Running distributed storage on Kubernetes is an orchestration problem, not a packaging one. Apache Ozone is an open-source, petabyte-scale, S3-compatible object store. While tools like Helm can deploy it, running it reliably in production requires an intelligent K8s Operator.

Distributed storage introduces unique challenges on Kubernetes: maintaining a strict boot order between metadata and storage nodes, ensuring stable network identities for Raft consensus, executing idempotent cluster formatting and managing dynamic topology changes at runtime.

The idea here is an opinionated Kubernetes Ozone Operator. It adopts an S3-only interface with erasure coding and a simplified one-disk-per-pod model using standard Persistent Volumes to streamline the pod lifecycle. Crucially, the Operator handles various Ozone specific operations by understanding Ozone’s internal data semantics. It manages component configurations, handles node discovery, security and observability into the system.

Attendees will leave with an understanding of a practical framework for running complex, consensus-driven stateful systems in a cloud-native environment.