Example #1
0
// CheckMigsAndNodes checks if all migs have all required nodes.
func CheckMigsAndNodes(nodes []*kube_api.Node, gceManager *gce.GceManager) error {
	migCount := make(map[string]int)
	migs := make(map[string]*config.MigConfig)
	for _, node := range nodes {
		instanceConfig, err := config.InstanceConfigFromProviderId(node.Spec.ProviderID)
		if err != nil {
			return err
		}

		migConfig, err := gceManager.GetMigForInstance(instanceConfig)
		if err != nil {
			return err
		}
		url := migConfig.Url()
		count, _ := migCount[url]
		migCount[url] = count + 1
		migs[url] = migConfig
	}
	for url, mig := range migs {
		size, err := gceManager.GetMigSize(mig)
		if err != nil {
			return err
		}
		count := migCount[url]
		if size != int64(count) {
			return fmt.Errorf("wrong number of nodes for mig: %s expected: %d actual: %d", url, size, count)
		}
	}
	return nil
}
Example #2
0
// GetNodeInfosForMigs finds NodeInfos for all migs used to manage the given nodes. It also returns a mig to sample node mapping.
func GetNodeInfosForMigs(nodes []*kube_api.Node, gceManager *gce.GceManager, kubeClient *kube_client.Client) (map[string]*schedulercache.NodeInfo,
	map[string]*kube_api.Node, error) {
	sampleNodes := make(map[string]*kube_api.Node)
	for _, node := range nodes {
		instanceConfig, err := config.InstanceConfigFromProviderId(node.Spec.ProviderID)
		if err != nil {
			return map[string]*schedulercache.NodeInfo{}, map[string]*kube_api.Node{}, err
		}

		migConfig, err := gceManager.GetMigForInstance(instanceConfig)
		if err != nil {
			return map[string]*schedulercache.NodeInfo{}, map[string]*kube_api.Node{}, err
		}
		url := migConfig.Url()
		sampleNodes[url] = node
	}
	result := make(map[string]*schedulercache.NodeInfo)
	for url, node := range sampleNodes {
		nodeInfo, err := simulator.BuildNodeInfoForNode(node.Name, kubeClient)
		if err != nil {
			return map[string]*schedulercache.NodeInfo{}, map[string]*kube_api.Node{}, err
		}
		result[url] = nodeInfo
	}
	return result, sampleNodes, nil
}
Example #3
0
// ScaleDown tries to scale down the cluster. It returns ScaleDownResult indicating if any node was
// removed and error if such occured.
func ScaleDown(
	nodes []*kube_api.Node,
	unneededNodes map[string]time.Time,
	unneededTime time.Duration,
	pods []*kube_api.Pod,
	gceManager *gce.GceManager,
	client *kube_client.Client,
	predicateChecker *simulator.PredicateChecker) (ScaleDownResult, error) {

	now := time.Now()
	candidates := make([]*kube_api.Node, 0)
	for _, node := range nodes {
		if val, found := unneededNodes[node.Name]; found {

			glog.V(2).Infof("%s was unneeded for %s", node.Name, now.Sub(val).String())

			// Check how long the node was underutilized.
			if !val.Add(unneededTime).Before(now) {
				continue
			}

			// Check mig size.
			instance, err := config.InstanceConfigFromProviderId(node.Spec.ProviderID)
			if err != nil {
				glog.Errorf("Error while parsing providerid of %s: %v", node.Name, err)
				continue
			}
			migConfig, err := gceManager.GetMigForInstance(instance)
			if err != nil {
				glog.Errorf("Error while checking mig config for instance %v: %v", instance, err)
				continue
			}
			size, err := gceManager.GetMigSize(migConfig)
			if err != nil {
				glog.Errorf("Error while checking mig size for instance %v: %v", instance, err)
				continue
			}

			if size <= int64(migConfig.MinSize) {
				glog.V(1).Infof("Skipping %s - mig min size reached", node.Name)
				continue
			}

			candidates = append(candidates, node)
		}
	}
	if len(candidates) == 0 {
		glog.Infof("No candidates for scale down")
		return ScaleDownNoUnneeded, nil
	}

	nodesToRemove, err := simulator.FindNodesToRemove(candidates, nodes, pods, client, predicateChecker, 1, false)
	if err != nil {
		return ScaleDownError, fmt.Errorf("Find node to remove failed: %v", err)
	}
	if len(nodesToRemove) == 0 {
		glog.V(1).Infof("No node to remove")
		return ScaleDownNoNodeDeleted, nil
	}
	nodeToRemove := nodesToRemove[0]
	glog.Infof("Removing %s", nodeToRemove.Name)
	instanceConfig, err := config.InstanceConfigFromProviderId(nodeToRemove.Spec.ProviderID)
	if err != nil {
		return ScaleDownError, fmt.Errorf("Failed to get instance config for %s: %v", nodeToRemove.Name, err)
	}

	err = gceManager.DeleteInstances([]*config.InstanceConfig{instanceConfig})
	if err != nil {
		return ScaleDownError, fmt.Errorf("Failed to delete %v: %v", instanceConfig, err)
	}

	return ScaleDownNodeDeleted, nil
}