fix: try to find suitable AZ for EFA, honor capacity-reservation for …

…non-EFA (#588) * combine nodegroup stacks for easier maintenance * preserve comment Co-authored-by: Nick Baker <[email protected]> * remove redundant instance type, explicit error for efa * Always create network interface * Add instance-types flag requirement to EFA flag doc * fix non-efa interface * fix creation timeout var * Fix non-EFA network interface case * Undo extra EFA validation * Remove EFA enabled parameter * Force EFA ASGs to use the same subnet * Isolate subnet & az selection for nodes * force nil pointer if networkcardindex empty, only render interfaces if passed --------- Co-authored-by: Nick Baker <[email protected]>
aws · Feb 27, 2025 · ecc47a9 · ecc47a9
1 parent 5e2e5c1
commit ecc47a9
Show file tree

Hide file tree

Showing 6 changed files with 266 additions and 453 deletions.
diff --git a/internal/deployers/eksapi/deployer.go b/internal/deployers/eksapi/deployer.go
@@ -62,7 +62,7 @@ type deployerOptions struct {
 	CapacityReservation         bool          `flag:"capacity-reservation" desc:"Use capacity reservation for the unmanaged nodegroup"`
 	ClusterCreationTimeout      time.Duration `flag:"cluster-creation-timeout" desc:"Time to wait for cluster to be created and become active."`
 	ClusterRoleServicePrincipal string        `flag:"cluster-role-service-principal" desc:"Additional service principal that can assume the cluster role"`
-	EFA                         bool          `flag:"efa" desc:"Create EFA interfaces on the node of an unmanaged nodegroup. Requires --unmanaged-nodes."`
+	EFA                         bool          `flag:"efa" desc:"Create EFA interfaces on the node of an unmanaged nodegroup. One instance type must be passed if set. Requires --unmanaged-nodes and --instance-types."`
 	EKSEndpointURL              string        `flag:"endpoint-url" desc:"Endpoint URL for the EKS API"`
 	EmitMetrics                 bool          `flag:"emit-metrics" desc:"Record and emit metrics to CloudWatch"`
 	ExpectedAMI                 string        `flag:"expected-ami" desc:"Expected AMI of nodes. Up will fail if the actual nodes are not utilizing the expected AMI. Defaults to --ami if defined."`

diff --git a/internal/deployers/eksapi/infra.go b/internal/deployers/eksapi/infra.go
@@ -73,19 +73,21 @@ func NewInfrastructureManager(clients *awsClients, resourceID string, metrics me
 }
 
 type Infrastructure struct {
-	vpc            string
-	subnetsPublic  []string
-	subnetsPrivate []string
-	clusterRoleARN string
-	nodeRoleARN    string
-	nodeRoleName   string
+	availabilityZones []string
+	vpc               string
+	subnetsPublic     []string
+	subnetsPrivate    []string
+	clusterRoleARN    string
+	nodeRoleARN       string
+	nodeRoleName      string
 }
 
 func (i *Infrastructure) subnets() []string {
 	return append(i.subnetsPublic, i.subnetsPrivate...)
 }
 
 func (m *InfrastructureManager) createInfrastructureStack(opts *deployerOptions) (*Infrastructure, error) {
+	// TODO: create a subnet in every AZ
 	// get two AZs for the subnets
 	azs, err := m.clients.EC2().DescribeAvailabilityZones(context.TODO(), &ec2.DescribeAvailabilityZonesInput{})
 	if err != nil {
@@ -165,6 +167,7 @@ func (m *InfrastructureManager) createInfrastructureStack(opts *deployerOptions)
 	}
 	klog.Infof("getting infrastructure stack resources: %s", *out.StackId)
 	infra, err := m.getInfrastructureStackResources()
+	infra.availabilityZones = subnetAzs
 	if err != nil {
 		return nil, fmt.Errorf("failed to get infrastructure stack resources: %w", err)
 	}