add IsNodeCandidateForScaleDown interface

This function allows cloud providers to specify when a node is not a
good candidate for scaling down. This will occur before the autoscaler has
begun to cordon, drain, and taint any node for scale down.

Also adds a unit test for the prefiltering node processor.

cluster-autoscaler/cloudprovider/alicloud/alicloud_cloud_provider.go

@@ -175,6 +175,11 @@ func (ali *aliCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (ali *aliCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // AliRef contains a reference to ECS instance or .
 type AliRef struct {
 	ID     string

cluster-autoscaler/cloudprovider/aws/aws_cloud_provider.go

@@ -193,6 +193,11 @@ func (aws *awsCloudProvider) Refresh() error {
 	return aws.awsManager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (aws *awsCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // AwsRef contains a reference to some entity in AWS world.
 type AwsRef struct {
 	Name string

cluster-autoscaler/cloudprovider/azure/azure_cloud_provider.go

@@ -174,6 +174,11 @@ func (azure *AzureCloudProvider) Refresh() error {
 	return azure.azureManager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (azure *AzureCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // azureRef contains a reference to some entity in Azure world.
 type azureRef struct {
 	Name string

cluster-autoscaler/cloudprovider/baiducloud/baiducloud_cloud_provider.go

@@ -228,6 +228,11 @@ func (baiducloud *baiducloudCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (baiducloud *baiducloudCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BaiducloudRef contains a reference to some entity in baiducloud world.
 type BaiducloudRef struct {
 	Name string

cluster-autoscaler/cloudprovider/bizflycloud/bizflycloud_cloud_provider.go

@@ -172,6 +172,11 @@ func (d *bizflycloudCloudProvider) Refresh() error {
 	return d.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (d *bizflycloudCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildBizflyCloud builds the Bizflycloud cloud provider.
 func BuildBizflyCloud(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/brightbox/brightbox_cloud_provider.go

@@ -219,6 +219,11 @@ func (b *brightboxCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (b *brightboxCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildBrightbox builds the Brightbox provider
 func BuildBrightbox(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/cherryservers/cherry_cloud_provider.go

@@ -170,6 +170,11 @@ func (ccp *cherryCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (ccp *cherryCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildCherry is called by the autoscaler to build a Cherry Servers cloud provider.
 //
 // The cherryManager is created here, and the node groups are created

cluster-autoscaler/cloudprovider/civo/civo_cloud_provider.go

@@ -167,6 +167,11 @@ func (d *civoCloudProvider) Refresh() error {
 	return d.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (d *civoCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildCivo builds the Civo cloud provider.
 func BuildCivo(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/cloud_provider.go

@@ -154,6 +154,14 @@ type CloudProvider interface {
 	// Refresh is called before every main loop and can be used to dynamically update cloud provider state.
 	// In particular the list of node groups returned by NodeGroups can change as a result of CloudProvider.Refresh().
 	Refresh() error
+
+	// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down. This function
+	// will be called during prefiltering of nodes for scaledown to allow cloud providers the opportunity
+	// to reject a node for scale down. This may be used in cases where nodes are undergoing upgrades or other
+	// cloud-specific behavior where the cluster autoscaler should not begin cordoning, draining, and tainting
+	// the node.
+	// Returns true if the node can be safely scaled down or false otherwise.
+	IsNodeCandidateForScaleDown(*apiv1.Node) (bool, error)
 }
 
 // ErrNotImplemented is returned if a method is not implemented.

