Skip to main content

Lakehouse and workflow experiments

Outcome​

In October 2026, Airflow, Spark, Trino, Flight Recorder, and Temporal were removed, along with their dedicated data. Shared storage, the registry, metrics, and Kubernetes logging stayed in place. ADR 0040 records the decision; ADRs 0031 through 0039 trace how the experiment grew.

What we learned​

  • A learning workload needs a review date and an exit path. Keeping a stack with no active users adds maintenance and alert noise.
  • Stop writers before deleting data.
  • Keep an operator installed until the resources it manages finish finalizing.

Limits​

Git can restore the definitions but not the deleted data. The old lakehouse instructions describe a retired experiment and do not authorize recreating it.