Progressive delivery
An evolution of continuous delivery where new versions are rolled out gradually to ever-larger portions of users, with automatic monitoring of metrics that governs whether the rollout should continue or roll back. "Deploy with a safety net".
Idea: instead of a deploy being binary (all or nothing) it's released in steps — 5% of traffic, then 25%, 50%, 100% — and at each step key metrics (error rate, latency, business metrics) are analyzed automatically. If the system sees a degradation, the rollout is stopped and rolled back automatically, often before a human even noticed. Techniques involved: canary (a gradual traffic share), feature flags (turn on features for selected groups), and automated analysis. Tools: Argo Rollouts, Flagger (on top of a service mesh). Win: it minimizes the "blast radius" of a bad deploy — only a small fraction is affected, and the rollback is automatic. It builds on canary and feature flags, but with automated, metric-driven progression. The modern ideal for risk-free delivery. Related to canary deployment and feature flag (ops).