cluster-autoscaler/cloudprovider/cloudstack/cloudstack_cloud_provider.go

@@ -115,6 +115,11 @@ func (provider *cloudStackCloudProvider) Pricing() (cloudprovider.PricingModel,
 	return nil, cloudprovider.ErrNotImplemented
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (provider *cloudStackCloudProvider) IsNodeCandidateForScaleDown(node *v1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // NewNodeGroup builds a theoretical node group based on the node definition provided. The node group is not automatically
 // created on the cloud provider side. The node group is not returned by NodeGroups() until it is created.
 func (provider *cloudStackCloudProvider) NewNodeGroup(machineType string, labels map[string]string, systemLabels map[string]string, taints []v1.Taint, extraResources map[string]resource.Quantity) (cloudprovider.NodeGroup, error) {

cluster-autoscaler/cloudprovider/clusterapi/clusterapi_provider.go

@@ -140,6 +140,11 @@ func (p *provider) GetNodeGpuConfig(node *corev1.Node) *cloudprovider.GpuConfig
 	return gpu.GetNodeGPUFromCloudProvider(p, node)
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (p *provider) IsNodeCandidateForScaleDown(node *corev1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 func newProvider(
 	name string,
 	rl *cloudprovider.ResourceLimiter,

cluster-autoscaler/cloudprovider/coreweave/coreweave_provider.go

@@ -181,6 +181,11 @@ func (c *CoreWeaveCloudProvider) Refresh() error {
 	return c.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (c *CoreWeaveCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildCoreWeave builds the CoreWeave cloud provider with the given options and returns it.
 func BuildCoreWeave(opts config.AutoscalingOptions, do cloudprovider.NodeGroupDiscoveryOptions, rl *cloudprovider.ResourceLimiter) cloudprovider.CloudProvider {
 	klog.V(4).Infof("Building CoreWeave cloud provider with options: %+v", opts)

cluster-autoscaler/cloudprovider/digitalocean/digitalocean_cloud_provider.go

@@ -169,6 +169,11 @@ func (d *digitaloceanCloudProvider) Refresh() error {
 	return d.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (d *digitaloceanCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildDigitalOcean builds the DigitalOcean cloud provider.
 func BuildDigitalOcean(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/equinixmetal/cloud_provider.go

@@ -176,6 +176,11 @@ func (pcp *equinixMetalCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (pcp *equinixMetalCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildCloudProvider is called by the autoscaler to build an Equinix Metal cloud provider.
 //
 // The equinixMetalManager is created here, and the node groups are created

cluster-autoscaler/cloudprovider/exoscale/exoscale_cloud_provider.go

@@ -224,6 +224,11 @@ func (e *exoscaleCloudProvider) Refresh() error {
 	return e.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (e *exoscaleCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildExoscale builds the Exoscale cloud provider.
 func BuildExoscale(_ config.AutoscalingOptions, discoveryOpts cloudprovider.NodeGroupDiscoveryOptions, rl *cloudprovider.ResourceLimiter) cloudprovider.CloudProvider {
 	manager, err := newManager(discoveryOpts)

cluster-autoscaler/cloudprovider/externalgrpc/externalgrpc_cloud_provider.go

@@ -316,6 +316,11 @@ func (e *externalGrpcCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (e *externalGrpcCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildExternalGrpc builds the externalgrpc cloud provider.
 func BuildExternalGrpc(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/gce/gce_cloud_provider.go

@@ -148,6 +148,11 @@ func (gce *GceCloudProvider) Refresh() error {
 	return gce.gceManager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (gce *GceCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // GceRef contains s reference to some entity in GCE world.
 type GceRef struct {
 	Project string

cluster-autoscaler/cloudprovider/hetzner/hetzner_cloud_provider.go

@@ -182,6 +182,11 @@ func (d *HetznerCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (d *HetznerCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildHetzner builds the Hetzner cloud provider.
 func BuildHetzner(_ config.AutoscalingOptions, do cloudprovider.NodeGroupDiscoveryOptions, rl *cloudprovider.ResourceLimiter) cloudprovider.CloudProvider {
 	manager, err := newManager()

cluster-autoscaler/cloudprovider/huaweicloud/huaweicloud_cloud_provider.go

@@ -179,6 +179,11 @@ func (hcp *huaweicloudCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (hcp *huaweicloudCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 func (hcp *huaweicloudCloudProvider) buildAsgs(specs []string) error {
 	asgs, err := hcp.cloudServiceManager.ListScalingGroups()
 	if err != nil {

cluster-autoscaler/cloudprovider/ionoscloud/ionoscloud_cloud_provider.go

@@ -308,6 +308,11 @@ func (ic *IonosCloudCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (ic *IonosCloudCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildIonosCloud builds the IonosCloud cloud provider.
 func BuildIonosCloud(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/kamatera/kamatera_cloud_provider.go

