Operations and management

Onehouse runs as a serverless data platform inside your account, operated by the team that built some of the largest data lakes in the world. You keep control of your data, your network, and your cloud costs, and still get a fully managed lakehouse. Most vendors do not offer that combination. The same design is what makes the bring your own Kubernetes path possible.

Operational excellence

Onehouse brings world-class lakehouse operators to your own environment. Onehouse monitors, tunes, upgrades, and supports your platform around the clock from the control plane, while the data, the compute, and infrastructure control stay in your account. Teams get the operational depth of a large platform team without building one, and the Onehouse engineer who answers an incident at 3 a.m. has seen it before.

That depth comes from a decade of running lakehouses in production. The same engineers created Apache Hudi at Uber, the first lakehouse in production, and lead Apache Hudi and Apache XTable (Incubating) today. They have resolved more than 2,000 issues for the open-source community over ten years, and the projects they lead run at Amazon, Uber, Walmart, Notion, and JD. AWS and Google Cloud ship these OSS projects pre-installed in their data services, and Google and Microsoft develop XTable alongside the team, so the open-source lakehouse support those clouds offer builds on this team’s work.

Day to day, that means 24/7 support for troubleshooting, tuning, and architecture guidance, proactive alerting on lag and failures, and fixes that reach your environment as platform rollouts. For security, the team works from metadata and metrics over the same audited management path used for installs and upgrades. Any further access follows a break-glass policy that you control. For large customers, Onehouse support engineers can work as contractors inside the customer’s organization, under its operational policies.

How the platform is operated

Everything is a container. The data plane is a set of Kubernetes workloads and nothing else: the agent, the Spark operator, the driver and executor pods that run your jobs, Prometheus and Grafana, and the telemetry collectors. There are no virtual machines to image, no hosts to patch beyond the node images your Kubernetes provider manages, and no software installed outside the cluster. The same deployment runs on EKS, GKE, AKS, any Kubernetes distribution, and a cluster you built yourself.

Installed and upgraded remotely. After the customer stack has created the two identities, the control plane installs the platform components onto your cluster through the Kubernetes API, over the management tunnel your cluster opens outbound. Upgrades use the same path: new container versions roll out across the data plane, driven from the control plane, with no engineer in your VPC and nothing to download or apply on your side. Onehouse manages versions across every customer, so a fix or a new capability reaches your environment as a rollout. Rollouts are staged across environments, and you can pin a version. Large deployments can set a maintenance window; otherwise upgrades roll every few weeks with minimal downtime.

Serverless inside your VPC. Compute is allocated when there is work and released when there is not. Between syncs, Flows use no compute beyond a lightweight poller that checks for new data on behalf of every Flow on the cluster. A cluster scales down to zero when idle, and an instance that sits idle for ten minutes is terminated. You set the ceilings, per cluster and across the project, and the runtime operates inside them. You get the experience of a hosted serverless service, with the data and the compute in your own account.

Operated at scale. Onehouse runs data planes of this design in production for large enterprises from a single control plane. The operational tooling (health checks, autoscaling, alerting on lag and failures, and the Grafana dashboards) is the same tooling Onehouse uses to run those environments, delivered into yours.

Onehouse is the only lakehouse vendor that delivers a fully managed platform on Kubernetes you control: installed, upgraded, and operated remotely, with your tables and the compute on them never leaving your account. Other vendors offer either a fully managed platform in their cloud or software in yours that your team operates.