Add logic to pick 3 AZs

pkg/eks/api.go

@@ -2,6 +2,7 @@ package eks
 
 import (
 	"fmt"
+	"math/rand"
 	"os"
 	"sync"
 	"time"
@@ -25,8 +26,9 @@ import (
 )
 
 const (
-	ClusterNameTag = "eksctl.cluster.k8s.io/v1alpha1/cluster-name"
-	AWSDebugLevel  = 5
+	ClusterNameTag            = "eksctl.cluster.k8s.io/v1alpha1/cluster-name"
+	AWSDebugLevel             = 5
+	RequiredAvailabilityZones = 3
 )
 
 var DefaultWaitTimeout = 20 * time.Minute
@@ -35,10 +37,9 @@ type ClusterProvider struct {
 	// core fields used for config and AWS APIs
 	cfg *ClusterConfig
 	svc *providerServices
-
 	// informative fields, i.e. used as outputs
-	iamRoleARN   string
-	sessionCreds *credentials.Credentials
+	// TODO refactor this as ProviderStatus
+	status providerStatus
 }
 
 type providerServices struct {
@@ -48,6 +49,12 @@ type providerServices struct {
 	sts stsiface.STSAPI
 }
 
+type providerStatus struct {
+	iamRoleARN        string
+	sessionCreds      *credentials.Credentials
+	availabilityZones []string
+}
+
 // simple config, to be replaced with Cluster API
 type ClusterConfig struct {
 	Region      string
@@ -101,36 +108,38 @@ func New(clusterConfig *ClusterConfig) *ClusterProvider {
 			ec2: ec2.New(s),
 			sts: sts.New(s),
 		},
-		sessionCreds: s.Config.Credentials,
+		status: providerStatus{
+			sessionCreds: s.Config.Credentials,
+		},
 	}
 
 	// override sessions if any custom endpoints specified
 	if endpoint, ok := os.LookupEnv("AWS_CLOUDFORMATION_ENDPOINT"); ok {
 		logger.Debug("Setting CloudFormation endpoint to %s", endpoint)
-		s := newSession(clusterConfig, endpoint, c.sessionCreds)
+		s := newSession(clusterConfig, endpoint, c.status.sessionCreds)
 		c.svc.cfn = cloudformation.New(s)
 	}
 	if endpoint, ok := os.LookupEnv("AWS_EKS_ENDPOINT"); ok {
 		logger.Debug("Setting EKS endpoint to %s", endpoint)
-		s := newSession(clusterConfig, endpoint, c.sessionCreds)
+		s := newSession(clusterConfig, endpoint, c.status.sessionCreds)
 		c.svc.eks = eks.New(s)
 	}
 	if endpoint, ok := os.LookupEnv("AWS_EC2_ENDPOINT"); ok {
 		logger.Debug("Setting EC2 endpoint to %s", endpoint)
-		s := newSession(clusterConfig, endpoint, c.sessionCreds)
+		s := newSession(clusterConfig, endpoint, c.status.sessionCreds)
 		c.svc.ec2 = ec2.New(s)
 	}
 	if endpoint, ok := os.LookupEnv("AWS_STS_ENDPOINT"); ok {
 		logger.Debug("Setting STS endpoint to %s", endpoint)
-		s := newSession(clusterConfig, endpoint, c.sessionCreds)
+		s := newSession(clusterConfig, endpoint, c.status.sessionCreds)
 		c.svc.sts = sts.New(s)
 	}
 
 	return c
 }
 
 func (c *ClusterProvider) GetCredentialsEnv() ([]string, error) {
-	creds, err := c.sessionCreds.Get()
+	creds, err := c.status.sessionCreds.Get()
 	if err != nil {
 		return nil, errors.Wrap(err, "getting effective credentials")
 	}
@@ -148,8 +157,8 @@ func (c *ClusterProvider) CheckAuth() error {
 		if err != nil {
 			return errors.Wrap(err, "checking AWS STS access – cannot get role ARN for current session")
 		}
-		c.iamRoleARN = *output.Arn
-		logger.Debug("role ARN for the current session is %q", c.iamRoleARN)
+		c.status.iamRoleARN = *output.Arn
+		logger.Debug("role ARN for the current session is %q", c.status.iamRoleARN)
 	}
 	{
 		input := &cloudformation.ListStacksInput{}
@@ -160,6 +169,41 @@ func (c *ClusterProvider) CheckAuth() error {
 	return nil
 }
 
