updatestrategy

package
v0.0.0-...-2828fdb Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Mar 3, 2025 License: MIT Imports: 34 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	KarpenterEC2NodeClassResource = "ec2nodeclasses.karpenter.k8s.aws"
)

Variables

This section is empty.

Functions

func InstanceConfigUpToDate 

func InstanceConfigUpToDate(instanceConfig, poolConfig *InstanceConfig) bool

InstanceConfigUpToDate compares current and desired InstanceConfig. It compares userdata, imageID and checks if the current config has all the desired tags. It does NOT check if the current config has too many EC2 tags as many tags are injected out of our control. This means removing a tag is not enough to make the configs unequal.

Types

type ASGNodePoolsBackend 

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

ASGNodePoolsBackend defines a node pool backed by an AWS Auto Scaling Group.

func NewASGNodePoolsBackend 

func NewASGNodePoolsBackend(cluster *api.Cluster, sess *session.Session) *ASGNodePoolsBackend

NewASGNodePoolsBackend initializes a new ASGNodePoolsBackend for the given cluster and AWS session and.

func (*ASGNodePoolsBackend) Get 

func (n *ASGNodePoolsBackend) Get(_ context.Context, nodePool *api.NodePool) (*NodePool, error)

Get gets the ASG matching to the node pool and gets all instances from the ASG. The node generation is set to 'current' for nodes with the latest launch configuration and 'outdated' for nodes with an older launch configuration.

func (*ASGNodePoolsBackend) MarkForDecommission 

func (n *ASGNodePoolsBackend) MarkForDecommission(_ context.Context, nodePool *api.NodePool) error

MarkForDecommission suspends autoscaling of the node pool if it was enabled and makes sure that the pool can be scaled down to 0. The implementation assumes the kubernetes cluster-autoscaler is used so it just removes a tag.

func (*ASGNodePoolsBackend) Scale 

func (n *ASGNodePoolsBackend) Scale(_ context.Context, nodePool *api.NodePool, replicas int) error

Scale sets the desired capacity of the ASGs to the number of replicas. If the node pool is backed by multiple ASGs the scale operation will try to balance the increment/decrement of nodes over all the ASGs.

func (*ASGNodePoolsBackend) Terminate 

func (n *ASGNodePoolsBackend) Terminate(_ context.Context, _ *api.NodePool, node *Node, decrementDesired bool) error

Terminate terminates an instance from the ASG and optionally decrements the DesiredCapacity. By default the desired capacity will not be decremented. In case the new desired capacity is less then the current min size of the ASG, it will also decrease the ASG minSize. This function will not return until the instance has been terminated in AWS.

type CLCUpdateStrategy 

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

func NewCLCUpdateStrategy 

func NewCLCUpdateStrategy(logger *log.Entry, nodePoolManager NodePoolManager, pollingInterval time.Duration) *CLCUpdateStrategy

NewCLCUpdateStrategy initializes a new CLCUpdateStrategy.

func (*CLCUpdateStrategy) PrepareForRemoval 

func (c *CLCUpdateStrategy) PrepareForRemoval(ctx context.Context, nodePoolDesc *api.NodePool) error

func (*CLCUpdateStrategy) Update 

func (c *CLCUpdateStrategy) Update(ctx context.Context, nodePoolDesc *api.NodePool) error

type DrainConfig 

type DrainConfig struct {
	// Start forcefully evicting pods <ForceEvictionGracePeriod> after node drain started
	ForceEvictionGracePeriod time.Duration

	// Only force evict pods that are at least <MinPodLifetime> old
	MinPodLifetime time.Duration

	// Wait until all healthy pods in the same PDB are at least <MinHealthyPDBSiblingCreationTime> old
	MinHealthyPDBSiblingLifetime time.Duration

	// Wait until all unhealthy pods in the same PDB are at least <MinUnhealthyPDBSiblingCreationTime> old
	MinUnhealthyPDBSiblingLifetime time.Duration

	// Wait at least <ForceEvictionInterval> between force evictions to allow controllers to catch up
	ForceEvictionInterval time.Duration

	// Wait for <PollInterval> between force eviction attempts
	PollInterval time.Duration
}

DrainConfig contains the various settings for the smart node draining algorithm

type EC2NodePoolBackend 

type EC2NodePoolBackend struct {
	// contains filtered or unexported fields
}

EC2NodePoolBackend defines a node pool consisting of EC2 instances managed externally by some component e.g. Karpenter.

func NewEC2NodePoolBackend 

func NewEC2NodePoolBackend(cluster *api.Cluster, sess *session.Session, crdResolverInitializer func() (*KarpenterCRDNameResolver, error)) *EC2NodePoolBackend

NewEC2NodePoolBackend initializes a new EC2NodePoolBackend for the given clusterID and AWS session and.

