5 Posts
0
1192
CSI Powerscale driver on infra nodes
Is there any way to run csi-powerscale driver in nodes different from workers?
I have tried labeling infra nodes as “csi-isilon.dellemc.com/XX.XX.XX.XX=csi-isilon.dellemc.com” and enabling CustomTopology, but it did not work as it keeps trying to use only worker nodes.
Flo_csI
166 Posts
0
March 17th, 2021 02:00
The driver has two main components:
* The controller that manages the volume/snapshots operation and which is a Deployment ; which you can decide to schedule on control-plane nodes with the help of that node selector : https://github.com/dell/csi-powerscale/blob/master/helm/csi-isilon/values.yaml#L80
* The node that mounts the share to the Pod and which is a DaemonSet ; that component is really useful to worker nodes that actually need to mount PV
You could always change the node selector of the isilon-node to go on control-plane nodes but I'm not sure to understand the use-case here.
HTH
Flo_csI
166 Posts
0
March 16th, 2021 10:00
Hi @fdgc,
The topology keys will ensure that, when you have a Pod with a PVC for the PowerScale storage class, it will be scheduled on a node with NFS access to the cluster.
The isilon-node component is still deployed as DaemonSet and therefore be present on all your nodes.
If you want to prevent the driver from running on particular nodes, you will have to configure a node selector manually.
You can find an example in another of our drivers right here : https://github.com/dell/csi-powermax/blob/master/helm/csi-powermax/values.yaml#L27 & https://github.com/dell/csi-powermax/blob/master/helm/csi-powermax/templates/node.yaml#L62
Today, the node selector variables are only available to configure the scheduling of the isilon-controller. Future releases will give that option for isilon-node too.
Let me know if that answers your question.
fdgc
5 Posts
0
March 17th, 2021 01:00
Hi, @Flo_csI
There is something I do not understand: you said "The isilon-node component is still deployed as DaemonSet and therefore be present on all your nodes", but I can see that the pods raised by that daemonset have nodeSelector:
node-role.kubernetes.io/worker: "", so they can only run in worker nodes.
Regards