devops

Choose ECS/Fargate when running fewer than 20 services on AWS without Kubernetes expertise. Choose managed Kubernetes (EKS, GKE, AKS) when you need multi-cloud portability and have a platform team. EKS on Fargate removes node management on AWS.

Use Terraform/OpenTofu for multi-cloud or compliance-heavy environments, Pulumi when you want general-purpose languages for complex logic, and AWS CDK for AWS-only stacks needing rich L2/L3 constructs. Always use remote state with locking and pin module versions.

Adopt multi-region only for contractual RTO under 30 minutes, data sovereignty requirements, or sub-50ms global latency needs. The default for most startups is three availability zones in one region with documented failover runbooks.

Use SLO-based alerting on error budget burn rate instead of infrastructure thresholds. Every alert must require immediate human action; otherwise delete it, ticket it, or move it to a dashboard. Target fewer than two pages per shift per day.

Elite performers deploy multiple times per day, achieve lead time under one hour, restore service in under one hour, and maintain a change failure rate of 0-15%. Track team-level trends over rolling 90-day windows rather than snapshots.

Avoid a service mesh when you have fewer than 15 services, no dedicated platform team, or when API gateway mTLS plus cloud-native private connectivity suffices. Istio adds roughly 20-30% overhead and significant operational complexity.