Pre-Production Cost Hygiene: Stop Paying for Idle Environments

Pre-production is the stealth budget killer. Staging, QA, and preview clusters rarely serve users, yet they run 24/7. A few guardrails can cut their cost in half while keeping release speed intact.

Right-size for intent

• Smaller node pools: Default to fewer, cheaper instances and scale up only during test windows.
• Lower HPA floors: Set min replicas to 1 or 2; avoid production-like padding unless load tests demand it.
• Use cheaper storage classes: GP2/GP3 over provisioned IOPS for anything non-prod.

Automate off-hours savings

• Scheduled scale-downs: Cron-based HPA caps at night and weekends; wake up before the workday.
• Suspend non-critical CronJobs: Archive runs during quiet hours; keep only synthetic monitoring alive.
• Ephemeral previews: Create namespaces on PR open, garbage collect on merge/close.

Control observability spend

• Shorter retention: 3–7 days of logs and metrics is plenty for staging.
• Sample traces: Keep SLO traffic at 100%, but sample aggressively for everything else.
• Sidecar limits: Clamp APM/logging sidecars to tight requests and limits.

Budget and enforce

• Separate pre-prod into its own projects/accounts so alerts and budgets are isolated.
• Apply stricter ResourceQuota and LimitRange than production; fail deployments that omit them.
• Report cost per environment weekly and set a target like “staging ≤ 40% of prod.”

Rollout checklist

• Baseline current staging spend and utilization.
• Apply template changes (limits, HPA floors, storage class) to the top 5 services.
• Add scheduled scale-downs and auto-clean preview namespaces.
• Re-measure after one week; adjust quotas and retention until the target is met.

Pre-production exists to de-risk releases, not to mirror every production cost. Shape it for short-lived testing, and your burn will drop without slowing shipping.***