Topology is a GA feature since Kubernetes v1.17, refer to CSI Topology Feature for more details.
In below example, there are two nodes with topology label: topology.azuredisk.csi.confidential.cloud/zone=eastus2-1
, which means these two nodes are both in zone 1.
$ kubectl get no --show-labels | grep topo
k8s-agentpool-83483713-vmss000000 Ready agent 62d v1.16.2 ...topology.azuredisk.csi.confidential.cloud/zone=eastus2-1
k8s-agentpool-83483713-vmss000001 Ready agent 62d v1.16.2 ...topology.azuredisk.csi.confidential.cloud/zone=eastus2-1
if node is in non-zone, topology label would be
topology.azuredisk.csi.confidential.cloud/zone=
Use volumeBindingMode: WaitForFirstConsumer
in storage class, pods would be scheduled on the agent node first, and then disk PVs would be created with the same zone as the agent node.
- available version: v1.5.0+
ZRS(Premium_ZRS
, StandardSSD_ZRS
) disk could be scheduled on the zone and non-zone agent node, without the restriction that disk volume should be co-located in the same zone as a given node. ZRS disk is supported only in a few regions, more details about Zone-redundant storage for managed disks.
- Use following storage class with ZRS disk support
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: managed-csi-zrs
provisioner: azuredisk.csi.confidential.cloud
parameters:
skuName: StandardSSD_ZRS
reclaimPolicy: Delete
volumeBindingMode: WaitForFirstConsumer