dynamically retrieve instance types

docs/aws/karpenter.cloudformation.yaml

@@ -52,6 +52,8 @@ Resources:
               - "ec2:DescribeInstances"
               - "ec2:DescribeSecurityGroups"
               - "ec2:DescribeSubnets"
+              - "ec2:DescribeInstanceTypes"
+              - "ec2:DescribeInstanceTypeOfferings"
               - "iam:GetInstanceProfile"
               - "ssm:GetParameter"
   KarpenterNodeInstanceProfile:

pkg/cloudprovider/aws/fake/ec2api.go

@@ -30,6 +30,7 @@ type EC2API struct {
 	DescribeLaunchTemplatesOutput *ec2.DescribeLaunchTemplatesOutput
 	DescribeSubnetsOutput         *ec2.DescribeSubnetsOutput
 	DescribeSecurityGroupsOutput  *ec2.DescribeSecurityGroupsOutput
+	DescribeInstanceTypesOutput   *ec2.DescribeInstanceTypesOutput
 	WantErr                       error
 
 	CalledWithCreateFleetInput []ec2.CreateFleetInput
@@ -97,3 +98,93 @@ func (a *EC2API) DescribeSecurityGroupsWithContext(context.Context, *ec2.Describ
 	}
 	return &ec2.DescribeSecurityGroupsOutput{SecurityGroups: []*ec2.SecurityGroup{{GroupId: aws.String("test-group")}}}, nil
 }
+
+func (a *EC2API) DescribeInstanceTypesPagesWithContext(ctx context.Context, input *ec2.DescribeInstanceTypesInput, fn func(*ec2.DescribeInstanceTypesOutput, bool) bool, opts ...request.Option) error {
+	if a.WantErr != nil {
+		return a.WantErr
+	}
+	if a.DescribeInstanceTypesOutput != nil {
+		fn(a.DescribeInstanceTypesOutput, false)
+		return nil
+	}
+	fn(&ec2.DescribeInstanceTypesOutput{
+		InstanceTypes: []*ec2.InstanceTypeInfo{
+			{
+				InstanceType:                 aws.String("m5.large"),
+				SupportedUsageClasses:        []*string{aws.String("on-demand")},
+				SupportedVirtualizationTypes: []*string{aws.String("hvm")},
+				VCpuInfo: &ec2.VCpuInfo{
+					DefaultVCpus: aws.Int64(2),
+				},
+				MemoryInfo: &ec2.MemoryInfo{
+					SizeInMiB: aws.Int64(8),
+				},
+				NetworkInfo: &ec2.NetworkInfo{
+					MaximumNetworkInterfaces:  aws.Int64(3),
+					Ipv4AddressesPerInterface: aws.Int64(30),
+				},
+			},
+			{
+				InstanceType:                 aws.String("m5.xlarge"),
+				SupportedUsageClasses:        []*string{aws.String("on-demand")},
+				SupportedVirtualizationTypes: []*string{aws.String("hvm")},
+				VCpuInfo: &ec2.VCpuInfo{
+					DefaultVCpus: aws.Int64(4),
+				},
+				MemoryInfo: &ec2.MemoryInfo{
+					SizeInMiB: aws.Int64(16),
+				},
+				NetworkInfo: &ec2.NetworkInfo{
+					MaximumNetworkInterfaces:  aws.Int64(4),
+					Ipv4AddressesPerInterface: aws.Int64(60),
+				},
+			},
+			{
+				InstanceType:                 aws.String("m5.2xlarge"),
+				SupportedUsageClasses:        []*string{aws.String("on-demand")},
+				SupportedVirtualizationTypes: []*string{aws.String("hvm")},
+				VCpuInfo: &ec2.VCpuInfo{
+					DefaultVCpus: aws.Int64(8),
+				},
+				MemoryInfo: &ec2.MemoryInfo{
+					SizeInMiB: aws.Int64(32),
+				},
+				NetworkInfo: &ec2.NetworkInfo{
+					MaximumNetworkInterfaces:  aws.Int64(4),
+					Ipv4AddressesPerInterface: aws.Int64(60),
+				},
+			},
+			{
+				InstanceType:                 aws.String("m5.4xlarge"),
+				SupportedUsageClasses:        []*string{aws.String("on-demand")},
+				SupportedVirtualizationTypes: []*string{aws.String("hvm")},
+				VCpuInfo: &ec2.VCpuInfo{
+					DefaultVCpus: aws.Int64(16),
+				},
+				MemoryInfo: &ec2.MemoryInfo{
+					SizeInMiB: aws.Int64(64),
+				},
+				NetworkInfo: &ec2.NetworkInfo{
+					MaximumNetworkInterfaces:  aws.Int64(8),
+					Ipv4AddressesPerInterface: aws.Int64(240),
+				},
+			},
+			{
+				InstanceType:                 aws.String("m5.8xlarge"),
+				SupportedUsageClasses:        []*string{aws.String("on-demand")},
+				SupportedVirtualizationTypes: []*string{aws.String("hvm")},
+				VCpuInfo: &ec2.VCpuInfo{
+					DefaultVCpus: aws.Int64(32),
+				},
+				MemoryInfo: &ec2.MemoryInfo{
+					SizeInMiB: aws.Int64(128),
+				},
+				NetworkInfo: &ec2.NetworkInfo{
+					MaximumNetworkInterfaces:  aws.Int64(8),
+					Ipv4AddressesPerInterface: aws.Int64(240),
+				},
+			},
+		},
+	}, false)
+	return nil
+}

