A StorageOS node is any machine (virtual or physical) that is running the StorageOS daemonset pod. A node must be running a daemonset pod in order to consume and/or present storage.

Nodes can be run in several modes.

Hyperconverged Mode

By default StorageOS nodes run in hyperconverged mode. This means that the node hosts data from StorageOS volumes and can present volumes to applications.

A hyperconverged node can store data from a volume and present volumes to applications regardless of whether the data for the volume consumed is placed on that node or is being served remotely. Remote volumes like this are handled by an internal protocol to present block device access to applications running on different nodes from the one to which their backing data store is attached.

StorageOS implements an extension of a Kubernetes Scheduler object that influences the placement of Pods on the same nodes as their data.

Compute-only Mode

Alternatively, a node can run in computeonly mode, which means no storage is consumed on the node itself and the node only presents volumes hosted by other nodes. Volumes presented to applications running on compute only nodes are therefore all remote. Compute only nodes can be very useful for topologies where nodes are ephemeral and should not host data, but the ephemeral nodes host applications that require StorageOS volumes. The nodes that are not intended to hold data, but just to present StorageOS volumes, can be set as computeonly.

A node can be marked as compute only at any point in time by adding the label, following the labels reference.

Storage Mode

Finally, nodes can be set to storage mode. Nodes set to storage mode don’t present data locally - instead all data is accessed through the network. This topology is enforced by tainting the relevant nodes to ensure that application workloads cannot be scheduled there.

This mode is ideal for ensuring maximum stability of data access as the node is isolated from resource drains that may occur due to applications running alongside. For redundancy purposes, in high load clusters it is ideal to have several nodes running in this mode.