func (*EC2NodePoolBackend) DecommissionKarpenterNodes 

func (n *EC2NodePoolBackend) DecommissionKarpenterNodes(ctx context.Context) error

func (*EC2NodePoolBackend) DecommissionNodePool 

func (n *EC2NodePoolBackend) DecommissionNodePool(ctx context.Context, nodePool *api.NodePool) error

func (*EC2NodePoolBackend) Get 

func (n *EC2NodePoolBackend) Get(ctx context.Context, nodePool *api.NodePool) (*NodePool, error)

Get gets the EC2 instances matching to the node pool by looking at node pool tag. The node generation is set to 'current' for nodes with up-to-date userData,ImageID and tags and 'outdated' for nodes with an outdated configuration.

func (*EC2NodePoolBackend) MarkForDecommission 

func (n *EC2NodePoolBackend) MarkForDecommission(context.Context, *api.NodePool) error

func (*EC2NodePoolBackend) Scale 

func (n *EC2NodePoolBackend) Scale(context.Context, *api.NodePool, int) error

func (*EC2NodePoolBackend) Terminate 

func (n *EC2NodePoolBackend) Terminate(ctx context.Context, pool *api.NodePool, node *Node, _ bool) error

type InstanceConfig 

type InstanceConfig struct {
	UserData string
	ImageID  string
	Tags     map[string]string
}

type KarpenterCRDNameResolver 

type KarpenterCRDNameResolver struct {
	NodePoolCRDName string
	// contains filtered or unexported fields
}

func NewKarpenterCRDResolver 

func NewKarpenterCRDResolver(ctx context.Context, k8sClients *kubernetes.ClientsCollection) (*KarpenterCRDNameResolver, error)

func (*KarpenterCRDNameResolver) NodePoolConfigGetter 

func (r *KarpenterCRDNameResolver) NodePoolConfigGetter(ctx context.Context, nodePool *api.NodePool) (*InstanceConfig, error)

func (*KarpenterCRDNameResolver) NodeTemplateCRDName 

func (r *KarpenterCRDNameResolver) NodeTemplateCRDName() string

type KubernetesNodePoolManager 

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

KubernetesNodePoolManager defines a node pool manager which uses the Kubernetes API along with a node pool provider backend to manage node pools.

func NewKubernetesNodePoolManager 

func NewKubernetesNodePoolManager(logger *log.Entry, kubeClient kubernetes.Interface, poolBackend ProviderNodePoolsBackend, drainConfig *DrainConfig, noScheduleTaint bool) *KubernetesNodePoolManager

NewKubernetesNodePoolManager initializes a new Kubernetes NodePool manager which can manage single node pools based on the nodes registered in the Kubernetes API and the related NodePoolBackend for those nodes e.g. ASGNodePool.

func (*KubernetesNodePoolManager) AbortNodeDecommissioning 

func (m *KubernetesNodePoolManager) AbortNodeDecommissioning(ctx context.Context, node *Node) error

func (*KubernetesNodePoolManager) CordonNode 

func (m *KubernetesNodePoolManager) CordonNode(ctx context.Context, node *Node) error

CordonNode marks a node unschedulable.

func (*KubernetesNodePoolManager) DisableReplacementNodeProvisioning 

func (m *KubernetesNodePoolManager) DisableReplacementNodeProvisioning(ctx context.Context, node *Node) error

func (*KubernetesNodePoolManager) GetPool 

func (m *KubernetesNodePoolManager) GetPool(ctx context.Context, nodePoolDesc *api.NodePool) (*NodePool, error)

GetPool gets the current node Pool from the node pool backend and attaches the Kubernetes node object name and labels to the corresponding nodes.

func (*KubernetesNodePoolManager) MarkNodeForDecommission 

func (m *KubernetesNodePoolManager) MarkNodeForDecommission(ctx context.Context, node *Node) error

func (*KubernetesNodePoolManager) MarkPoolForDecommission 

func (m *KubernetesNodePoolManager) MarkPoolForDecommission(ctx context.Context, nodePool *api.NodePool) error

func (*KubernetesNodePoolManager) ScalePool 

func (m *KubernetesNodePoolManager) ScalePool(ctx context.Context, nodePool *api.NodePool, replicas int) error

ScalePool scales a nodePool to the specified number of replicas. On scale down it will attempt to do it gracefully by draining the nodes before terminating them.

func (*KubernetesNodePoolManager) TerminateNode 

func (m *KubernetesNodePoolManager) TerminateNode(ctx context.Context, nodePool *api.NodePool, node *Node, decrementDesired bool) error

TerminateNode terminates a node and optionally decrement the desired size of the node pool. Before a node is terminated it's drained to ensure that pods running on the nodes are gracefully terminated.

type Node 

type Node struct {
	Name            string
	Annotations     map[string]string
	Labels          map[string]string
	Taints          []v1.Taint
	Cordoned        bool
	ProviderID      string
	FailureDomain   string
	Generation      int
	VolumesAttached bool
	Ready           bool
	Master          bool
}