@@ -129,6 +129,11 @@ func (k *kamateraCloudProvider) Refresh() error {
 	return k.manager.refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (k *kamateraCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildKamatera builds the Kamatera cloud provider.
 func BuildKamatera(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/kubemark/kubemark_linux.go

@@ -180,6 +180,11 @@ func (kubemark *KubemarkCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (kubemark *KubemarkCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // NodeGroup implements NodeGroup interface.
 type NodeGroup struct {
 	Name               string

cluster-autoscaler/cloudprovider/kwok/kwok_provider.go

@@ -170,6 +170,11 @@ func (kwok *KwokCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (kwok *KwokCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildKwok builds kwok cloud provider.
 func BuildKwok(opts config.AutoscalingOptions,
 	do cloudprovider.NodeGroupDiscoveryOptions,

cluster-autoscaler/cloudprovider/linode/linode_cloud_provider.go

@@ -151,6 +151,11 @@ func (l *linodeCloudProvider) Refresh() error {
 	return l.manager.refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (l *linodeCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 func newLinodeCloudProvider(config io.Reader, rl *cloudprovider.ResourceLimiter) (cloudprovider.CloudProvider, error) {
 	m, err := newManager(config)
 	if err != nil {

cluster-autoscaler/cloudprovider/magnum/magnum_cloud_provider.go

@@ -203,6 +203,11 @@ func (mcp *magnumCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (mcp *magnumCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // refreshNodeGroups gets the list of node groups which meet the requirements for autoscaling,
 // creates magnumNodeGroups for any that do not exist in the cloud provider,
 // and drops any node groups which are present in the cloud provider but not in the

cluster-autoscaler/cloudprovider/mocks/CloudProvider.go

@@ -276,3 +276,8 @@ func (_m *CloudProvider) Refresh() error {
 
 	return r0
 }
+
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (_m *CloudProvider) IsNodeCandidateForScaleDown(node *v1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}

cluster-autoscaler/cloudprovider/oci/instancepools/oci_cloud_provider.go

@@ -147,6 +147,11 @@ func (ocp *OciCloudProvider) Refresh() error {
 	return ocp.poolManager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (ocp *OciCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildOCI constructs the OciCloudProvider object that implements the could provider interface (InstancePoolManager).
 func BuildOCI(opts config.AutoscalingOptions, do cloudprovider.NodeGroupDiscoveryOptions, rl *cloudprovider.ResourceLimiter) cloudprovider.CloudProvider {
 	ocidType, err := ocicommon.GetAllPoolTypes(opts.NodeGroups)

cluster-autoscaler/cloudprovider/oci/nodepools/oci_cloud_provider.go

@@ -149,3 +149,8 @@ func (ocp *OciCloudProvider) Cleanup() error {
 func (ocp *OciCloudProvider) Refresh() error {
 	return ocp.manager.Refresh()
 }
+
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (ocp *OciCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}

cluster-autoscaler/cloudprovider/ovhcloud/ovh_cloud_provider.go

@@ -318,3 +318,8 @@ func (provider *OVHCloudProvider) Refresh() error {
 
 	return nil
 }
+
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (provider *OVHCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}

cluster-autoscaler/cloudprovider/rancher/rancher_provider.go

@@ -219,6 +219,11 @@ func (provider *RancherCloudProvider) Cleanup() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (provider *RancherCloudProvider) IsNodeCandidateForScaleDown(node *corev1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 func (provider *RancherCloudProvider) scalableNodeGroups() ([]*nodeGroup, error) {
 	var result []*nodeGroup
 

