Progressive delivery with ArgoCD Rollouts - phased rollout strategy

[hortison]

Tier-1 services (SLO-backed) should use progressive delivery to reduce blast radius on deployments. Options:

1. ArgoCD Rollouts - canary or blue-green, integrates with ArgoCD, Prometheus metrics for automated rollback
2. Flagger - service mesh aware, more complex setup
3. Manual staged deploys - release to staging namespace first, then prod via PR

For v1 I'd recommend ArgoCD Rollouts with canary strategy:

• 10% initial canary weight
• Pause for 5 minutes, analyze error rate and p99 latency
• Auto-promote if metrics healthy, auto-rollback if SLO violated

This gives us automated quality gates without needing a service mesh.

[arjunmehta-git]

ArgoCD Rollouts is the right call - it integrates natively with ArgoCD's UI so teams can see rollout status in the same place as everything else. Prometheus analysis template design is the tricky part - we'll need a standard AnalysisTemplate per tier that teams reference.

[miacycle]

The UX is important here - ArgoCD Rollouts UI is excellent. Teams can see canary traffic split and can manually promote or abort from the UI without needing kubectl. This is a DX win.

[jamieplu]

For the AnalysisTemplate - let's define: error rate < 1% over 5min (Prometheus query), and p99 latency < 500ms. Teams can override thresholds in their Rollout spec but these are the safe defaults.