Azure Machine Learning Workspace
Does ZopNight manage Azure Machine Learning Workspace?
Azure ML workspaces cost almost nothing themselves. The spend sits in the compute instances, clusters, and deployments they contain. ZopNight discovers each workspace via Resource Graph, then enumerates its compute, endpoints, deployments, and jobs through the AML enricher, attributing spend per child resource from Cost Management.
Rules that fire on Azure Machine Learning Workspace
No active rule family targets Azure Machine Learning 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 | discovery and topology only; compute instances and clusters within it are scheduled individually. |
Azure Machine Learning workspaces are the top-level container for ML assets: compute, endpoints, models, and jobs. The workspace is nearly free, but it anchors compute instances and clusters that are among Azure’s most-forgotten spend.
A nearly-free container that anchors real money
The workspace object itself carries no meaningful meter. What it anchors does: every compute instance, compute cluster, online deployment, and managed-compute deployment created inside it bills on its own VM-hour terms, and the workspace also drags along a storage account, a Key Vault, and an Application Insights resource that accrue their own charges. When an ML platform bill surprises a team, the workspace is where the investigation starts, but never where the meter runs.
How ZopNight maps a workspace’s contents
Discovered via Azure Resource Graph, then enriched through the AML enricher to enumerate compute, endpoints, deployments, and jobs. Each child lands in the inventory as its own resource with the workspace as parent, so Cost Management billing can be attributed across child resources rather than pooling into one opaque line. ML-compute recommendations apply to the children this enumeration surfaces.
Why nothing schedules at the workspace level
There is no stop operation on a workspace, and ZopNight does not pretend otherwise: compute instances and clusters inside it are scheduled individually, each with its own stop semantics. A workspace with ten notebook VMs is ten separate scheduling decisions, which is what lets one shared research workspace carry both an always-on team cluster and nightly-stopped personal instances.
The forgotten-workspace pattern
Abandoned experiment workspaces are the recurring leak: a project ends, the workspace survives, and inside it a compute instance or a cluster with a nonzero minimum node count keeps billing. Because the workspace looks like a single quiet resource in a resource-group listing, the child spend hides until something enumerates it. That enumeration is precisely the visibility the AML enrichment adds.
Reviewing a workspace’s inventory
Azure portal → Azure Machine Learning lists workspaces; opening one and launching Azure ML studio shows Compute, Endpoints, Models, and Jobs in the left navigation: the same children ZopNight enumerates, with their states.