pkg/cloudprovider/aws/packing/nodecapacity.go

@@ -15,84 +15,40 @@ limitations under the License.
 package packing
 
 import (
+	"fmt"
+
+	"github.com/aws/aws-sdk-go/service/ec2"
 	resourcesUtil "github.com/awslabs/karpenter/pkg/utils/resources"
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
 )
 
-// TODO get this information from node-instance-selector
-var (
-	nodeCapacities = []*nodeCapacity{
-
-		{
-			instanceType: "m5.24xlarge",
-			total: v1.ResourceList{
-				v1.ResourceCPU:    resource.MustParse("96000m"),
-				v1.ResourceMemory: resource.MustParse("384Gi"),
-				v1.ResourcePods:   resource.MustParse("737"),
-			},
-			reserved: v1.ResourceList{
-				v1.ResourceCPU:    resource.Quantity{},
-				v1.ResourceMemory: resource.Quantity{},
-			},
-		},
-		{
-			instanceType: "m5.8xlarge",
-			total: v1.ResourceList{
-				v1.ResourceCPU:    resource.MustParse("32000m"),
-				v1.ResourceMemory: resource.MustParse("128Gi"),
-				v1.ResourcePods:   resource.MustParse("234"),
-			},
-			reserved: v1.ResourceList{
-				v1.ResourceCPU:    resource.Quantity{},
-				v1.ResourceMemory: resource.Quantity{},
-			},
-		},
-		{
-			instanceType: "m5.2xlarge",
-			total: v1.ResourceList{
-				v1.ResourceCPU:    resource.MustParse("8000m"),
-				v1.ResourceMemory: resource.MustParse("32Gi"),
-				v1.ResourcePods:   resource.MustParse("58"),
-			},
-			reserved: v1.ResourceList{
-				v1.ResourceCPU:    resource.Quantity{},
-				v1.ResourceMemory: resource.Quantity{},
-			},
-		},
-		{
-			instanceType: "m5.xlarge",
-			total: v1.ResourceList{
-				v1.ResourceCPU:    resource.MustParse("4000m"),
-				v1.ResourceMemory: resource.MustParse("16Gi"),
-				v1.ResourcePods:   resource.MustParse("58"),
-			},
-			reserved: v1.ResourceList{
-				v1.ResourceCPU:    resource.Quantity{},
-				v1.ResourceMemory: resource.Quantity{},
-			},
-		},
-		{
-			instanceType: "m5.large",
-			total: v1.ResourceList{
-				v1.ResourceCPU:    resource.MustParse("2000m"),
-				v1.ResourceMemory: resource.MustParse("8Gi"),
-				v1.ResourcePods:   resource.MustParse("29"),
-			},
-			reserved: v1.ResourceList{
-				v1.ResourceCPU:    resource.Quantity{},
-				v1.ResourceMemory: resource.Quantity{},
-			},
-		},
-	}
-)
-
 type nodeCapacity struct {
 	instanceType string
 	reserved     v1.ResourceList
 	total        v1.ResourceList
 }
 
