kubescaler

package module

v0.1.2 Latest Latest Go to latest Published: Jan 22, 2022 License: MIT Imports: 17 Imported by: 0

Details

Valid go.mod file

The Go module system was introduced in Go 1.11 and is the official dependency management solution for Go.
Redistributable license

Redistributable licenses place minimal restrictions on how software can be used, modified, and redistributed.
Tagged version

Modules with tagged versions give importers more predictable builds.
Stable version

When a project reaches major version v1 it is considered stable.
Learn more about best practices

Repository

github.com/theredrad/kubescaler

Links

Open Source Insights

README ¶

Kubescaler

The kubescaler is an app to scale k8s cluster node pool to run a bunch of stateful dedicated servers (like dedicated game server using memory state) using kubernetes API inspired by Mark Mandel blog post . This application is a simple scaler and you can use Agones for more advanced features.

Scale Strategy

The scaler follows a simple rule to scale the nodes. There is a buffer size and if the resource request is more than the buffer size, needed new nodes are created and if the resource request decreases, the extra nodes are deleted.

The scaler runs the scale method in a loop. If the available resource is smaller than the buffer size, first check the unschedulable nodes and if the resource isn't enough, then resize the pool size to the needed nodes. Otherwise, if the available resource is greater than the buffer size, first calculate the extra nodes and then mark them as unschedulable to prevent scheduling new pods on them. at the end check the unschedulable nodes and deletes expired nodes with no dedicated server pods.

Permissions

kubescaler uses cluster config to manage nodes & pods, so permissions and roles must be applied to the kubescaler Deployment

Configs

The example config file exists as config.yaml.exmaple file. Also, you can set the configs as environment variables in uppercase and snail case format.

cloud-provider : Currently only digitalocean is implemented as a cloud provider, but the app supports driver, so you can implement any other provider and only register it (or contribute to this repo and send a pull request). Cloud provider manages the node pool size and deletes extra nodes.
cloud-provider-token : Access token for cloud API
cluster-name : Cluster name
node-pool-name : Node pool name
node-selector : Kubernetes selector to target the dedicated servers nodes. (ex: "role=scalable"). This selector is used to find the nodes that host the dedicated server pods.
cluster-kube-config-master-url : cluster kube config master URL to use k8s API (leave empty if using in-cluster config)
cluster-kube-config-path : cluster kube config path (leave empty if using in-cluster config)
minimum-node-pool-size : minimum node pool size
maximum-node-pool-size : maximum node pool size
server-pod-label-name : dedicated server pod label name (using to filter the dedicated server pods)
server-pod-label-value : dedicated server pod label value (using to filter the dedicated server pods)
buffer-slot-size : buffer slot size
scale-loop-tick-sec : scale loop tick duration in seconds
server-cpu-resource-request : dedicated server pod CPU resource request (in MilliValue)
empty-node-expiration-sec : empty node expiration duration in seconds (delete node after this time if no pods scheduled)

TODOs

Add kubernetes deployment
Add health check

Documentation ¶

Index ¶

Variables
func ClientSet(masterURL, kubeConfigPath string) (kubernetes.Interface, error)
func NewDefaultLogger(infoLogger *log.Logger, debugLogger *log.Logger, errorLogger *log.Logger) *defaultLogger
func SortNodesByPods(nodes []*Node)
func SortNodesByPodsDesc(nodes []*Node)
type Config
type K8S
- func NewK8S(i kubernetes.Interface) *K8S
- func NewK8SFromKubeConfig(masterURL, kubeConfigPath string) (*K8S, error)
- func (k *K8S) NewPodWatcher(ctx context.Context, namespace, labelSelector string) (*PodWatcher, error)
- func (k *K8S) NodePods(ctx context.Context, nodeName string) (*v1.PodList, error)
- func (k *K8S) Nodes(ctx context.Context, selector string) (*NodeList, error)
- func (k *K8S) UpdateNode(ctx context.Context, node *Node) error
type Kubernetes
type Logger
type Node
- func (n *Node) AvailableResource(resource v1.ResourceName) int64
- func (n *Node) IsReady() bool
- func (n *Node) IsSchedulable() bool
- func (n *Node) MarkAsSchedulable() error
- func (n *Node) MarkAsUnschedulable() error
- func (n *Node) ResourceCapacity(resource v1.ResourceName) int64
- func (n *Node) SchedulingMarkTimestamp() (time.Time, error)
- func (n *Node) UsingResources(resource v1.ResourceName) int64
type NodeList
- func (n *NodeList) AvailableNodes() []*Node
- func (n *NodeList) AvailableResource(resource v1.ResourceName) int64
- func (n *NodeList) AvailableSlot(need Resource) int64
- func (n *NodeList) SchedulableNodes() []*Node
- func (n *NodeList) UnschedulableNodes() []*Node
type PodWatcher
- func (pw *PodWatcher) Stop()
- func (pw *PodWatcher) Watch()
type Resource
type Scaler
- func NewScaler(npm nodepoolmanager.Provider, k8s Kubernetes, config *Config) *Scaler
- func (s *Scaler) Start() error
- func (s *Scaler) Stop()

Constants ¶

This section is empty.