+func (c *ClusterProvider) SelectAvailabilityZones() error {
+	avoidZones := map[string]bool{
+		// well-known over-populated zones
+		"us-east1-a": true,
+		"us-east1-b": true,
+	}
+
+	input := &ec2.DescribeAvailabilityZonesInput{}
+	output, err := c.svc.ec2.DescribeAvailabilityZones(input)
+	if err != nil {
+		return errors.Wrapf(err, "getting availibility zones for %s", c.cfg.Region)
+	}
+
+	usableZones := []string{}
+	for _, zone := range output.AvailabilityZones {
+		_, avoidZone := avoidZones[*zone.ZoneName]
+		if !avoidZone && *zone.State == ec2.AvailabilityZoneStateAvailable {
+			usableZones = append(usableZones, *zone.ZoneName)
+		}
+	}
+
+	zones := []string{}
+	for len(zones) < RequiredAvailabilityZones {
+		rand := rand.New(rand.NewSource(time.Now().UnixNano()))
+		for _, r := range rand.Perm(len(usableZones)) {
+			zones = append(zones, usableZones[r])
+			if len(zones) == RequiredAvailabilityZones {
+				break
+			}
+		}
+	}
+	c.status.availabilityZones = zones
+	return nil
+}
+
 func (c *ClusterProvider) runCreateTask(tasks map[string]func(chan error) error, taskErrs chan error) {
 	wg := &sync.WaitGroup{}
 	wg.Add(len(tasks))

pkg/eks/assets/1.10.3/2018-06-05/amazon-eks-vpc-sample.yaml

@@ -9,6 +9,9 @@ Parameters:
     Default: 192.168.0.0/16
     Description: The CIDR range for the VPC. This should be a valid private (RFC 1918) CIDR range.
 
+ AvailabilityZones:
+    Type: List<AWS::EC2::AvailabilityZone::Name>
+
   Subnet01Block:
     Type: String
     Default: 192.168.64.0/18
@@ -82,8 +85,7 @@ Resources:
       AvailabilityZone:
         Fn::Select:
         - '0'
-        - Fn::GetAZs:
-            Ref: AWS::Region
+        - !Ref AvailabilityZones
       CidrBlock:
         Ref: Subnet01Block
       VpcId:
@@ -100,8 +102,7 @@ Resources:
       AvailabilityZone:
         Fn::Select:
         - '1'
-        - Fn::GetAZs:
-            Ref: AWS::Region
+        - !Ref AvailabilityZones
       CidrBlock:
         Ref: Subnet02Block
       VpcId:
@@ -118,8 +119,7 @@ Resources:
       AvailabilityZone:
         Fn::Select:
         - '2'
-        - Fn::GetAZs:
-            Ref: AWS::Region
+        - !Ref AvailabilityZones
       CidrBlock:
         Ref: Subnet03Block
       VpcId:

pkg/eks/auth.go

@@ -135,7 +135,7 @@ func (c *ClusterProvider) MaybeDeletePublicSSHKey() {
 }
 
 func (c *ClusterProvider) getUsername() string {
-	usernameParts := strings.Split(c.iamRoleARN, "/")
+	usernameParts := strings.Split(c.status.iamRoleARN, "/")
 	if len(usernameParts) > 1 {
 		return usernameParts[len(usernameParts)-1]
 	}
@@ -179,7 +179,7 @@ func (c *ClusterProvider) NewClientConfig() (*ClientConfig, error) {
 		},
 		ClusterName: clusterName,
 		ContextName: contextName,
-		roleARN:     c.iamRoleARN,
+		roleARN:     c.status.iamRoleARN,
 		sts:         c.svc.sts,
 	}
 

pkg/eks/cfn.go

@@ -151,6 +151,13 @@ func (c *ClusterProvider) stackNameVPC() string {
 	return "EKS-" + c.cfg.ClusterName + "-VPC"
 }
 
+func (c *ClusterProvider) stackParamsVPC() map[string]string {
+	params := map[string]string{
+		"AvailabilityZones": strings.Join(c.status.availabilityZones, ","),
+	}
+	return params
+}
+
 func (c *ClusterProvider) createStackVPC(errs chan error) error {
 	name := c.stackNameVPC()
 	logger.Info("creating VPC stack %q", name)
@@ -162,7 +169,7 @@ func (c *ClusterProvider) createStackVPC(errs chan error) error {
 	stackChan := make(chan Stack)
 	taskErrs := make(chan error)
 
-	if err := c.CreateStack(name, templateBody, nil, false, stackChan, taskErrs); err != nil {
+	if err := c.CreateStack(name, templateBody, c.stackParamsVPC(), false, stackChan, taskErrs); err != nil {
 		return err
 	}