Azure Databricks Workspace
Does ZopNight manage Azure Databricks Workspace?
Azure Databricks workspace scheduling fans out to child resources: interactive clusters are terminated, pools scale to 0 idle instances, and SQL warehouses stop. ZopNight enumerates them through the Databricks API using ARM and AAD tokens, applies 15 RC-2200-family rules, and attributes cost at the VM layer; DBU charges are not priced.
Rules that fire on Azure Databricks Workspace
No active rule family targets Azure Databricks Workspace today. Rules that used to are retired, and retired rules publish no pages and fire no findings. Scheduling and permissions coverage are unaffected.
At a glance
| Field | Value |
|---|---|
| Scheduling notes | workspace-level scheduling fans out to child resources: interactive clusters are terminated, pools are scaled to 0 idle instances, and SQL warehouses are stopped. |
Azure Databricks is the first-party Databricks offering on Azure for data engineering, analytics, and ML. The workspace itself is the billing parent for clusters, pools, and SQL warehouses whose VMs dominate spend.
Two meters: DBUs above, VMs below
Every hour of Databricks compute is charged twice: once as DBUs (Databricks units, priced by workload tier) and once as the underlying Azure VMs, disks, and networking in the managed resource group. Both meters run whenever a cluster or warehouse is up, and both go quiet when it is not. ZopNight prices the VM layer through Cost Management; DBU charges are not priced, so the attributed cost understates the true total. The savings direction is the same, the magnitude is larger than shown.
Workspace fan-out scheduling
Stopping “a workspace” really means stopping the things inside it. A ZopNight workspace schedule fans out to child resources: interactive clusters are terminated, instance pools are scaled to 0 idle instances, and SQL warehouses are stopped. Job clusters need no scheduling, since they exist only for the duration of their run. Fan-out matters because the expensive objects live behind the Databricks control plane where ARM-level automation alone cannot reach them.
How ZopNight reaches inside the workspace
Discovered via Azure Resource Graph, then explored through the Databricks integration (ARM + AAD token) to enumerate clusters, pools, SQL warehouses, jobs, and model endpoints. Cost is attributed at the VM layer through Cost Management billing, and 15 Azure Databricks recommendation rules (the RC-2200 family) apply. This access requires adding the service principal to the workspace admins group. The child resource types (databricks-cluster, databricks-pool, databricks-sqlwarehouse) are documented in the Databricks catalog.
Databricks waste starts with idle clusters
The recurring leaks: interactive clusters with auto-termination disabled or set to hours, idling between notebook sessions on full VM billing; instance pools holding warm minimum capacity around the clock for daytime-only teams; and SQL warehouses left running after a dashboard review ends.
The workspace in the Azure portal
Azure portal → Azure Databricks lists each workspace, but the objects that bill live in the Databricks UI itself. Open the workspace and check Compute for clusters and pools, and SQL Warehouses for warehouse state.