Netflix engineers from the JVM ecosystem team share how they scaled automated dependency updates across 3,000+ applications and 5,000+ repositories. They cover five pillars: build opinions via a custom Gradle wrapper (Nebula) with resolution rules and dependency locking; artifact observability through an internal tool called Astrid that indexes every deployment and tracks library versions in production; distributed refactoring using nightly automated PRs and a campaign tool for migrations and security incidents; continuous delivery improvements that pushed daily deployment rates from under 50% to over 60%; and testing/validation infrastructure including testing scorecards, predictive test selection, flaky test detection, and remote test execution via Gradle Dev Velocity. The talk emphasizes that tooling alone is insufficient — cultural change and developer confidence are equally critical to making automated code changes stick.
Sort: