Zero-downtime deploy FAQ

Will blue/green double our infrastructure cost?

You will momentarily pay for both blue and green during rollout windows, but automation should decommission the idle environment within minutes of promotion. Track cost per release so finance understands the short-lived overhead and plan reserved capacity accordingly.

Can we adopt this without a service mesh?

Yes. Many ingress controllers (ALB, NGINX, Traefik) support weighted routing or header-based routing. If that is not available, use load balancer target groups or DNS weighted records as an interim step while evaluating a mesh.

How do we handle database schema changes?

Design migrations to be backward compatible. Ship schema updates first, keep both versions live during traffic shifting, and only remove deprecated fields after the new application version is stable. For destructive changes, use feature flags or copy-on-write strategies.

What happens if a health gate fails?

Argo Rollouts pauses the deployment. Your runbook should trigger an automated rollback to blue, capture diagnostic logs, and page the owning team. The failed version stays available for investigation without pressuring engineers to fix forward.

Who owns the rollout?

Platform engineering typically operates the tooling, but application teams must be on bridge calls or chat channels to validate business metrics and make go/no-go decisions. The checklist ensures everyone knows their role ahead of time.