krabka-operator

The Kubernetes operator for krabka: a Kafka cluster, its node pools, listeners, topics, users, connectors, schema registry and gRPC gateway, reconciled from custom resources.

Custom resources

All in group krabka.io, version v1alpha1:

The manifests live in charts/krabka-operator/crds, one krabka.io_<plural>.yaml per kind. Regenerate them from the Rust types with:

tools/regen-crds.sh

CI runs the same script and fails when the working tree changes, so a change to a CRD type has to reach the manifests in the same commit.

Chart

charts/krabka-operator installs the operator, its RBAC, and the CRDs. Helm creates the files in crds/ on install. It does not touch them on upgrade, so a schema change needs a kubectl apply of that directory.

Scope

This operator reconciles Kafka. Gres — the Postgres-compatible layer — is not here: its controllers reached about 21k lines of storage engine across gres-substrate, pgexec, pgkv and gres-ranges, none of which have been extracted, and it is under active development. It returns once those crates land in the organisation.

The chart does not install the Gres and GresTenant CRDs. They come back, from the generator, with the Gres controllers.

Layering

Depends on three sibling repositories, pinned by revision in Cargo.toml’s [patch.crates-io]:

Build

cargo test --workspace

bazel test //...

Both are supported and both are gated in CI. Cargo stays the dependency source of truth; Bazel reads the same Cargo.toml and Cargo.lock.