Variables ¶

View Source

var (
	ErrNotEnoughResources = errors.New("not enough resources")
)

Functions ¶

func ClientSet ¶

func ClientSet(masterURL, kubeConfigPath string) (kubernetes.Interface, error)

func NewDefaultLogger ¶

func NewDefaultLogger(infoLogger *log.Logger, debugLogger *log.Logger, errorLogger *log.Logger) *defaultLogger

func SortNodesByPods ¶

func SortNodesByPods(nodes []*Node)

func SortNodesByPodsDesc ¶

func SortNodesByPodsDesc(nodes []*Node)

Types ¶

type Config ¶

type Config struct {
	NodeSelector string

	MinimumNode int
	MaximumNode int

	PodCPURequest int64
	PodLabelName  string
	PodLabelValue string

	EmptyNodeExpiration time.Duration

	BufferSlotSize int64

	ScaleLoopDuration time.Duration

	Logger Logger
}

type K8S ¶

type K8S struct {
	// contains filtered or unexported fields
}

func NewK8S ¶

func NewK8S(i kubernetes.Interface) *K8S

func NewK8SFromKubeConfig ¶

func NewK8SFromKubeConfig(masterURL, kubeConfigPath string) (*K8S, error)

func (*K8S) NewPodWatcher ¶

func (k *K8S) NewPodWatcher(ctx context.Context, namespace, labelSelector string) (*PodWatcher, error)

func (*K8S) NodePods ¶

func (k *K8S) NodePods(ctx context.Context, nodeName string) (*v1.PodList, error)

func (*K8S) Nodes ¶

func (k *K8S) Nodes(ctx context.Context, selector string) (*NodeList, error)

func (*K8S) UpdateNode ¶

func (k *K8S) UpdateNode(ctx context.Context, node *Node) error

type Kubernetes ¶

type Kubernetes interface {
	Nodes(ctx context.Context, selector string) (*NodeList, error)
	UpdateNode(ctx context.Context, node *Node) error
	NewPodWatcher(ctx context.Context, namespace, labelSelector string) (*PodWatcher, error)
}

type Logger ¶

type Logger interface {
	Info(v ...interface{})
	Infof(format string, v ...interface{})
	Debug(v ...interface{})
	Debugf(format string, v ...interface{})
	Error(v ...interface{})
	Errorf(format string, v ...interface{})
}

type Node ¶

type Node struct {
	N    *v1.Node
	Pods []v1.Pod
}

func (*Node) AvailableResource ¶

func (n *Node) AvailableResource(resource v1.ResourceName) int64

func (*Node) IsReady ¶

func (n *Node) IsReady() bool

func (*Node) IsSchedulable ¶

func (n *Node) IsSchedulable() bool

func (*Node) MarkAsSchedulable ¶

func (n *Node) MarkAsSchedulable() error

func (*Node) MarkAsUnschedulable ¶

func (n *Node) MarkAsUnschedulable() error

func (*Node) ResourceCapacity ¶

func (n *Node) ResourceCapacity(resource v1.ResourceName) int64

func (*Node) SchedulingMarkTimestamp ¶

func (n *Node) SchedulingMarkTimestamp() (time.Time, error)

func (*Node) UsingResources ¶

func (n *Node) UsingResources(resource v1.ResourceName) int64

type NodeList ¶

type NodeList struct {
	Nodes []*Node
}

func (*NodeList) AvailableNodes ¶

func (n *NodeList) AvailableNodes() []*Node

func (*NodeList) AvailableResource ¶

func (n *NodeList) AvailableResource(resource v1.ResourceName) int64

func (*NodeList) AvailableSlot ¶

func (n *NodeList) AvailableSlot(need Resource) int64

func (*NodeList) SchedulableNodes ¶

func (n *NodeList) SchedulableNodes() []*Node

func (*NodeList) UnschedulableNodes ¶

func (n *NodeList) UnschedulableNodes() []*Node

type PodWatcher ¶

type PodWatcher struct {
	Events chan watch.Event
	// contains filtered or unexported fields
}

func (*PodWatcher) Stop ¶

func (pw *PodWatcher) Stop()

func (*PodWatcher) Watch ¶

func (pw *PodWatcher) Watch()

type Resource ¶

type Resource struct {
	Name  v1.ResourceName
	Value int64
}

type Scaler ¶

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

func NewScaler ¶

func NewScaler(npm nodepoolmanager.Provider, k8s Kubernetes, config *Config) *Scaler

func (*Scaler) Start ¶

func (s *Scaler) Start() error

func (*Scaler) Stop ¶

func (s *Scaler) Stop()

Source Files ¶

View all Source files

Directories ¶

Path	Synopsis
cmd
kubescaler
mocks Package mocks is a generated GoMock package.	Package mocks is a generated GoMock package.
nodepoolmanager
providers/digitalocean

?	: This menu
/	: Search site
f or F	: Jump to
y or Y	: Canonical URL