Refactor

cli/cmd/cluster.go

@@ -349,7 +349,7 @@ var _clusterConfigureCmd = &cobra.Command{
 		out, exitCode, err := runManagerWithClusterConfig("/root/install.sh --configure", newClusterConfig, awsClient, nil, nil, []string{
 			"CORTEX_NODEGROUP_NAMES_TO_SCALE=" + strings.Join(configureChanges.NodeGroupsToScale, " "),   // NodeGroupsToScale contain the cluster config node-group names
 			"CORTEX_NODEGROUP_NAMES_TO_ADD=" + strings.Join(configureChanges.NodeGroupsToAdd, " "),       // NodeGroupsToAdd contain the cluster config node-group names
-			"CORTEX_NODEGROUP_NAMES_TO_REMOVE=" + strings.Join(configureChanges.NodeGroupsToRemove, " "), // NodeGroupsToRemove contain the EKS node-group names
+			"CORTEX_NODEGROUP_NAMES_TO_REMOVE=" + strings.Join(configureChanges.StaleEKSNodeGroups, " "), // StaleEKSNodeGroups contain the EKS node-group names
 		})
 		if err != nil {
 			exit.Error(err)

cli/cmd/lib_cluster_config.go

@@ -31,7 +31,6 @@ import (
 	libmath "github.com/cortexlabs/cortex/pkg/lib/math"
 	"github.com/cortexlabs/cortex/pkg/lib/pointer"
 	"github.com/cortexlabs/cortex/pkg/lib/prompt"
-	"github.com/cortexlabs/cortex/pkg/lib/slices"
 	s "github.com/cortexlabs/cortex/pkg/lib/strings"
 	"github.com/cortexlabs/cortex/pkg/lib/table"
 	"github.com/cortexlabs/cortex/pkg/types/clusterconfig"
@@ -158,22 +157,12 @@ func getConfigureClusterConfig(awsClient *aws.Client, stacks clusterstate.Cluste
 	newUserClusterConfig.Telemetry = isTelemetryEnabled()
 	cachedClusterConfig.Telemetry = newUserClusterConfig.Telemetry
 
-	configureChanges, err := newUserClusterConfig.ValidateOnConfigure(awsClient, cachedClusterConfig)
+	configureChanges, err := newUserClusterConfig.ValidateOnConfigure(awsClient, cachedClusterConfig, stacks.NodeGroupsStacks)
 	if err != nil {
 		err = errors.Append(err, fmt.Sprintf("\n\ncluster configuration schema can be found at https://docs.cortex.dev/v/%s/", consts.CortexVersionMinor))
 		return nil, clusterconfig.ConfigureChanges{}, errors.Wrap(err, newClusterConfigFile)
 	}
 
-	// intersect with the stale eks node groups
-	eksNodeGroupsToRemove := []string{}
-	staleEKSNgs, staleEKSNgAvailabilities := stacks.GetStaleNodeGroupNames(*newUserClusterConfig)
-	for i := range staleEKSNgs {
-		if slices.HasString(configureChanges.NodeGroupsToRemove, staleEKSNgs[i]) {
-			eksNodeGroupsToRemove = append(eksNodeGroupsToRemove, clusterstate.GetStackName(newUserClusterConfig.ClusterName, staleEKSNgAvailabilities[i], staleEKSNgs[i]))
-		}
-	}
-	configureChanges.NodeGroupsToRemove = eksNodeGroupsToRemove
-
 	confirmConfigureClusterConfig(configureChanges, cachedClusterConfig, *newUserClusterConfig, _flagClusterDisallowPrompt)
 
 	return newUserClusterConfig, configureChanges, nil

pkg/types/clusterconfig/cluster_config.go

@@ -28,6 +28,7 @@ import (
 	"time"
 
 	"github.com/aws/amazon-vpc-cni-k8s/pkg/awsutils"
+	"github.com/aws/aws-sdk-go/service/cloudformation"
 	"github.com/aws/aws-sdk-go/service/iam"
 	"github.com/cortexlabs/cortex/pkg/consts"
 	"github.com/cortexlabs/cortex/pkg/lib/aws"
@@ -190,10 +191,11 @@ type ConfigureChanges struct {
 	NodeGroupsToAdd    []string
 	NodeGroupsToRemove []string
 	NodeGroupsToScale  []string
+	StaleEKSNodeGroups []string
 }
 
 func (c *ConfigureChanges) HasChanges() bool {
-	return len(c.NodeGroupsToAdd) != 0 || len(c.NodeGroupsToRemove) != 0 || len(c.NodeGroupsToScale) != 0
+	return len(c.NodeGroupsToAdd) != 0 || len(c.NodeGroupsToRemove) != 0 || len(c.NodeGroupsToScale) != 0 || len(c.StaleEKSNodeGroups) != 0
 }
 
 // NewForFile initializes and validates the cluster config from the YAML config file
@@ -1101,7 +1103,7 @@ func (cc *Config) ValidateOnInstall(awsClient *aws.Client) error {
 	return nil
 }
 
-func (cc *Config) ValidateOnConfigure(awsClient *aws.Client, oldConfig Config) (ConfigureChanges, error) {
+func (cc *Config) ValidateOnConfigure(awsClient *aws.Client, oldConfig Config, eksNodeGroupStacks []*cloudformation.StackSummary) (ConfigureChanges, error) {
 	fmt.Print("verifying your configuration ...\n\n")
 
 	cc.ClusterUID = oldConfig.ClusterUID
@@ -1144,6 +1146,7 @@ func (cc *Config) ValidateOnConfigure(awsClient *aws.Client, oldConfig Config) (
 		NodeGroupsToAdd:    GetNodeGroupNames(ngsToBeAdded),
 		NodeGroupsToRemove: GetNodeGroupNames(ngsToBeRemoved),
 		NodeGroupsToScale:  GetNodeGroupNames(ngNamesToBeScaled),
+		StaleEKSNodeGroups: getStaleEksNodeGroups(cc.ClusterName, eksNodeGroupStacks, ngsToBeRemoved),
 	}, nil
 }
 
@@ -1523,6 +1526,30 @@ func (ng *NodeGroup) SpotConfigOnDemandValues() (int64, int64) {
 	return onDemandBaseCapacity, onDemandPercentageAboveBaseCapacity
 }
 
+func getStaleEksNodeGroups(clusterName string, eksNodeGroupStacks []*cloudformation.StackSummary, ngsMarkedForRemoval []*NodeGroup) []string {
+	eksNodeGroupsToRemove := []string{}
+	for _, ng := range ngsMarkedForRemoval {
+		availability := "d"
+		if ng.Spot {
+			availability = "s"
+		}
+
+		eksNgName := fmt.Sprintf("cx-w%s-%s", availability, ng.Name)
+		eksStackName := fmt.Sprintf("eksctl-%s-nodegroup-cx-w%s-%s", clusterName, availability, ng.Name)
+		for _, eksNgStack := range eksNodeGroupStacks {
+			if eksNgStack == nil || eksNgStack.StackName == nil {
+				continue
+			}
+			if *eksNgStack.StackName == eksStackName {
+				eksNodeGroupsToRemove = append(eksNodeGroupsToRemove, eksNgName)
+				break
+			}
+		}
+	}
+
+	return eksNodeGroupsToRemove
+}
+
 func (cc *CoreConfig) TelemetryEvent() map[string]interface{} {
 	event := make(map[string]interface{})