in parallel; traffic is switched to the new environment once it is ready. Canary releases expose changes to a small portion of users for validation before full rollout. Feature flags allow disabling new code paths if issues arise.

Database and data migrations pose special challenges; patterns include online schema changes, shadow writes, dual writes with validation, and using read replicas to absorb load. Maintenance tasks can be scheduled with extendable automation and backout plans.

Operational practices include thorough testing, monitoring, canary metrics, and well-defined rollback procedures. Reliability targets such as service level objectives (SLOs), recovery point objectives (RPO), and recovery time objectives (RTO) guide design decisions.

Limitations: true zero-downtime is not always feasible, especially for services with strong external dependencies or long transactions. It often involves increased complexity and cost, and requires disciplined change management and testing.