orphanedpv

Documentation

Index

• Constants
• type Controller
  • func NewController(kubeClient kubernetes.Interface, ...) (*Controller, error)
  • func (opc *Controller) Run(ctx context.Context, workers int)
• type PVItem

Constants

const (
	ControllerName = "OrphanedPVController"
)

Types

type Controller

type Controller struct {
	// contains filtered or unexported fields
}

Controller watches all PVs actively belonging to a ScyllaDBDatacenter and replace scylla node on any PV that is orphaned, if enabled on the ScyllaDBDatacenter. Orphaned PV is a volume that is hard bound to a node that doesn't exist anymore. The controller is based on a ScyllaDBDatacenter key and listing matching PVs because using a PV key and trying to find a corresponding ScyllaDBDatacenter would be quite hard, there are no ownerRefs and we can't propagate the "enabled" information from a ScyllaDBDatacenter to a PVC annotation because PVCs are not reconciled. TODO: When we support auto-replacing nodes, we could replace it with generic controller

deleting PVs bound to nodes that don't exists anymore, without knowing about ScyllaDBDatacenter.
It would also process PVs instead of ScyllaDBDatacenter which is currently complicating the logic
that has to handle multiple PVs at once, artificial requeues / not watching PVs and different error paths.

func NewController

func NewController(
	kubeClient kubernetes.Interface,
	pvInformer corev1informers.PersistentVolumeInformer,
	pvcInformer corev1informers.PersistentVolumeClaimInformer,
	nodeInformer corev1informers.NodeInformer,
	scyllaDBDatacenterInformer scyllav1alpha1informers.ScyllaDBDatacenterInformer,
) (*Controller, error)

func (*Controller) Run

func (opc *Controller) Run(ctx context.Context, workers int)

type PVItem

type PVItem struct {
	PV          *corev1.PersistentVolume
	ServiceName string
}