knext
Learn knext

9 · Going further

What you built, how to take it down, and what to read next.

What you built

  • ACME, deployed to Knative on your own cluster, on a real URL
  • Scaled to zero — no pods, no container compute, when idle
  • A Postgres database bound by Secret, which knext neither created nor manages
  • That database also scaled to zero, so idle ACME costs nothing at either tier
  • A CI pipeline that redeploys on every push, with no cloud credentials in it

The through-line: you describe what you want, the operator makes it true. You never created a Knative Service by hand, and nothing edited the cluster behind the NextApp resource's back.

Taking it down

shell
kubectl delete nextapp acme

The operator removes what it created.

Deleting the NextApp also reaps app-scoped external state. A finalizer clears the object-store keys under ACME's own <app-name>/ prefix and the Redis keys under its cache prefix. If you have parked anything else under that prefix in the same bucket, it goes too.

Registry images are not touched — those remain yours. Your database is never touched — knext did not create it and does not delete it.

To remove scale-zero-pg as well, delete its namespace. That does destroy the data, because that data lives in the storage plane you deployed in chapter 7.

Where to go next

Operating it

  • Scale to zero — the autoscaling knobs, and what to change when cold start matters more than idle cost
  • Rollback — pin traffic to a previous revision when a deploy goes wrong
  • Previews — per-branch environments
  • Observability — Prometheus metrics and Grafana dashboards

Understanding it

Running it well

  • Databases — poolers, the writer/reader split, the typed client
  • Security — digest pinning, NetworkPolicy defaults, supply chain
  • Upgrading — why the operator is upgraded before the CLI
  • Multi-cloud — what is verified where

Two honest caveats

Cold start is environment-dependent. The numbers in chapter 5 and chapter 7 were measured on specific clusters and are published as distributions rather than headline figures. Measure your own before designing around a number.

Verified is not the same as claimed. The compatibility matrix is explicit about which capability rows are backed by a check that goes red on failure and which are not. When you need a guarantee, read that page rather than a feature list.

Building something real

You have the whole path now: adapter, config, deploy, scale-to-zero, database, CI. The natural next step is to build an actual app on it rather than a scaffold — a repo of your own, using these tools end to end.

If you hit something this course did not prepare you for, that is a documentation bug worth telling us about.

On this page