Regional retail brand · India · 2025
Zero-downtime releases for a peak-traffic storefront
Problem
Manual releases took ~4 hours with storefront downtime. Staging drifted from production, so bugs only appeared after cutover.
Constraints
- No multi-week freeze during festival traffic windows
- Ops team of three had to own runbooks after handoff
- Budget capped — reuse existing AWS account, no greenfield org
Architecture & stack
- EKS workloads with health-gated rollouts
- GitLab CI: build → scan → deploy to staging → promote
- Blue-green service cutover with instant rollback path
“We finally ship on Fridays without a war room.” — Head of Engineering
Post go-live: Client keeps cluster access, GitLab pipelines, Grafana, and Terraform/Helm repos. We stayed on a light retainer for the first two peak weekends.
Discuss a similar engagement