Cluster topology awareness in airgapped bare metal

You can provide your cluster topology information to Portworx using Kubernetes node labels. To understand how Portworx uses these for volumes, refer to this page.

Region and Zone information

On-premises

You can label Kubernetes nodes with following labels to inform Portworx about region and zone of the nodes.

Label Name	Purpose
px/region	Region in which the node resides
px/zone	Zone in which the node resides

Rack information

To provide rack information to Portworx, you need to label Kubernetes nodes with px/rack=rack1, where px/rack is the key and rack1 is the value identifying the rack of which the node is a part of. Make sure the label is a string not starting with a special character or a number.

Example

Following example updates rack information for a node.

Run the following command to list the existing nodes and their labels.

kubectl get nodes --show-labels

NAME      STATUS    AGE       VERSION   LABELS
vm-1      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-1,node-role.kubernetes.io/master=
vm-2      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-2
vm-3      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-3

To indicate node vm-2 is placed on rack1 update the node label in the following way:

kubectl label nodes vm-2 px/rack=rack1

Now let's check your updated node labels.

kubectl get nodes --show-labels

NAME      STATUS    AGE       VERSION   LABELS
vm-1      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-1,node-role.kubernetes.io/master=
vm-2      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-2,px/rack=rack1
vm-3      Ready     14d       v1.7.4    beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/hostname=vm-3

This verifies that node vm-2 has the new px/rack label.

Double check if the rack information is reflected in the Portworx cluster.

pxctl cluster provision-status

NODE        NODE STATUS        POOL        POOL STATUS .....   ZONE           REGION        RACK
vm-2        Online                0        Online      .....   default        default       rack1
vm-3        Online                0        Online      .....   default        default       default

The node vm-2 which was labelled rack1 is reflected on the Portworx node while the unlabelled node vm-3 is still using the default rack info.

All the subsequent updates to the node labels will be automatically picked up by the Portworx nodes. A deletion of a px/rack label will also be reflected.

Specifying replica placement for volumes

Once the nodes are updated with rack info you can specify how the volume data can spread across your different racks.

Following is an example of a storage class that replicates its volume data across racks rack1 and rack2

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
    name: px-postgres-sc
provisioner: pxd.portworx.com
parameters:
   repl: "2"
   shared: "true"
   racks: "rack1,rack2"

Any PVC created using the above storage class will have a replication factor of 2 and will have one copy of its data on rack1 and the other copy on rack2

To do the same for regions and zones, you can use regions and zones as parameters in the StorageClass respectively.