Architecture

Replication design and architecture

Container Storage Modules (CSM) for Replication project consists of the following components:

• DellCSIReplicationGroup, a Kubernetes Custom Resource.
• CSM Replication controller which replicates the resources across (or within) Kubernetes clusters.
• CSM Replication sidecar container which is part of each CSI driver controller pod.
• repctl - Multi cluster Kubernetes client for managing replication related objects.

DellCSIReplicationGroup

DellCSIReplicationGroup (RG) is a cluster scoped Custom Resource that represents a protection group on the backend storage array. It is used to group volumes with the same replication related properties together. DellCSIReplicationGroup’s spec contains an action field which can be used to perform replication related operations on the backing protection groups on the storage arrays. This includes operations like Failover, Reprotect, Suspend, Synchronize, etc. Any replication related operation is always carried out on all the volumes present in the group.

Specification

kind: DellCSIReplicationGroup
apiVersion: replication.storage.dell.com/v1
metadata:
  name: rg-e6be24c0-145d-4b62-8674-639282ebdd13
spec:
  driverName: driver.dellemc.com # Name of the CSI driver (same as provisioner name in StorageClass)
  action: "" # Name of the replication action to be performed on the protection group
  protectionGroupAttributes:
    localAttributeKey: value
  protectionGroupId: protection-group-id # Identifier of the backing protection group on the Storage Array
  remoteClusterId: tgtClusterID # A unique identifier for the remote Kubernetes Cluster
  remoteProtectionGroupAttributes:
    remoteAttributeKey: value
  remoteProtectionGroupId: csi-rep-sg-test-5-ASYNC # Identifier for the protection group on the remote Storage Array

Status

The status sub resource of DellCSIReplicationGroup contains information about the state of replication & any actions which have been performed on the object.

status:
    conditions:
    - condition: Action REPROTECT_REMOTE succeeded
      time: "2021-08-11T12:22:05Z"
    - condition: Replication Link State:IsSource changed from (true) to (false)
      time: "2021-08-11T12:18:50Z"
    - condition: Action FAILOVER_REMOTE succeeded
      time: "2021-08-11T12:18:50Z"
    lastAction:
      condition: Action REPROTECT_REMOTE succeeded
      time: "2021-08-11T12:22:05Z"
    replicationLinkState:
      isSource: false
      lastSuccessfulUpdate: "2021-08-11T17:18:12Z"
      state: Synchronized
    state: Ready

Here is a diagram representing how the state of the CustomResource changes based on actions:

CSM Replication Sidecar

CSM Replication sidecar is deployed as sidecar container in each CSI driver’s controller pod. This container is similar to Kubernetes CSI Sidecar containers and runs a Controller Manager which manages the following controllers:

• PersistentVolume(PV) Controller
• PersistentVolumeClaim(PVC) Controller
• DellCSIReplicationGroup(RG) Controller

The PV & PVC controllers watch for PV/PVC creation events and use dell-csi-extensions APIs to communicate with the CSI Driver controller plugin to discover/create replication enabled volumes and protection groups on the backend storage array. The PersistentVolume controller then uses these details to create DellCSIReplicationGroup objects in the cluster. These controllers are also responsible for associating the PV & PVC objects with DellCSIReplicationGroup objects. This association is established by applying annotations & labels on the PV & PVC objects.

The RG controller manages DellCSIReplicationGroup instances and processes any change requests. It is primarily responsible for the following:

• Perform actions on the protection groups
• Monitor status of replication
• Updates to the status sub resource

CSM Replication Controller

CSM Replication Controller is a Kubernetes application deployed independently of CSI drivers and is responsible for the communication between Kubernetes clusters.

The details about the clusters it needs to connect to are provided in the form of a ConfigMap with references to secrets containing the details(KubeConfig/ServiceAccount tokens) required to connect to the respective clusters.

It consists of Controller Manager which manages the following controllers:

• PersistentVolume(PV) Controller
• PersistentVolumeClaim(PVC) Controller
• DellCSIReplicationGroup(RG) Controller

The PV controller is responsible for creating PV objects (representing the replicated volumes on the backend storage array) in the remote Kubernetes cluster. This controller also enables deletion of the remote PV object, if enabled through the storage class’ RemotePVRetentionPolicy, by propagating the deletion request across clusters.

Similarly, the RG controller is responsible for creating RG objects in the remote Kubernetes cluster. These RG objects represent the remote protection groups on the backend storage array. This controller can also propagate the deletion request of RG objects across clusters, if enabled through the storage class’ RemoteRGRetentionPolicy.

Both the PV & RG objects in the remote cluster have extra metadata associated with them in form of annotations & labels. This metadata includes information about the respective objects in the source cluster.

The PVC objects are never replicated across the clusters. Instead, the remote PV objects have annotations related to the source PVC objects. This information can be easily used to create the PVCs whenever required using repctl or kubectl.

Supported Cluster Topologies

Click here for details for the various types of supported cluster topologies