Skip to content

One brain for your lakehouse

The control plane is the single place you manage your catalog, your roles, and your compute workers. It ships as one binary, so there is no separate metastore or scheduler service to operate. It runs on your infrastructure. Your data stays on your storage.

Catalog of schemas, tables, columns
Roles, grants, and audit log
Compute-worker registry and scheduled jobs
Admin manage Control Plane one place to manage everything assign work Worker your VM Worker your VM read & write Your Storage (S3 / ADLS / Disk)

What the control plane stores

Metadata about your platform, not your data

Objects

Schemas, tables, columns, views, and pipelines, along with the storage credentials that point to them.

Access

Users, built-in and custom roles, and GRANT or REVOKE on catalog objects.

Activity

An audit log of who did what, and a registry of compute workers with heartbeats.

Jobs

Cron-scheduled SQL and pipeline runs, with run history.

How access is enforced

Roles and grants, enforced by the SQL engine before any read or write

Users

Identities that authenticate to the control plane and act on its behalf.

Roles

Built-in administrative roles, plus your own custom roles.

Grants

GRANT and REVOKE at schema and table level.

Scheduling and worker lifecycle

Cron jobs and a live worker registry, in the same binary

Scheduled jobs run SQL and pipelines on a cron expression, and every run is recorded with its history, so there is no external scheduler cluster to deploy or babysit. Maintenance work like OPTIMIZE and VACUUM, nightly loads, and pipeline refreshes all live on the same schedule surface you manage in the control plane.

Compute workers register with the control plane when they start and send heartbeats while they run. The registry always knows which workers are alive and what they are doing, which is the signal that lets the platform add workers under load and retire them when queries drain.

Where the control plane fits

One service alongside your workers, not a stack of them

Most self-hosted lakehouse guides have you assemble a Hive metastore, a separate catalog service, and a scheduler cluster before you can run a single query. The DeltaForge control plane replaces that assembly with one binary: the catalog, role enforcement, scheduling, and the worker registry all live in the same process you already installed.

The control plane is one piece of a larger picture. The architecture overview shows how it sits between your users, your compute workers, and your object storage. The security model covers how roles, grants, the credential vault, and the audit log protect your tables. And compute autoscaling explains how the worker registry grows and shrinks compute with query load.

Further reading: the Delta maintenance runbook shows the OPTIMIZE and VACUUM work you can put on a control-plane schedule, and GDPR right to be forgotten in SQL walks through a deletion workflow where grants and the audit log earn their keep.

Run the brain inside your network

You install and operate the control plane. There is no managed service to depend on.