EKS Anywhere, Part-1 Dell EMC PowerScale CSI

 

This article is part of the EKS Anywhere series EKS Anywhere., extending the Hybrid cloud momentum

The article is a deep-dive and hence split into 2 parts. This is the Part-1 of the PowerScale CSI and EKS Anywhere article. In this article, we are going to explore the implementation, integration and testing of Dell PowerScale CSI with EKS Anywhere.

PowerScale is Dell Technologies award winning1 and a revolutionary scale-out NAS solution. It is particularly designed with the extreme demands of AI/ML, Big Data, NAS use cases, etc. As customers begin to design or transition their critical file based stateful workloads to Kubernetes, the persistence layer offered by PowerScale CSI becomes a quintessential enabler for these use-cases.

As an introductory context, one can observe DellEMC’s CSI coverage CSI Drivers | Dell Technologies. In addition to the CSI implementations, DellEMC has also innovated and instrumented CSM (Container Storage Modules), which further heighten the range of experiences by adding authentication, authorization, observability, replication, etc.

Our use-case:

We will be leveraging an NFS based ReadWriteMany implementation pattern, where the persistence layer for our stateful workload is implemented over the Dell EMC PowerScale CSI.

The below visual represents a high-level summary of the same. The use case capabilities can be further extrapolated to suit file-sharing workloads such as CI-CD systems, Advanced Analytics, AI/ML, etc. that require high performance file sharing solutions on top of Kubernetes stack.

Goals

1. Implement PowerScale CSI drivers on EKS Anywhere cluster
2. Implement snapshotting capabilities via external-snapshotter
3. Deploy file sharing workload with RWX persistence layer delivered over PowerScale CSI
4. Test various use-cases around persistence, volume expansion, snapshotting (backup & restore)

Prerequisites

1. EKS Administrative machine is setup as per the earlier article
2. One can do this on a new or an existing cluster either standalone or a dedicated management cluster mode
3. A PowerScale cluster or OneFS simulator configured and ready to be used for CSI testing

Let's begin

Assuming, I am starting with a fresh standalone workload cluster (you can have an existing one or create a new one). My EKS Anywhere cluster’s name will be eksa1 and the static IP for the API server will be 172.24.165.11

SSH into the EKS Anywhere administration and run the cluster creation script as shown below to create the standalone workload cluster

 

cd $HOMEsource create-eksa-cluster.sh#######################IMPORTANT NOTE#################################keep the name of workload and management cluster EXACTLY THE SAMEin case of creating standlone workload clusters or management clusters######################################################################Workload cluster name: eksa1Management cluster name: eksa1staticIp for API server High Availability: 172.24.165.11Kubernetes version 1.21, 1.22, 1.23, etc.: 1.23

 

EKS Anywhere Cluster creation log

Next switch the kubectl context to work with the standalone EKS Anywhere

 

cd $HOMEsource eks-anywhere/cluster-ops/switch-cluster.shclusterName: eksa1eksa1kubectl config get-contextsCURRENT   NAME                CLUSTER   AUTHINFO      NAMESPACE*         eksa1-admin@eksa1   eksa1     eksa1-admin

 

As one can see from the below kubectl output for my EKS Anywhere workload cluster, AWS EKS Anywhere ships with the standard storage class which is mapped to VMware’s CNS CSI. This default CSI is also covered in the other article 

 

kubectl get storageclassNAME                 PROVISIONER              RECLAIMPOLICY   VOLUMEBINDINGMODE   ALLOWVOLUMEEXPANSION   AGEstandard (default)   csi.vsphere.vmware.com   Delete          Immediate           false                  17m

 

A handy script has been created that automates the entire process. This includes:

• Deployment of PowerScale CSI driver on EKS Anywhere
• Deployment of External snapshotter
• Creation of Storage class for use with PowerScale CSI driver

You will need to gather the following details for your PowerScale cluster before proceeding, IP/FQDN, username, password, PowerScale cluster name

SSH into the EKS Administrative machine and execute the below script

 

cd $HOMEsource eks-anywhere/powerscale/install-powerscale-csi-driver.shEnter EKSA Cluster Name on which CSI driver needs to be installedclusterName: eksa1Enter PowerScale CSI release version, e.g. 2.2.0, 2.3.0, 2.4.0, 2.5.0csiReleaseNumber: 2.5.0Enter PowerScale Cluster NamepowerScaleClusterName: powerscale01Enter IP or FQDN of the PowerScale ClusteripOrFqdnOfPowerStoreArray: 172.24.165.41Enter username of the PowerScale ClusteruserNameOfPowerScaleCluster: adminEnter password of the PowerScale ClusterpasswordOfPowerScaleCluster:

 

A snip of the script installation log is shown below.

As seen from the below logs, CSI driver along with external snapshotter and storage class has been created on the EKS Anywhere cluster. As a part of the CSI driver installation the below pods are created in the csi-powerscale namespace

• ReplicaSet named isilon-controller (2 pods with 5 containers each namely attacher, provisioner, snapshotter, resizer, driver)
• DaemonSet named isilon-node (2 pods with 2 containers each namely driver, registrar)

The setup is completed and is ready to deploy NFS based workloads via PowerScale CSI ReadWriteMany mode

Additional observations

One can perform additional observations for the various pods/containers and logs for the CSI artefacts by running the below commands.

 

pscalecontrollerpod1=$(kubectl get pods --selector=app=isilon-controller -n csi-powerscale -o=jsonpath='{.items[0].metadata.name}')pscalecontrollerpod2=$(kubectl get pods --selector=app=isilon-controller -n csi-powerscale -o=jsonpath='{.items[1].metadata.name}')You can also run the below commands using -c and changing the container name to any of the below mentioned 5 containers>>> attacher, provisioner, snapshotter, resizer, driverkubectl describe pod $pscalecontrollerpod1 -n csi-powerscalekubectl describe pod $pscalecontrollerpod2 -n csi-powerscalekubectl logs $pscalecontrollerpod1 -n csi-powerscale -c driverkubectl logs $pscalecontrollerpod2 -n csi-powerscale -c driver# powerscale NODE PODSpscalenodepod1=$(kubectl get pods --selector=app=isilon-node -n csi-powerscale -o=jsonpath='{.items[0].metadata.name}')pscalenodepod2=$(kubectl get pods --selector=app=isilon-node -n csi-powerscale -o=jsonpath='{.items[1].metadata.name}')You can also run the below commands using -c and changing the container name to any of the below mentioned 5 containers>>> driver, registrarkubectl describe pod $pscalenodepod1 -n csi-powerscalekubectl describe pod $pscalenodepod1 -n csi-powerscalekubectl logs $pscalenodepod1 -n csi-powerscale -c driverkubectl logs $pscalenodepod2 -n csi-powerscale -c driver

 

A detailed output for the above commands is captured in the below URL for greater comprehension and understanding

eksa-powerscale-csi-outputs.md (github.com)

OK, we are all done with implementing the PowerScale CSI on our EKS Anywhere cluster. With this step completed you can move over to part-2 where we will deploy and test use-cases leveraging the PowerScale CSI.

cheers,

Ambar Hassani

#iwork4dell