cluster-autoscaler/cloudprovider/scaleway/scaleway_cloud_provider.go

@@ -277,3 +277,8 @@ func (scw *scalewayCloudProvider) Refresh() error {
 
 	return nil
 }
+
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (scw *scalewayCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}

cluster-autoscaler/cloudprovider/tencentcloud/tencentcloud_cloud_provider.go

@@ -173,6 +173,11 @@ func (tencentcloud *tencentCloudProvider) Refresh() error {
 	return tencentcloud.tencentcloudManager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (tencentcloud *tencentCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildTencentcloud returns tencentcloud provider
 func BuildTencentcloud(opts config.AutoscalingOptions, do cloudprovider.NodeGroupDiscoveryOptions, rl *cloudprovider.ResourceLimiter) cloudprovider.CloudProvider {
 	var config io.ReadCloser

cluster-autoscaler/cloudprovider/test/test_cloud_provider.go

@@ -45,20 +45,25 @@ type OnNodeGroupDeleteFunc func(string) error
 // HasInstance is a function called to determine if a node has been removed from the cloud provider.
 type HasInstance func(string) (bool, error)
 
+// IsNodeCandidateForScaleDown is a function called to determine if a cloud provider considers
+// a node a good candidate for scaling down.
+type IsNodeCandidateForScaleDown func(*apiv1.Node) (bool, error)
+
 // TestCloudProvider is a dummy cloud provider to be used in tests.
 type TestCloudProvider struct {
 	sync.Mutex
-	nodes             map[string]string
-	groups            map[string]cloudprovider.NodeGroup
-	onScaleUp         func(string, int) error
-	onScaleDown       func(string, string) error
-	onNodeGroupCreate func(string) error
-	onNodeGroupDelete func(string) error
-	hasInstance       func(string) (bool, error)
-	machineTypes      []string
-	machineTemplates  map[string]*framework.NodeInfo
-	priceModel        cloudprovider.PricingModel
-	resourceLimiter   *cloudprovider.ResourceLimiter
+	nodes                       map[string]string
+	groups                      map[string]cloudprovider.NodeGroup
+	onScaleUp                   func(string, int) error
+	onScaleDown                 func(string, string) error
+	onNodeGroupCreate           func(string) error
+	onNodeGroupDelete           func(string) error
+	hasInstance                 func(string) (bool, error)
+	isNodeCandidateForScaleDown func(*apiv1.Node) (bool, error)
+	machineTypes                []string
+	machineTemplates            map[string]*framework.NodeInfo
+	priceModel                  cloudprovider.PricingModel
+	resourceLimiter             *cloudprovider.ResourceLimiter
 }
 
 // TestCloudProviderBuilder is used to create CloudProvider
@@ -127,6 +132,14 @@ func (b *TestCloudProviderBuilder) WithHasInstance(hasInstance HasInstance) *Tes
 	return b
 }
 
+// WithIsNodeCandidateForScaleDown adds an IsNodeCandidateForScaleDown handler to provider.
+func (b *TestCloudProviderBuilder) WithIsNodeCandidateForScaleDown(isNodeCandidateForScaleDown IsNodeCandidateForScaleDown) *TestCloudProviderBuilder {
+	b.builders = append(b.builders, func(p *TestCloudProvider) {
+		p.isNodeCandidateForScaleDown = isNodeCandidateForScaleDown
+	})
+	return b
+}
+
 // Build returns a built test cloud provider
 func (b *TestCloudProviderBuilder) Build() *TestCloudProvider {
 	p := &TestCloudProvider{
@@ -352,6 +365,15 @@ func (tcp *TestCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (tcp *TestCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	if tcp.isNodeCandidateForScaleDown == nil {
+		return true, cloudprovider.ErrNotImplemented
+	}
+
+	return tcp.isNodeCandidateForScaleDown(node)
+}
+
 // TestNodeGroup is a node group used by TestCloudProvider.
 type TestNodeGroup struct {
 	sync.Mutex

