The problem with traditional Kubernetes deploys

  • Previously we used kubectl apply or helm install
  • This is a "push" model
  • Even if we automate, traditional CI/CD is typically, "one and done"
  • It has limits:
    • Status is hard to get. CI/CD will usually show "success" or "failure"
    • CI/CD-aware rollbacks are unlikely
    • Drift (out-of-band change after deploy) is hard to detect or nonexistent
    • CI/CD now has admin cluster access. To all clusters! 😱
slide 106 (click 0 of 8)