pkg/platform/aks: implement PostApplyHook

This commit implements newly introduced platform.PostApplyHook for AKS
clusters, to address issue, where components depending on the storage
gets installed when default storage class has not been yet created by
the AKS controller, as this causes components to get stuck, which makes
cluster provisioning to fail.

The implemented hook lists available storage classes on the cluster and
returns when class with default storage class annotation is found.
Usually default storage class appears on the cluster within 5 minutes
after cluster creation, so 10 minutes timeout seems like a sane default
for this operation.

Closes #855

Signed-off-by: Mateusz Gozdek <mateusz@kinvolk.io>

Author: invidian

pkg/platform/aks/aks.go

@@ -17,15 +17,21 @@
 package aks
 
 import (
+	"context"
 	"fmt"
 	"os"
 	"path/filepath"
 	"text/template"
+	"time"
 
 	"github.com/hashicorp/hcl/v2"
 	"github.com/hashicorp/hcl/v2/gohcl"
 	"github.com/mitchellh/go-homedir"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/wait"
+	v1 "k8s.io/client-go/kubernetes/typed/storage/v1"
 
+	"github.com/kinvolk/lokomotive/pkg/k8sutil"
 	"github.com/kinvolk/lokomotive/pkg/platform"
 	"github.com/kinvolk/lokomotive/pkg/terraform"
 )
@@ -309,6 +315,48 @@ func (c *config) Initialize(ex *terraform.Executor) error {
 	return createTerraformConfigFile(c, terraformRootDir)
 }
 
+const (
+	retryInterval = 5 * time.Second
+	timeout       = 10 * time.Minute
+)
+
+// PostApplyHook implements platform.PostApplyHook interface and defines hooks
+// which should be executed after AKS cluster is created.
+func (c *config) PostApplyHook(kubeconfig []byte) error {
+	client, err := k8sutil.NewClientset(kubeconfig)
+	if err != nil {
+		return fmt.Errorf("creating clientset from kubeconfig: %w", err)
+	}
+
+	return waitForDefaultStorageClass(client.StorageV1().StorageClasses())
+}
+
+// waitForDefaultStorageClass wait until default storage class appears on a given cluster.
+// If it doesn't within defined time range, error is returned.
+func waitForDefaultStorageClass(sci v1.StorageClassInterface) error {
+	// AKS still uses annotation with .beta.
+	defaultStorageClassAnnotation := "storageclass.beta.kubernetes.io/is-default-class"
+
+	if err := wait.PollImmediate(retryInterval, timeout, func() (done bool, err error) {
+		scs, err := sci.List(context.TODO(), metav1.ListOptions{})
+		if err != nil {
+			return false, fmt.Errorf("getting storage classes %w", err)
+		}
+
+		for _, sc := range scs.Items {
+			if v, ok := sc.ObjectMeta.Annotations[defaultStorageClassAnnotation]; ok && v == "true" {
+				return true, nil
+			}
+		}
+
+		return false, nil
+	}); err != nil {
+		return fmt.Errorf("waiting for the default storage class to be configured: %w", err)
+	}
+
+	return nil
+}
+
 // createTerraformConfigFiles create Terraform config files in given directory.
 func createTerraformConfigFile(cfg *config, terraformRootDir string) error {
 	t := template.Must(template.New("t").Parse(terraformConfigTmpl))