stardelt Operator

Kubernetes operator for stardelt. Reconciles the Lakehouse, PlatformInstance, Tenant, Pipeline, StreamApp, and MLWorkspace CRDs — composing upstream best-of-breed projects (Trino, Lakekeeper, Iceberg, Spark, Airflow, RisingWave, KubeRay, …) into a single declarative experience.

Status

Pre-alpha — skeleton only, no reconcile logic yet. Part of the broader stardelt MVP / vibecoding push; everything here is expected to change fast.

Watch github.com/stardelt/stardelt-operator/issues/1 for the implementation milestone.

Planned CRDs

CRD	Purpose
PlatformInstance	Cluster-scoped: declares which stardelt pillars + foundations are installed
Tenant	Namespaced scope of data + identities; isolation boundary
Lakehouse	A managed Trino + Lakekeeper + Iceberg unit on object storage
Pipeline	Batch ETL workload (Spark/Airflow/dbt) bound to a Tenant + Lakehouse
StreamApp	Streaming workload (Kafka/Flink/RisingWave)
MLWorkspace	ML/AI environment (Ray/Kubeflow/MLflow/KServe)

Architecture

Built with kube-rs on Tokio. Each CRD has a dedicated controller (crates/stardelt-operator/src/controllers/<crd>.rs — planned). The operator does not re-implement upstream charts; it watches CRDs and applies/reconciles upstream Helm releases and native manifests via kube-rs server-side apply.

Dev

cargo check
cargo run   # logs a stub message and sleeps — no cluster required

Repo layout

Cargo.toml                       # workspace
crates/stardelt-operator/
  Cargo.toml
  src/main.rs                    # entry — currently a stub
chart/                           # Helm chart for the operator — TBD

Links

• stardelt.io/roadmap
• stardelt.io/architecture/overview
• stardelt.io/design/master-spec