tenzen-y / imperator Goto Github PK

View Code? Open in Web Editor NEW

2.0 2.0 0.0 418 KB

Imperator: The Kubernetes Operator to provide virtual resource groups.

License: Apache License 2.0

Dockerfile 0.45% Makefile 3.31% Go 89.69% Shell 6.54%

kubernetes-operator

imperator's People

Contributors

Stargazers

Watchers

imperator's Issues

Support updating pod managed by impelator

Now, Impelator does not support updating Pods managed by imperator, pod-resource-injector reject to update Pod.

imperator/pkg/api/v1alpha1/pod_resource_injector.go

Lines 95 to 119 in 5ea7494

    
           func (r *resourceInjector) replacePods(ctx context.Context, pod *corev1.Pod) error { 
        
           	// fetch old Pod 
        
           	oldPod := &corev1.Pod{} 
        
           	if err := r.Client.Get(ctx, client.ObjectKey{Name: pod.Name, Namespace: pod.Namespace}, oldPod); errors.IsNotFound(err) { 
        
           		return nil 
        
           	} 
        
           	return fmt.Errorf("it is forbidden to update the Pod") 
        
           	// TODO: Implementing replace pod 
        
           	//if len(oldPod.OwnerReferences) == 0 && os.Getenv("SKIP_OWNER_CHECK") != "true" { 
        
           	//	return fmt.Errorf("name: %s, namespace: %s; pods that are not managed by the parent resource can not be updated", 
        
           	//		oldPod.Name, oldPod.Namespace) 
        
           	//} 
        
           	// 
        
           	//if !oldPod.GetDeletionTimestamp().IsZero() { 
        
           	//	return fmt.Errorf("name: %s, namespace: %s; the pod is replacing now", pod.Name, pod.Namespace) 
        
           	//} 
        
           	// 
        
           	//if err := r.Client.Delete(ctx, oldPod, &client.DeleteOptions{}); err != nil { 
        
           	//	return err 
        
           	//} 
        
           	// 
        
           	//return nil 
        
           }

Must set label for nodeAffinity to Node

Currently, machinenodepool controller sets labels or taints to use as nodeAffinity at imperator-reservation and guest Pod although taints can not use as nodeAffinity. So it must set labels to Node and change MachineNodePool API to make taints optional.

design

type NodePool struct {

	// +kubebuilder:validation:Required
	Name string `json:"name"`

	// +kubebuilder:validation:Required
	// +kubebuilder:validation:Enum=ready;maintenance
	Mode NodePoolMode `json:"mode"`

	// +optional
- 	// default=label
- 	AssignmentType NodePoolAssignmentType `json:"assignmentType,omitempty"`
+       // default=false
+       Taint *bool `json:"taint,omitempty"`

	// +kubebuilder:validation:Required
	MachineType NodePoolMachineType `json:"machineType"`
}

Support more selectors for GPUs

Now, Impletor is only supporting nvidia.com/gpu.family to select GPU type below.

imperator/pkg/api/v1alpha1/machine_types.go

Lines 56 to 57 in 5ea7494

    
           // +optional 
        
           Family string `json:"generation,omitempty"`

I think it needs to support more types of selectors.

Example:

nvidia.com/gpu.product
nvidia.com/gpu.machine

https://github.com/NVIDIA/gpu-feature-discovery#generated-labels

Add additional some unit tests for machine webhook

I think it needs to add some unit tests for the machine webhook to verify func (r *Machine) Default() and func (r *Machine) ValidateNodePoolMachineTypeName() error

tenzen-y / imperator Goto Github PK

imperator's People

Contributors

Stargazers

Watchers

imperator's Issues

Support updating pod managed by impelator

Must set label for nodeAffinity to Node

Support more selectors for GPUs

Add additional some unit tests for machine webhook

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	func (r resourceInjector) replacePods(ctx context.Context, pod corev1.Pod) error {

	// fetch old Pod
	oldPod := &corev1.Pod{}
	if err := r.Client.Get(ctx, client.ObjectKey{Name: pod.Name, Namespace: pod.Namespace}, oldPod); errors.IsNotFound(err) {
	return nil
	}
	return fmt.Errorf("it is forbidden to update the Pod")

	// TODO: Implementing replace pod
	//if len(oldPod.OwnerReferences) == 0 && os.Getenv("SKIP_OWNER_CHECK") != "true" {
	// return fmt.Errorf("name: %s, namespace: %s; pods that are not managed by the parent resource can not be updated",
	// oldPod.Name, oldPod.Namespace)
	//}
	//
	//if !oldPod.GetDeletionTimestamp().IsZero() {
	// return fmt.Errorf("name: %s, namespace: %s; the pod is replacing now", pod.Name, pod.Namespace)
	//}
	//
	//if err := r.Client.Delete(ctx, oldPod, &client.DeleteOptions{}); err != nil {
	// return err
	//}
	//
	//return nil
	}