Architecture

Volume Manager is implemented according to the CSI specification. It comes with the below components run as Pods in Kubernetes:

• Controller
• Node server

When Volume Manager contains legacy volumes from DirectCSI, the following additional components also run as Pods:

• Legacy controller
• Legacy node server

The Controller runs as Deployment Pods named controller. The Volume Manager controller has three replicas. The replicas elect one instance to serve requests.

Each pod has the following containers:

• Controller

  Honors CSI requests to create, delete and expand volumes.

• CSI provisioner

  Bridges volume creation and deletion requests from a Persistent Volume Claim to the CSI controller.

• CSI resizer

  Bridges volume expansion requests from Persistent Volume Claim to CSI controller.

The controller server runs as a container named controller in a controller Deployment Pod.

It handles the following requests:

• Create volume

  The controller server creates a new Volume ManagerVolume CRD after reversing requested storage space on a suitable Volume ManagerDrive CRD. For more information, refer to the Volume scheduling guide.

• Delete volume

  The controller server deletes an existing Volume ManagerVolume CRD for unbound volumes after releasing previously reserved space in a Volume ManagerDrive CRD.

• Expand volume

  The controller server expands an existing Volume ManagerVolume CRD after reversing requested storage space in a Volume ManagerDrive CRD.

The Legacy controller runs as Deployment Pods named legacy-controller. The Legacy controller has three replicas located in any Kubernetes nodes. The replicas elect one instance to serve requests.

Each pod has the following containers:

• CSI provisioner

  Bridges legacy volume creation and deletion requests from Persistent Volume Claim to the CSI controller.

• Controller

  Honors CSI requests to delete and expand volumes only. Create volume requests are prohibited. The legacy controller works only for legacy volumes previously created in DirectCSI.

• CSI resizer

  Bridges legacy volume expansion requests from Persistent Volume Claim to the CSI controller.

The Legacy controller server runs as a container controller in a legacy-controller Deployment Pod. It handles the following requests:

• Create volume

  The Controller server returns an error for this request. Volume Manager does not create new legacy DirectCSI volumes.

• Delete volume

  The Controller server deletes the Volume ManagerVolume CRD for unbound volumes after releasing previously reserved space in the Volume ManagerDrive CRD.

• Expand volume

  The Controller server expands the Volume ManagerVolume CRD after reversing requested storage space in the Volume ManagerDrive CRD.

The node server runs as DaemonSet Pods named node-server in all or selected Kubernetes nodes. Each node server Pod runs on a node independently.

Each pod has the following containers:

• Node driver registrar

  Registers node server to kubelet to get CSI RPC calls.

• Node server

  Honors stage, unstage, publish, unpublish and expand volume RPC requests.

• Node controller

  Honors CRD events from Volume ManagerDrive, Volume ManagerVolume, Volume ManagerNode and Volume ManagerInitRequest.

• Liveness probe

  Exposes a /healthz endpoint to check node server liveness by Kubernetes.

The legacy node server runs as DaemonSet Pods named legacy-node-server in all or selected Kubernetes nodes. Each legacy node server Pod runs on a node independently.

Each pod contains the below running containers:

• Node driver registrar

  Registers legacy node server to kubelet to get CSI RPC calls.

• Node server

  Honors stage, unstage, publish, unpublish and expand volume RPC requests.

• Liveness probe

  Exposes /healthz endpoint to check legacy node server liveness by Kubernetes.