Add parameter provisioningTimeout to Kubernetes driver options.

Signed-off-by: Arnold Sobanski <arnold@l4g.dev>

driver/kubernetes/driver.go

@@ -41,15 +41,16 @@ type Driver struct {
 
 	// if you add fields, remember to update docs:
 	// https://github.com/docker/docs/blob/main/content/build/drivers/kubernetes.md
-	minReplicas      int
-	deployment       *appsv1.Deployment
-	configMaps       []*corev1.ConfigMap
-	clientset        *kubernetes.Clientset
-	deploymentClient clientappsv1.DeploymentInterface
-	podClient        clientcorev1.PodInterface
-	configMapClient  clientcorev1.ConfigMapInterface
-	podChooser       podchooser.PodChooser
-	defaultLoad      bool
+	minReplicas         int
+	deployment          *appsv1.Deployment
+	configMaps          []*corev1.ConfigMap
+	clientset           *kubernetes.Clientset
+	deploymentClient    clientappsv1.DeploymentInterface
+	podClient           clientcorev1.PodInterface
+	configMapClient     clientcorev1.ConfigMapInterface
+	podChooser          podchooser.PodChooser
+	defaultLoad         bool
+	provisioningTimeout time.Duration
 }
 
 func (d *Driver) IsMobyDriver() bool {
@@ -88,7 +89,7 @@ func (d *Driver) Bootstrap(ctx context.Context, l progress.Logger) error {
 			}
 		}
 		return sub.Wrap(
-			fmt.Sprintf("waiting for %d pods to be ready", d.minReplicas),
+			fmt.Sprintf("waiting for %d pods to be ready, timeout: %ds", d.minReplicas, d.provisioningTimeout/time.Second),
 			func() error {
 				return d.wait(ctx)
 			})
@@ -101,22 +102,33 @@ func (d *Driver) wait(ctx context.Context) error {
 		err  error
 		depl *appsv1.Deployment
 	)
-	for try := 0; try < 100; try++ {
-		depl, err = d.deploymentClient.Get(ctx, d.deployment.Name, metav1.GetOptions{})
-		if err == nil {
-			if depl.Status.ReadyReplicas >= int32(d.minReplicas) {
-				return nil
-			}
-			err = errors.Errorf("expected %d replicas to be ready, got %d",
-				d.minReplicas, depl.Status.ReadyReplicas)
-		}
+	timeoutChan := time.After(d.provisioningTimeout)
+	ticker := time.NewTicker(100 * time.Millisecond)
+	tryCounter := 0
+	defer ticker.Stop()
+
+	for {
 		select {
+		case <-ticker.C:
+			if tryCounter < 100 {
+				tryCounter++
+			}
+			ticker.Stop()
+			ticker = time.NewTicker(time.Duration(100+tryCounter*20) * time.Millisecond)
+			depl, err = d.deploymentClient.Get(ctx, d.deployment.Name, metav1.GetOptions{})
+			if err == nil {
+				if depl.Status.ReadyReplicas >= int32(d.minReplicas) {
+					return nil
+				}
+				err = errors.Errorf("expected %d replicas to be ready, got %d", d.minReplicas, depl.Status.ReadyReplicas)
+			}
+		case <-timeoutChan:
+			return errors.Errorf("timeout after %s: last error: %v", d.provisioningTimeout, err)
 		case <-ctx.Done():
 			return ctx.Err()
-		case <-time.After(time.Duration(100+try*20) * time.Millisecond):
 		}
 	}
-	return err
+
 }
 
 func (d *Driver) Info(ctx context.Context) (*driver.Info, error) {

driver/kubernetes/factory.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"strconv"
 	"strings"
+	"time"
 
 	corev1 "k8s.io/api/core/v1"
 
@@ -82,6 +83,8 @@ func (f *factory) New(ctx context.Context, cfg driver.InitConfig) (driver.Driver
 
 	d.minReplicas = deploymentOpt.Replicas
 
+	d.provisioningTimeout = deploymentOpt.ProvisioningTimeout
+
 	d.deploymentClient = clientset.AppsV1().Deployments(namespace)
 	d.podClient = clientset.CoreV1().Pods(namespace)
 	d.configMapClient = clientset.CoreV1().ConfigMaps(namespace)
@@ -104,13 +107,14 @@ func (f *factory) New(ctx context.Context, cfg driver.InitConfig) (driver.Driver
 
 func (f *factory) processDriverOpts(deploymentName string, namespace string, cfg driver.InitConfig) (*manifest.DeploymentOpt, string, string, bool, error) {
 	deploymentOpt := &manifest.DeploymentOpt{
-		Name:          deploymentName,
-		Image:         bkimage.DefaultImage,
-		Replicas:      1,
-		BuildkitFlags: cfg.BuildkitdFlags,
-		Rootless:      false,
-		Platforms:     cfg.Platforms,
-		ConfigFiles:   cfg.Files,
+		Name:                deploymentName,
+		Image:               bkimage.DefaultImage,
+		Replicas:            1,
+		ProvisioningTimeout: 120 * time.Second,
+		BuildkitFlags:       cfg.BuildkitdFlags,
+		Rootless:            false,
+		Platforms:           cfg.Platforms,
+		ConfigFiles:         cfg.Files,
 	}
 
 	defaultLoad := false
@@ -229,6 +233,8 @@ func (f *factory) processDriverOpts(deploymentName string, namespace string, cfg
 			if err != nil {
 				return nil, "", "", false, err
 			}
+		case "provisioningTimeout":
+			deploymentOpt.ProvisioningTimeout, err = time.ParseDuration(v)
 		default:
 			return nil, "", "", false, errors.Errorf("invalid driver option %s for driver %s", k, DriverName)
 		}

driver/kubernetes/factory_test.go

@@ -2,6 +2,7 @@ package kubernetes
 
 import (
 	"testing"
+	"time"
 