+func nodeCapacityFrom(instanceTypeInfo ec2.InstanceTypeInfo) *nodeCapacity {
+	return &nodeCapacity{
+		instanceType: *instanceTypeInfo.InstanceType,
+		total: v1.ResourceList{
+			v1.ResourceCPU:    resource.MustParse(fmt.Sprint(*instanceTypeInfo.VCpuInfo.DefaultVCpus)),
+			v1.ResourceMemory: resource.MustParse(fmt.Sprintf("%dMi", *instanceTypeInfo.MemoryInfo.SizeInMiB)),
+			// The number of pods per node is calculated using the formula:
+			// max number of ENIs * (IPv4 Addresses per ENI -1) + 2
+			// https://github.com/awslabs/amazon-eks-ami/blob/master/files/eni-max-pods.txt#L20
+			v1.ResourcePods: resource.MustParse(fmt.Sprint(
+				*instanceTypeInfo.NetworkInfo.MaximumNetworkInterfaces*(*instanceTypeInfo.NetworkInfo.Ipv4AddressesPerInterface-1) + 2)),
+		},
+		reserved: v1.ResourceList{
+			v1.ResourceCPU:    resource.Quantity{},
+			v1.ResourceMemory: resource.Quantity{},
+			v1.ResourcePods:   resource.Quantity{},
+		},
+	}
+}
+
 func (nc *nodeCapacity) Copy() *nodeCapacity {
 	return &nodeCapacity{nc.instanceType, nc.reserved.DeepCopy(), nc.total.DeepCopy()}
 }

pkg/cloudprovider/aws/packing/packing.go

@@ -16,9 +16,13 @@ package packing
 
 import (
 	"context"
+	"fmt"
 	"sort"
 
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/ec2"
 	"github.com/aws/aws-sdk-go/service/ec2/ec2iface"
+	provisioningv1alpha1 "github.com/awslabs/karpenter/pkg/apis/provisioning/v1alpha1"
 	"github.com/awslabs/karpenter/pkg/cloudprovider"
 	"github.com/awslabs/karpenter/pkg/utils/binpacking"
 	"github.com/awslabs/karpenter/pkg/utils/resources"
@@ -27,10 +31,14 @@ import (
 )
 
 type podPacker struct {
-	// TODO use this ec2 API to get the instance types
 	ec2 ec2iface.EC2API
 }
 
+type packingResult struct {
+	packed   []*v1.Pod
+	unpacked []*v1.Pod
+}
+
 // Packer helps pack the pods and calculates efficient placement on the instances.
 type Packer interface {
 	// TODO use ctx when calling ec2 API
@@ -60,11 +68,15 @@ func (p *podPacker) Pack(ctx context.Context, constraints *cloudprovider.Constra
 	// Sort pods in decreasing order by the amount of CPU requested, if
 	// CPU requested is equal compare memory requested.
 	sort.Sort(sort.Reverse(binpacking.ByResourcesRequested{SortablePods: constraints.Pods}))
-	packings := []*Packing{}
+	var packings []*Packing
 	var packing *Packing
+	var err error
 	remainingPods := constraints.Pods
 	for len(remainingPods) > 0 {
-		packing, remainingPods = p.packWithLargestPod(remainingPods, constraints)
+		packing, remainingPods, err = p.packWithLargestPod(remainingPods, constraints)
+		if err != nil {
+			return packings, err
+		}
 		// checked all instance type and found no packing option
 		if len(packing.Pods) == 0 {
 			zap.S().Warnf("Failed to find instance type for pod %s/%s ", remainingPods[0].Namespace, remainingPods[0].Name)
@@ -77,33 +89,73 @@ func (p *podPacker) Pack(ctx context.Context, constraints *cloudprovider.Constra
 	return packings, nil
 }
 
-// TODO filter instance types based on node contraints like availability zones etc.
-func (p *podPacker) getNodeCapacities(constraints *cloudprovider.Constraints) []*nodeCapacity {
-	result := make([]*nodeCapacity, 0)
-	for _, nc := range nodeCapacities {
-		ncc := nc.Copy()
-		kubeletOverhead := binpacking.CalculateKubeletOverhead(ncc.total)
-		if ok := ncc.reserve(resources.Merge(constraints.Overhead, kubeletOverhead)); !ok {
-			zap.S().Errorf("Failed to reserve kubelet overhead for node capacity type %v", ncc.instanceType)
+func (p *podPacker) getNodeCapacities(constraints *cloudprovider.Constraints) ([]*nodeCapacity, error) {
+	result := []*nodeCapacity{}
+
+	describeInstanceTypesInput := &ec2.DescribeInstanceTypesInput{
+		Filters: describeInstanceTypesFiltersFrom(constraints),
+	}
+
+	err := p.ec2.DescribeInstanceTypesPagesWithContext(context.TODO(), describeInstanceTypesInput, func(page *ec2.DescribeInstanceTypesOutput, lastPage bool) bool {
+		for _, instanceTypeInfo := range page.InstanceTypes {
+			nc := nodeCapacityFrom(*instanceTypeInfo)
+			kubeletOverhead := binpacking.CalculateKubeletOverhead(nc.total)
+			if ok := nc.reserve(resources.Merge(constraints.Overhead, kubeletOverhead)); !ok {
+				zap.S().Infof("Excluding instance type %s because there are not enough resources for the kubelet overhead", nc.instanceType)
+			}
+			result = append(result, nc)
 		}
-		result = append(result, ncc)
+		return lastPage
+	})
+
+	if err != nil {
+		return nil, fmt.Errorf("fetching instance types using ec2.DescribeInstanceTypes, %w", err)
 	}
-	return result
+	return result, nil
 }
 
