extender

README

How to build and deploy scheduler extender

1. Build binary and push image (if you don't change extender' code you don't have to rebuild binary and image):

       make build-extener
       make image-extender
       make push-extender
   

2. On node where kubernetes scheduler pod is run create folder /etc/kubernetes/scheduler e.g.

ssh root@provo-goop mkdir -p /etc/kubernetes/scheduler

1. Copy files config.yaml and policy.yaml from pkg/scheduler folder to the /etc/kubernetes/scheduler on the node. e.g.

scp policy.yaml config.yaml root@provo-goop:/etc/kubernetes/scheduler

1. Install extender chart:

   helm install se charts/scheduler-extender --set log.level=debug --set image.tag=`make version`
   

2. Modify kube-scheduler manifest on the node. Config file is located in /etc/kubernetes/manifests/kube-scheduler.yaml

   • add next volumes in .spec:

   - name: scheduler-config
     hostPath:
       path: /etc/kubernetes/scheduler/config.yaml
       type: File
   - name: scheduler-policy
     hostPath:
       path: /etc/kubernetes/scheduler/policy.yaml
       type: File
   

   • add volume mounts in .spec.containers[0].volumeMounts:

   - mountPath: /etc/kubernetes/scheduler/config.yaml
     name: scheduler-config
     readOnly: true
   - mountPath: /etc/kubernetes/scheduler/policy.yaml
     name: scheduler-policy
     readOnly: true
   

   • add next params for kube-scheduler entrypoint in .spec.containers[0].command:

   - --config=/etc/kubernetes/scheduler/config.yaml
   

   After you save changes in kube-scheduler.yaml kubernetes will restart scheduler.

3. Apply some pod manifest

4. Run

   kubectl logs -f -n %NAMESPACE_NAME% `kubectl get pods -n %NAMESPACE_NAME% --selector=app=csi-baremetal-se --no-headers | awk '{print $1}'`
   

   and observe as scheduler extender works

Documentation

Index

• type Extender
  • func NewExtender(logger *logrus.Logger, kubeClient *k8s.KubeClient, kubeCache *k8s.KubeCache, ...) (*Extender, error)
  • func (e *Extender) BindHandler(w http.ResponseWriter, req *http.Request)
  • func (e *Extender) FilterHandler(w http.ResponseWriter, req *http.Request)
  • func (e *Extender) PrioritizeHandler(w http.ResponseWriter, req *http.Request)

Types

type Extender

type Extender struct {
	sync.Mutex
	// contains filtered or unexported fields
}

Extender holds http handlers for scheduler extender endpoints and implements logic for nodes filtering based on pod volumes requirements and Available Capacities

func NewExtender

func NewExtender(logger *logrus.Logger, kubeClient *k8s.KubeClient,
	kubeCache *k8s.KubeCache, provisioner string, featureConf fc.FeatureChecker, annotationKey string) (*Extender, error)

NewExtender returns new instance of Extender struct

func (*Extender) BindHandler

func (e *Extender) BindHandler(w http.ResponseWriter, req *http.Request)

BindHandler does bind of a pod to specific node todo - not implemented. Was used for testing purposes ONLY (fault injection)!

func (*Extender) FilterHandler

func (e *Extender) FilterHandler(w http.ResponseWriter, req *http.Request)

FilterHandler extracts ExtenderArgs struct from req and writes ExtenderFilterResult to the w

func (*Extender) PrioritizeHandler

func (e *Extender) PrioritizeHandler(w http.ResponseWriter, req *http.Request)

PrioritizeHandler helps with even distribution of the volumes across the nodes. It will set priority based on the formula: rank of node X = max number of volumes - number of volume on node X.