Remove deployed cluster components, destroy Azure infrastructure, and clean up resources. Component cleanup preserves Azure infrastructure by default; destroy operations remove Terraform-managed resources.
[!NOTE] This guide is part of the Deploy Hub. Return there for the full deployment lifecycle.
Run component cleanup before destroying infrastructure. Follow this order to avoid dependency issues.
| Step | Action | Detail |
|---|---|---|
| 1 | Uninstall OSMO Backend | Backend operator, workflow namespaces |
| 2 | Uninstall OSMO Control Plane | OSMO service, router, web-ui |
| 3 | Uninstall AzureML Extension | ML extension, compute target, FICs |
| 4 | Uninstall GPU Infrastructure | GPU Operator, KAI Scheduler |
| 5 | Destroy VPN (if deployed) | VPN Gateway, connections |
| 6 | Destroy Main Infrastructure | All Terraform-managed Azure resources |
Cleanup scripts remove Kubernetes resources from the AKS cluster without affecting Azure infrastructure.
| Script | Removes |
|---|---|
cleanup/uninstall-osmo-backend.sh |
Backend operator, workflow namespaces |
cleanup/uninstall-osmo-control-plane.sh |
OSMO service, router, web-ui |
cleanup/uninstall-azureml-extension.sh |
ML extension, compute target, FICs |
cleanup/uninstall-robotics-charts.sh |
GPU Operator, KAI Scheduler |
Run scripts from the infrastructure/setup/cleanup/ directory:
cd infrastructure/setup/cleanup
./uninstall-osmo-backend.sh
./uninstall-osmo-control-plane.sh
./uninstall-azureml-extension.sh
./uninstall-robotics-charts.sh
Uninstall scripts preserve data by default. Use flags for complete removal.
| Script | Flag | Description |
|---|---|---|
uninstall-osmo-backend.sh |
--delete-container |
Deletes blob container with workflow artifacts |
uninstall-osmo-control-plane.sh |
--delete-mek |
Removes encryption key ConfigMap |
uninstall-osmo-control-plane.sh |
--purge-postgres |
Drops OSMO tables from PostgreSQL |
uninstall-osmo-control-plane.sh |
--purge-redis |
Flushes OSMO keys from Redis |
uninstall-robotics-charts.sh |
--delete-namespaces |
Removes gpu-operator, kai-scheduler namespaces |
uninstall-robotics-charts.sh |
--delete-crds |
Removes GPU Operator CRDs |
Full cleanup including all data:
cd infrastructure/setup/cleanup
./uninstall-osmo-backend.sh --delete-container
./uninstall-osmo-control-plane.sh --purge-postgres --purge-redis --delete-mek
./uninstall-azureml-extension.sh --force
./uninstall-robotics-charts.sh --delete-namespaces --delete-crds
Selective cleanup for specific components:
# OSMO only (preserve AzureML and GPU infrastructure)
./uninstall-osmo-backend.sh
./uninstall-osmo-control-plane.sh
# AzureML only (preserve OSMO)
./uninstall-azureml-extension.sh
After removing cluster components, destroy Azure infrastructure using one of two approaches.
Recommended approach. Preserves state files and allows clean redeployment.
cd infrastructure/terraform
# Destroy VPN first (if deployed)
cd vpn && terraform destroy -var-file=terraform.tfvars && cd ..
# Preview changes
terraform plan -destroy -var-file=terraform.tfvars
# Destroy main infrastructure
terraform destroy -var-file=terraform.tfvars
Fastest cleanup method. Removes all resources regardless of how they were created.
# Get resource group name from Terraform outputs
terraform output -raw resource_group | jq -r '.name'
# Delete resource group
az group delete --name <resource-group-name> --yes --no-wait
[!WARNING] Resource group deletion removes everything in the group, including resources not managed by Terraform. Terraform state becomes orphaned after this operation.
Terraform removes resources in dependency order. Private Endpoints, AKS clusters, and PostgreSQL servers take 5-10 minutes each. Full destruction typically takes 20-30 minutes.
Monitor remaining resources during destruction:
az resource list --resource-group <resource-group> \
--query "[].{name:name, type:type}" -o table
Azure retains certain deleted resources in a soft-deleted state. Redeployment fails when Terraform creates a resource with the same name as a soft-deleted one.
| Resource | Soft Delete | Retention Period | Blocks Redeployment |
|---|---|---|---|
| Key Vault | Mandatory | 7-90 days (configurable) | Yes |
| Azure ML Workspace | Mandatory | 14 days (fixed) | Yes |
| Container Registry | Opt-in (preview) | 1-90 days (configurable) | No (disabled by default) |
| Storage Account | Recovery only | 14 days | No (same-name creation allowed) |
Purge soft-deleted Key Vault:
az keyvault list-deleted --subscription <subscription-id> \
--resource-type vault -o table
az keyvault purge --subscription <subscription-id> \
--name <key-vault-name>
[!NOTE] Key Vaults with
purge_protection_enabled = truecannot be purged and must wait for retention expiry. This configuration defaults toshould_enable_purge_protection = false.
Purge soft-deleted Azure ML Workspace:
az ml workspace delete \
--name <workspace-name> \
--resource-group <resource-group> \
--permanently-delete
Resources manually deleted or created outside Terraform cause state mismatches.
Refresh state for resources deleted outside Terraform:
cd infrastructure/terraform
terraform refresh -var-file=terraform.tfvars
terraform plan -var-file=terraform.tfvars
Import resources created outside Terraform into state:
terraform plan -var-file=terraform.tfvars
terraform import -var-file=terraform.tfvars \
'<resource_address>' '<azure_resource_id>'
After import, run terraform plan to verify the imported resource matches configuration.
Management locks block deletion operations:
az lock list --resource-group <resource-group> -o table
az lock delete --name <lock-name> --resource-group <resource-group>
🤖 Crafted with precision by ✨Copilot following brilliant human instruction, then carefully refined by our team of discerning human reviewers.