cluster-autoscaler/cloudprovider/utho/utho_cloud_provider.go

@@ -162,6 +162,11 @@ func (u *uthoCloudProvider) Refresh() error {
 	return u.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (u *uthoCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // BuildUtho builds the Utho cloud provider.
 func BuildUtho(
 	opts config.AutoscalingOptions,

cluster-autoscaler/cloudprovider/volcengine/volcengine_cloud_provider.go

@@ -120,6 +120,11 @@ func (v *volcengineCloudProvider) Refresh() error {
 	return nil
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (v *volcengineCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // GetNodeGpuConfig returns the label, type and resource name for the GPU added to node. If node doesn't have
 // any GPUs, it returns nil.
 func (v *volcengineCloudProvider) GetNodeGpuConfig(node *apiv1.Node) *cloudprovider.GpuConfig {

cluster-autoscaler/cloudprovider/vultr/vultr_cloud_provider.go

@@ -140,6 +140,11 @@ func (v *vultrCloudProvider) Refresh() error {
 	return v.manager.Refresh()
 }
 
+// IsNodeCandidateForScaleDown returns whether the node is a good candidate for scaling down.
+func (v *vultrCloudProvider) IsNodeCandidateForScaleDown(node *apiv1.Node) (bool, error) {
+	return true, cloudprovider.ErrNotImplemented
+}
+
 // toProviderID returns a provider ID from the given node ID.
 func toProviderID(nodeID string) string {
 	return fmt.Sprintf("%s%s", vultrProviderIDPrefix, nodeID)

cluster-autoscaler/processors/nodes/pre_filtering_processor.go

@@ -22,6 +22,7 @@ import (
 	apiv1 "k8s.io/api/core/v1"
 	klog "k8s.io/klog/v2"
 
+	"k8s.io/autoscaler/cluster-autoscaler/cloudprovider"
 	"k8s.io/autoscaler/cluster-autoscaler/context"
 	"k8s.io/autoscaler/cluster-autoscaler/utils"
 	"k8s.io/autoscaler/cluster-autoscaler/utils/errors"
@@ -48,6 +49,15 @@ func (n *PreFilteringScaleDownNodeProcessor) GetScaleDownCandidates(ctx *context
 	nodeGroupSize := utils.GetNodeGroupSizeMap(ctx.CloudProvider)
 
 	for _, node := range nodes {
+		if candidate, err := ctx.CloudProvider.IsNodeCandidateForScaleDown(node); err != nil {
+			if err != cloudprovider.ErrNotImplemented {
+				klog.Warningf("Error while checking if node is a candidate for deletion %s: %v", node.Name, err)
+				continue
+			}
+		} else if !candidate {
+			klog.V(5).Infof("Node %s is not a candidate for deletion (cloud provider determined)", node.Name)
+			continue
+		}
 		nodeGroup, err := ctx.CloudProvider.NodeGroupForNode(node)
 		if err != nil {
 			klog.Warningf("Error while checking node group for %s: %v", node.Name, err)

cluster-autoscaler/processors/nodes/pre_filtering_processor_test.go

@@ -17,6 +17,7 @@ limitations under the License.
 package nodes
 
 import (
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -44,23 +45,75 @@ func TestPreFilteringScaleDownNodeProcessor_GetScaleDownCandidateNodes(t *testin
 	ng1_1 := BuildTestNode("ng1-1", 1000, 1000)
 	ng1_2 := BuildTestNode("ng1-2", 1000, 1000)
 	ng2_1 := BuildTestNode("ng2-1", 1000, 1000)
+	ng2_2 := BuildTestNode("ng2-2", 1000, 1000)
 	noNg := BuildTestNode("no-ng", 1000, 1000)
-	provider := testprovider.NewTestCloudProviderBuilder().Build()
-	provider.AddNodeGroup("ng1", 1, 10, 2)
-	provider.AddNodeGroup("ng2", 1, 10, 1)
-	provider.AddNode("ng1", ng1_1)
-	provider.AddNode("ng1", ng1_2)
-	provider.AddNode("ng2", ng2_1)
 
