custom

Documentation

Index

• type Builder
  • func NewControllerManagedBy(ctx context.Context, mgr manager.Manager) *Builder
  • func (b *Builder) Complete(reconciler Reconciler) (healthz.Checker, error)
  • func (b *Builder) Named(name string) *Builder
  • func (b *Builder) Options(options Options) *Builder
  • func (b *Builder) SetDefaults()
  • func (b *Builder) UsingConditions(conditions condition.Conditions) *Builder
  • func (b *Builder) UsingConverter(converter Converter) *Builder
  • func (b *Builder) UsingDataStore(dataStore cache.Indexer) *Builder
  • func (b *Builder) WithClientSet(clientSet *kubernetes.Clientset) *Builder
  • func (b *Builder) WithLogger(logger logr.Logger) *Builder
• type Converter
• type CustomController
  • func NewCustomController(log logr.Logger, options Options, config *cache.Config, reconciler Reconciler, ...) *CustomController
  • func (c *CustomController) CustomCheck() healthz.Checker
  • func (c *CustomController) Start(ctx context.Context) error
  • func (c *CustomController) WaitForCacheSync(controller cache.Controller)
• type Options
• type Reconciler
• type Request

Types

type Builder

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

func NewControllerManagedBy

func NewControllerManagedBy(ctx context.Context, mgr manager.Manager) *Builder

func (*Builder) Complete

func (b *Builder) Complete(reconciler Reconciler) (healthz.Checker, error)

Complete adds the controller to manager's Runnable. The Controller runnable will start when the manager starts

func (*Builder) Named

func (b *Builder) Named(name string) *Builder

func (*Builder) Options

func (b *Builder) Options(options Options) *Builder

func (*Builder) SetDefaults

func (b *Builder) SetDefaults()

SetDefaults sets the default options for controller

func (*Builder) UsingConditions

func (b *Builder) UsingConditions(conditions condition.Conditions) *Builder

func (*Builder) UsingConverter

func (b *Builder) UsingConverter(converter Converter) *Builder

func (*Builder) UsingDataStore

func (b *Builder) UsingDataStore(dataStore cache.Indexer) *Builder

func (*Builder) WithClientSet

func (b *Builder) WithClientSet(clientSet *kubernetes.Clientset) *Builder

func (*Builder) WithLogger

func (b *Builder) WithLogger(logger logr.Logger) *Builder

type Converter

type Converter interface {
	// ConvertObject takes an object and returns the modified object which will be
	// stored in the data store
	ConvertObject(originalObj interface{}) (convertedObj interface{}, err error)
	// ConvertList takes an object and returns the modified list of objects which
	// will be returned to the Simple Pager function to aggregate the list pagination
	// response
	ConvertList(originalList interface{}) (convertedList interface{}, err error)
	// Resource returns the K8s resource name to list/watch
	Resource() string
	// ResourceType returns the k8s object to list/watch
	ResourceType() runtime.Object
	// Indexer returns the key for indexing custom converted object
	Indexer(obj interface{}) (string, error)
}

Converter for converting k8s object and object list used in watches and list operation to the desired format.

type CustomController

type CustomController struct {

	// Reconciler will be called on all the K8s object events
	Do Reconciler
	// contains filtered or unexported fields
}

This Controller can be used for any type of K8s object, but is used for Pod Objects in this repository. There are two reasons why we are using a wrapper over the low level controllers instead of using controllers from controller-runtime.

1. We don't want to cache the entire Pod Object because of Memory constraints. We need specific details from metadata and Pod Spec. To do this we intercept the request at List; and watch, optimize it before it's stored in cache. Long term plan is to use MetaData only cache or disable Pod caching altogether
2. We want the Deleted Object when Pod is Terminating. Pod Networking should only be deleted once the Pod has deleted or all containers have exited. Long term plan is to consider migrating to using finalizers and delete only when all containers have exited.

In future, we may be able to switch to upstream controller for reconciling Pods if the long term solutions are in place

func NewCustomController

func NewCustomController(
	log logr.Logger,
	options Options,
	config *cache.Config,
	reconciler Reconciler,
	workQueue workqueue.RateLimitingInterface,
	conditions condition.Conditions) *CustomController

func (*CustomController) CustomCheck

func (c *CustomController) CustomCheck() healthz.Checker

func (*CustomController) Start

func (c *CustomController) Start(ctx context.Context) error

Starts the low level controller

func (*CustomController) WaitForCacheSync

func (c *CustomController) WaitForCacheSync(controller cache.Controller)

WaitForCacheSync tills the cache has synced, this must be done under mutex lock to prevent other controllers from starting at same time

type Options

type Options struct {
	// Name of the controller used for creating named work queues
	Name string
	// PageLimit is the number of objects returned per page on a list operation
	PageLimit int
	// Namespace to list and watch for
	Namespace string
	// ResyncPeriod how often to sync using list with the API Server
	ResyncPeriod time.Duration
	// MaxConcurrentReconciles to parallelize processing of worker queue
	MaxConcurrentReconciles int
}

Options contains the configurable parameters of the Custom Controller

type Reconciler

type Reconciler interface {
	Reconcile(request Request) (ctrl.Result, error)
}

type Request

type Request struct {
	// Add/Update Request will contain the Namespaced Name only. The
	// item can be retrieved from the indexer for add/update events
	NamespacedName types.NamespacedName
	// Delete Event will contain the DeletedObject only.
	DeletedObject interface{}
}

Request for Add/Update only contains the Namespace/Name Request for Delete contains the Pod Object as by the time Delete Request is reconciled the cache will not have it