-type packingResult struct {
-	packed   []*v1.Pod
-	unpacked []*v1.Pod
+func describeInstanceTypesFiltersFrom(constraints *cloudprovider.Constraints) []*ec2.Filter {
+	architecture := "x86_64"
+	if constraints.Architecture != nil && *constraints.Architecture == provisioningv1alpha1.ArchitectureArm64 {
+		architecture = string(*constraints.Architecture)
+	}
+
+	filters := []*ec2.Filter{
+		{
+			Name:   aws.String("processor-info.supported-architecture"),
+			Values: []*string{&architecture},
+		},
+		{
+			Name:   aws.String("supported-usage-class"),
+			Values: []*string{aws.String("on-demand")},
+		},
+		{
+			Name:   aws.String("supported-virtualization-type"),
+			Values: []*string{aws.String("hvm")},
+		},
+	}
+	if len(constraints.InstanceTypes) != 0 {
+		instanceTypesFilter := &ec2.Filter{
+			Name:   aws.String("instance-type"),
+			Values: aws.StringSlice(constraints.InstanceTypes),
+		}
+		filters = append(filters, instanceTypesFilter)
+	}
+	return filters
 }
 
 // packWithLargestPod will try to pack max number of pods with largest pod in
 // pods across all available node capacities. It returns Packing: max pod count
 // that fit; with their node capacities and list of leftover pods
-func (p *podPacker) packWithLargestPod(unpackedPods []*v1.Pod, constraints *cloudprovider.Constraints) (*Packing, []*v1.Pod) {
+func (p *podPacker) packWithLargestPod(unpackedPods []*v1.Pod, constraints *cloudprovider.Constraints) (*Packing, []*v1.Pod, error) {
 	bestPackedPods := []*v1.Pod{}
 	bestCapacities := []*nodeCapacity{}
 	remainingPods := unpackedPods
-	for _, nc := range p.getNodeCapacities(constraints) {
+	nodeCapacities, err := p.getNodeCapacities(constraints)
+	if err != nil {
+		return nil, nil, err
+	}
+	for _, nc := range nodeCapacities {
 		// check how many pods we can fit with the available capacity
 		result := p.packPodsForCapacity(nc, unpackedPods)
 		if len(result.packed) == 0 {
@@ -125,7 +177,7 @@ func (p *podPacker) packWithLargestPod(unpackedPods []*v1.Pod, constraints *clou
 	for _, capacity := range bestCapacities {
 		capacityNames = append(capacityNames, capacity.instanceType)
 	}
-	return &Packing{Pods: bestPackedPods, InstanceTypes: capacityNames}, remainingPods
+	return &Packing{Pods: bestPackedPods, InstanceTypes: capacityNames}, remainingPods, nil
 }
 
 func (p *podPacker) packPodsForCapacity(capacity *nodeCapacity, pods []*v1.Pod) *packingResult {

pkg/controllers/provisioning/v1alpha1/allocation/constraints.go

@@ -81,7 +81,7 @@ func (c *Constraints) getNodeOverhead(ctx context.Context, node *v1.Node) (v1.Re
 	}
 
 	// 2. filter DaemonSets to include those that will schedule on this node
-	podSpecs := make([]*v1.PodSpec, 0)
+	podSpecs := []*v1.PodSpec{}
 	for _, daemonSet := range daemonSetList.Items {
 		if scheduling.IsSchedulable(&daemonSet.Spec.Template.Spec, node) {
 			podSpecs = append(podSpecs, &daemonSet.Spec.Template.Spec)