-	ctx := &context.AutoscalingContext{
-		CloudProvider: provider,
+	testCases := map[string]struct {
+		buildProvider     func() *testprovider.TestCloudProvider
+		configureProvider func(p *testprovider.TestCloudProvider)
+		expectedNodes     []*apiv1.Node
+		inputNodes        []*apiv1.Node
+	}{
+		// Expectation: only node groups not at minimum size should be candidates.
+		"1 scale down candidate, 1 node group at minimum size, 1 node with no node group, 1 node group above minimum size.": {
+			configureProvider: func(p *testprovider.TestCloudProvider) {
+				p.AddNodeGroup("ng1", 1, 10, 2)
+				p.AddNodeGroup("ng2", 1, 10, 1)
+				p.AddNode("ng1", ng1_1)
+				p.AddNode("ng1", ng1_2)
+				p.AddNode("ng2", ng2_1)
+			},
+			expectedNodes: []*apiv1.Node{ng1_1, ng1_2},
+			inputNodes:    []*apiv1.Node{ng1_1, ng1_2, ng2_1, noNg},
+		},
+		// Expectation: only node groups that contain nodes the cloud provider considers candidates for deletion should be candidates.
+		"1 scale down candidate, 1 node group with nodes that are not candidates for deletion, 1 node group above minimum size.": {
+			buildProvider: func() *testprovider.TestCloudProvider {
+				provider := testprovider.
+					NewTestCloudProviderBuilder().
+					WithIsNodeCandidateForScaleDown(func(n *apiv1.Node) (bool, error) {
+						if strings.HasPrefix(n.Name, "ng2") {
+							return false, nil
+						}
+						return true, nil
+					}).
+					Build()
+				return provider
+			},
+			configureProvider: func(p *testprovider.TestCloudProvider) {
+				p.AddNodeGroup("ng1", 1, 10, 2)
+				p.AddNodeGroup("ng2", 1, 10, 2)
+				p.AddNode("ng1", ng1_1)
+				p.AddNode("ng1", ng1_2)
+				p.AddNode("ng2", ng2_1)
+				p.AddNode("ng2", ng2_2)
+			},
+			expectedNodes: []*apiv1.Node{ng1_1, ng1_2},
+			inputNodes:    []*apiv1.Node{ng1_1, ng1_2, ng2_1, ng2_2},
+		},
 	}
 
-	expectedNodes := []*apiv1.Node{ng1_1, ng1_2}
-	defaultProcessor := NewPreFilteringScaleDownNodeProcessor()
-	inputNodes := []*apiv1.Node{ng1_1, ng1_2, ng2_1, noNg}
-	result, err := defaultProcessor.GetScaleDownCandidates(ctx, inputNodes)
+	for description, testCase := range testCases {
+		t.Run(description, func(t *testing.T) {
+			var provider *testprovider.TestCloudProvider
+			if testCase.buildProvider == nil {
+				provider = testprovider.NewTestCloudProviderBuilder().Build()
+			} else {
+				provider = testCase.buildProvider()
+			}
+			assert.NotNil(t, provider)
 
-	assert.NoError(t, err)
-	assert.Equal(t, result, expectedNodes)
+			testCase.configureProvider(provider)
+
+			ctx := &context.AutoscalingContext{
+				CloudProvider: provider,
+			}
+
+			defaultProcessor := NewPreFilteringScaleDownNodeProcessor()
+			result, err := defaultProcessor.GetScaleDownCandidates(ctx, testCase.inputNodes)
+
+			assert.NoError(t, err)
+			assert.Equal(t, result, testCase.expectedNodes)
+		})
+	}
 }