Learning experiments and platform scope
Anton welcomes learning, but each experiment needs a reason, an owner, and a way out.
The old playground
The earlier playground collected examples for ClickHouse, Dragonfly, Redpanda, ScyllaDB, and TiDB. They showed how to try each product; they were never a reliable list of what was deployed. Their install recipes are no longer current guidance.
Some older notes also ran applications imperatively, outside Flux. That was a different ownership model, not a way around today's GitOps and approval rules.
The rule since the rebuild
The structured rebuild separated two reasons to adopt a component: a concrete consumer, or contained learning. A learning experiment needs:
- a review date,
- agreement that its data is disposable, and
- an exit plan.
How recent experiments ended
| Experiment | Outcome |
|---|---|
| ClickStack | Ended; Kubernetes logs moved to OTel and Loki |
| Lakehouse and workflows | Retired |
Database operators still running today serve platform consumers; they do not revive the playground. The capabilities map shows that boundary.