Edit

Delete Microsoft Discovery resources

Applies to: Microsoft Discovery (Public Preview)

When you no longer need a Microsoft Discovery deployment, delete its resources in a specific order. Discovery resources form a parent-child hierarchy. A parent resource refuses deletion while any child still exists.

Discovery resources fall into two planes. Each plane has its own deletion experience:

Plane Resources Where to delete
Data plane Agents, workflow agents, knowledge bases Microsoft Discovery Studio
Control plane Workspaces, projects, chat model deployments, supercomputers, node pools, storage containers, storage assets, bookshelves, tools Azure portal or Azure CLI

Important

Delete data plane resources first. Control plane deletion fails while agents still reference a project, bookshelf, or tool.

Review the deletion order

  1. Agents and workflow agents (data plane, per project)
  2. Knowledge bases (data plane, per bookshelf)
  3. Storage assets, chat model deployments, node pools, bookshelves, tools
  4. Projects and storage containers
  5. Workspaces
  6. Supercomputers
  7. Supporting Azure resources: virtual network, user-assigned managed identity, container registry, storage account, Event Grid system topic

Step 3 resources don't depend on each other, so you can delete them in any order or in parallel. Steps 4 through 6 are strictly sequential.

Delete data plane resources in Discovery Studio

Agents, workflow agents, and knowledge bases aren't Azure Resource Manager resources, so they don't appear in the Azure portal. Delete them in Microsoft Discovery Studio:

  1. Open your workspace in Discovery Studio.
  2. Select the project, then delete every agent and workflow agent it contains.
  3. Open each bookshelf and delete its knowledge bases.

If you skip this step, control plane deletion fails with one of these errors:

  • ResourceHasDependentResources — Cannot delete the project while agents are present.
  • ValidationError — Knowledge base <name> (version <version>) is linked to N agent(s).

Delete control plane resources

Use either the Azure portal or the Azure CLI. The portal deletes resources one at a time from each resource blade. The following Azure CLI examples are easier to script.

Before you begin

Sign in and select the subscription that contains the resources. This approach ensures Azure issues the token for the subscription's tenant:

az login
az account set --subscription <subscription-id>

Set variables that the following commands reuse:

SUB=<subscription-id>
RG=<resource-group>
API=2026-06-01

All Microsoft Discovery commands require the --api-version parameter. Without it, the Azure CLI can't resolve the resource type and reports a generic failure.

Delete child resources

Delete storage assets, chat model deployments, and node pools before their parents:

# Storage assets
az resource delete --subscription $SUB --api-version $API \
  --ids "/subscriptions/$SUB/resourceGroups/$RG/providers/Microsoft.Discovery/storagecontainers/<storage-container>/storageassets/<asset>"

# Chat model deployments
az resource delete --subscription $SUB --api-version $API \
  --ids "/subscriptions/$SUB/resourceGroups/$RG/providers/Microsoft.Discovery/workspaces/<workspace>/chatmodeldeployments/<deployment>"

# Node pools
az resource delete --subscription $SUB --api-version $API \
  --ids "/subscriptions/$SUB/resourceGroups/$RG/providers/Microsoft.Discovery/supercomputers/<supercomputer>/nodepools/<nodepool>"

To delete every child of a given type at once, list the resources first:

IDS=$(az resource list --subscription $SUB --resource-group $RG \
  --query "[?type=='Microsoft.Discovery/storagecontainers/storageassets'].id" -o tsv)
az resource delete --subscription $SUB --api-version $API --ids $IDS

Delete bookshelves and tools

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <bookshelf> --resource-type Microsoft.Discovery/bookshelves

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <tool> --resource-type Microsoft.Discovery/tools

Delete projects and storage containers

az resource delete --subscription $SUB --api-version $API \
  --ids "/subscriptions/$SUB/resourceGroups/$RG/providers/Microsoft.Discovery/workspaces/<workspace>/projects/<project>"

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <storage-container> --resource-type Microsoft.Discovery/storagecontainers

Delete the workspace

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <workspace> --resource-type Microsoft.Discovery/workspaces

Delete the supercomputer

Delete the supercomputer only after the workspace is gone. A workspace references every attached supercomputer through its supercomputerIds property:

az resource show --subscription $SUB --resource-group $RG --api-version $API \
  --name <workspace> --resource-type Microsoft.Discovery/workspaces \
  --query properties.supercomputerIds

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <supercomputer> --resource-type Microsoft.Discovery/supercomputers

When you delete a supercomputer, Azure also removes its managed resource group (mrg-dscmp-<supercomputer>-<suffix>) and the underlying Azure Kubernetes Service (AKS) cluster.

Delete supporting resources

After you delete all Microsoft Discovery resources, delete the virtual network, user-assigned managed identity, container registry, storage account, and Event Grid system topic that the deployment used. Delete these resources last, because Discovery and AKS resources reference them.

Azure automatically removes managed on behalf of (MOBO) broker resources named mobr-* with their parent resource. Don't delete them directly.

Timing expectations

Workspace, bookshelf, and supercomputer deletions are long-running operations that can take 30 minutes or more. Add --no-wait to submit the request and poll separately.

az resource delete --subscription $SUB --resource-group $RG --api-version $API \
  --name <workspace> --resource-type Microsoft.Discovery/workspaces --no-wait

Azure Resource Manager caches resource listings, so a deleted child resource can still appear in az resource list for a few minutes. Confirm the actual state with az resource show, which returns ResourceNotFound after deletion completes.

Troubleshoot a node pool that fails to delete

If node pool deletion returns InternalServerError with the target internalMetadata, the backing AKS cluster is likely unhealthy. The cluster that backs the supercomputer shows a Failed or Deallocated state.

  1. Check the cluster in the supercomputer's managed resource group:

    az aks show --resource-group mrg-dscmp-<supercomputer>-<suffix> --name aks-dscmp-<suffix> \
  --query "{provisioningState:provisioningState, powerState:powerState.code}"

  2. Reconcile the cluster:

    az aks update --resource-group mrg-dscmp-<supercomputer>-<suffix> --name aks-dscmp-<suffix> --yes

  3. When provisioningState returns Succeeded, retry the node pool deletion.

A direct node pool reprovision doesn't clear this state. Repair the AKS cluster first.

Related content

  • Last updated on