Node is an abstract node object which combines the node information from the node pool backend along with the corresponding Kubernetes node object.

type NodePool 

type NodePool struct {
	Min        int
	Desired    int
	Current    int
	Max        int
	Generation int
	Nodes      []*Node
}

NodePool defines a node pool including all nodes.

func WaitForDesiredNodes 

func WaitForDesiredNodes(ctx context.Context, logger *log.Entry, n NodePoolManager, nodePoolDesc *api.NodePool) (*NodePool, error)

WaitForDesiredNodes waits for the current number of nodes to match the desired number. The final node pool will be returned.

func (*NodePool) ReadyNodes 

func (n *NodePool) ReadyNodes() []*Node

ReadyNodes returns a list of nodes which are marked as ready.

type NodePoolManager 

type NodePoolManager interface {
	GetPool(ctx context.Context, nodePool *api.NodePool) (*NodePool, error)
	MarkNodeForDecommission(ctx context.Context, node *Node) error
	AbortNodeDecommissioning(ctx context.Context, node *Node) error
	ScalePool(ctx context.Context, nodePool *api.NodePool, replicas int) error
	TerminateNode(ctx context.Context, nodePool *api.NodePool, node *Node, decrementDesired bool) error
	MarkPoolForDecommission(ctx context.Context, nodePool *api.NodePool) error
	DisableReplacementNodeProvisioning(ctx context.Context, node *Node) error
	CordonNode(ctx context.Context, node *Node) error
}

NodePoolManager defines an interface for managing node pools when performing update operations.

type ProfileNodePoolProvisioner 

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

ProfileNodePoolProvisioner is a NodePoolProvisioner which selects the backend provisioner based on the node pool profile. It has a default provisioner and a mapping of profile to provisioner for those profiles which can't use the default provisioner.

func NewProfileNodePoolsBackend 

func NewProfileNodePoolsBackend(defaultProvisioner ProviderNodePoolsBackend, profileMapping map[string]ProviderNodePoolsBackend) *ProfileNodePoolProvisioner

NewProfileNodePoolsBackend initializes a new ProfileNodePoolProvisioner.

func (*ProfileNodePoolProvisioner) Get 

func (n *ProfileNodePoolProvisioner) Get(ctx context.Context, nodePool *api.NodePool) (*NodePool, error)

Get the specified node pool using the right node pool provisioner for the profile.

func (*ProfileNodePoolProvisioner) MarkForDecommission 

func (n *ProfileNodePoolProvisioner) MarkForDecommission(ctx context.Context, nodePool *api.NodePool) error

MarkForDecommission marks a node pool for decommissioning using the right node pool provisioner for the profile.

func (*ProfileNodePoolProvisioner) Scale 

func (n *ProfileNodePoolProvisioner) Scale(ctx context.Context, nodePool *api.NodePool, replicas int) error

Scale scales a node pool using the right node pool provisioner for the profile.

func (*ProfileNodePoolProvisioner) Terminate 

func (n *ProfileNodePoolProvisioner) Terminate(ctx context.Context, nodePool *api.NodePool, node *Node, decrementDesired bool) error

Terminate terminates a node pool using the default provisioner.

type ProviderNodePoolsBackend 

type ProviderNodePoolsBackend interface {
	Get(ctx context.Context, nodePool *api.NodePool) (*NodePool, error)
	Scale(ctx context.Context, nodePool *api.NodePool, replicas int) error
	MarkForDecommission(ctx context.Context, nodePool *api.NodePool) error
	Terminate(ctx context.Context, nodePool *api.NodePool, node *Node, decrementDesired bool) error
}

ProviderNodePoolsBackend is an interface for describing a node pools provider backend e.g. AWS Auto Scaling Groups.

type RollingUpdateStrategy 

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

RollingUpdateStrategy is a cluster node update strategy which will roll the nodes with a specified surge.

func NewRollingUpdateStrategy 

func NewRollingUpdateStrategy(logger *log.Entry, nodePoolManager NodePoolManager, surge int) *RollingUpdateStrategy

NewRollingUpdateStrategy initializes a new RollingUpdateStrategy.

func (*RollingUpdateStrategy) PrepareForRemoval 

func (r *RollingUpdateStrategy) PrepareForRemoval(ctx context.Context, nodePoolDesc *api.NodePool) error

func (*RollingUpdateStrategy) Update 

func (r *RollingUpdateStrategy) Update(ctx context.Context, nodePoolDesc *api.NodePool) error

Update performs a rolling update of a single node pool. Passing a context allows stopping the update loop in case the context is canceled.

type UpdateStrategy 

type UpdateStrategy interface {
	Update(ctx context.Context, nodePool *api.NodePool) error
	PrepareForRemoval(ctx context.Context, nodePool *api.NodePool) error
}

UpdateStrategy defines an interface for performing cluster node updates.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.