Skip to main content
You want to tidy up a Stack - rename a manifest, split one into two, or drop one you no longer need - and the manifest owns a namespace or a PersistentVolumeClaim. Done carelessly, that change deletes the volume and the data on it. This guide shows how to check what is at risk and make the change safely.

​
Why a Stack Change Can Delete Data

Ankra identifies a manifest by its name. When a Stack is applied - with ankra cluster apply, a GitOps sync, or an edit in the portal - and a manifest name is no longer there, Ankra deletes every object in that manifest’s stored YAML. A rename is the same thing: the old name is deleted and the new name is created as a new manifest, with nothing ordering the two.
  • A manifest that holds a Namespace takes the namespace with it, and every PersistentVolumeClaim inside it.
  • A manifest that holds PersistentVolumeClaims deletes those claims.
  • What happens to the disk depends on the PersistentVolume’s reclaim policy. With Delete, the default for most CSI drivers, the disk is destroyed with the claim. With Retain, the PersistentVolume and its data stay behind in the Released state.
The field-level rules are in the ImportCluster reference.

​
Prerequisites

  • Access to the cluster in Ankra, and permission to edit the Stack.
  • The Ankra CLI, logged in, if you use the commands below. Everything can also be done in the portal.

​
Steps

1

Find the volumes the change touches

List the claims in the namespaces the manifest owns or deploys into. In the portal, open the cluster and go to Kubernetes → Storage → Persistent Volume Claims. From the CLI:
Note the volume each claim is bound to. A claim created by a StatefulSet’s volumeClaimTemplates is not in any manifest, but it is still deleted when the namespace around it is.
2

Check each volume's reclaim policy

Go to Kubernetes → Storage → Persistent Volumes and read the Reclaim Policy column for those volumes, or list them:
Any volume with Delete loses its data if its claim is deleted.
3

Switch the volumes you need to Retain

A volume’s reclaim policy can be changed in place, with no downtime. From the CLI:
Or open the volume, go to its Manifest tab, click Edit, set persistentVolumeReclaimPolicy: Retain, and Apply.For volumes created later, a StorageClass’s reclaimPolicy cannot be changed after it is created. Add a second StorageClass with reclaimPolicy: Retain and use it for stateful workloads.
4

Take a copy if the data matters

Retain keeps the disk, but it does not give you a copy. If your organisation has Backups enabled (closed beta), take a restore point of the Stack first. Otherwise use your database’s or storage provider’s own backup.
5

Make the Stack change

Rename or remove the manifest. Where you can, avoid renaming a manifest that owns a namespace or a claim at all: keep its name fixed and change its contents instead. If you rename a parent, update every parents entry that names it in the same change.

​
Verify

After the Stack has synced, check Kubernetes → Storage → Persistent Volumes again. Volumes you switched to Retain whose claims were deleted show the status Released, with their data intact. To use a Released volume again, clear its spec.claimRef and create a claim that names it with spec.volumeName. A released volume keeps costing money until you reuse or delete it; Cloud Cost lists released persistent volumes among its savings findings.

​
Changing the Cluster’s GitOps Repository

Pointing a cluster at a different GitOps repository or branch carries the same risk on a larger scale. Ankra writes the cluster’s current state to the new source first, but from then on that source is authoritative: anything that later leaves it is removed, including workloads that own claims. ankra cluster apply refuses a repoint unless you pass --allow-repoint, and on a cluster that holds PersistentVolumeClaims it also needs --allow-repoint-destroying-data. Only pass the second flag after you have done the steps above. See spec.git_repository in the ImportCluster reference.

​
Troubleshooting

  • The claim is gone and the volume was Delete. The disk has been removed by the storage provider. Restore from a backup or restore point; Ankra cannot bring it back.
  • The recreated claim is empty. A new claim with the old name is bound to a new, empty volume by the StorageClass, not to the retained one. To get the data back, clear the retained volume’s spec.claimRef and create the claim with spec.volumeName set to that volume.

​
Next Steps

Before your next Stack change, switch the volumes of every stateful workload to Retain now, so a later rename cannot take data with it. For help with a volume you have already lost, contact support.