 	"github.com/docker/buildx/driver"
 	"github.com/docker/buildx/driver/bkimage"
@@ -37,22 +38,23 @@ func TestFactory_processDriverOpts(t *testing.T) {
 	t.Run(
 		"ValidOptions", func(t *testing.T) {
 			cfg.DriverOpts = map[string]string{
-				"namespace":       "test-ns",
-				"image":           "test:latest",
-				"replicas":        "2",
-				"requests.cpu":    "100m",
-				"requests.memory": "32Mi",
-				"limits.cpu":      "200m",
-				"limits.memory":   "64Mi",
-				"rootless":        "true",
-				"nodeselector":    "selector1=value1,selector2=value2",
-				"tolerations":     "key=tolerationKey1,value=tolerationValue1,operator=Equal,effect=NoSchedule,tolerationSeconds=60;key=tolerationKey2,operator=Exists",
-				"annotations":     "example.com/expires-after=annotation1,example.com/other=annotation2",
-				"labels":          "example.com/owner=label1,example.com/other=label2",
-				"loadbalance":     "random",
-				"qemu.install":    "true",
-				"qemu.image":      "qemu:latest",
-				"default-load":    "true",
+				"namespace":           "test-ns",
+				"image":               "test:latest",
+				"replicas":            "2",
+				"provisioningTimeout": "300s",
+				"requests.cpu":        "100m",
+				"requests.memory":     "32Mi",
+				"limits.cpu":          "200m",
+				"limits.memory":       "64Mi",
+				"rootless":            "true",
+				"nodeselector":        "selector1=value1,selector2=value2",
+				"tolerations":         "key=tolerationKey1,value=tolerationValue1,operator=Equal,effect=NoSchedule,tolerationSeconds=60;key=tolerationKey2,operator=Exists",
+				"annotations":         "example.com/expires-after=annotation1,example.com/other=annotation2",
+				"labels":              "example.com/owner=label1,example.com/other=label2",
+				"loadbalance":         "random",
+				"qemu.install":        "true",
+				"qemu.image":          "qemu:latest",
+				"default-load":        "true",
 			}
 			r, loadbalance, ns, defaultLoad, err := f.processDriverOpts(cfg.Name, "test", cfg)
 
@@ -91,6 +93,7 @@ func TestFactory_processDriverOpts(t *testing.T) {
 			require.Equal(t, "test-ns", ns)
 			require.Equal(t, "test:latest", r.Image)
 			require.Equal(t, 2, r.Replicas)
+			require.Equal(t, 300*time.Second, r.ProvisioningTimeout)
 			require.Equal(t, "100m", r.RequestsCPU)
 			require.Equal(t, "32Mi", r.RequestsMemory)
 			require.Equal(t, "200m", r.LimitsCPU)
@@ -118,6 +121,7 @@ func TestFactory_processDriverOpts(t *testing.T) {
 			require.Equal(t, "test", ns)
 			require.Equal(t, bkimage.DefaultImage, r.Image)
 			require.Equal(t, 1, r.Replicas)
+			require.Equal(t, 120*time.Second, r.ProvisioningTimeout)
 			require.Equal(t, "", r.RequestsCPU)
 			require.Equal(t, "", r.RequestsMemory)
 			require.Equal(t, "", r.LimitsCPU)
@@ -148,6 +152,7 @@ func TestFactory_processDriverOpts(t *testing.T) {
 			require.Equal(t, "test", ns)
 			require.Equal(t, bkimage.DefaultRootlessImage, r.Image)
 			require.Equal(t, 1, r.Replicas)
+			require.Equal(t, 120*time.Second, r.ProvisioningTimeout)
 			require.Equal(t, "", r.RequestsCPU)
 			require.Equal(t, "", r.RequestsMemory)
 			require.Equal(t, "", r.LimitsCPU)
@@ -174,6 +179,16 @@ func TestFactory_processDriverOpts(t *testing.T) {
 		},
 	)
 
+	t.Run(
+		"InvalidProvisioningTimeout", func(t *testing.T) {
+			cfg.DriverOpts = map[string]string{
+				"ProvisioningTimeout": "invalid",
+			}
+			_, _, _, _, err := f.processDriverOpts(cfg.Name, "test", cfg)
+			require.Error(t, err)
+		},
+	)
+
 	t.Run(
 		"InvalidRootless", func(t *testing.T) {
 			cfg.DriverOpts = map[string]string{

driver/kubernetes/manifest/manifest.go

@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"path"
 	"strings"
+	"time"
 
 	"github.com/docker/buildx/util/platformutil"
 	v1 "github.com/opencontainers/image-spec/specs-go/v1"
@@ -15,12 +16,13 @@ import (
 )
 
 type DeploymentOpt struct {
-	Namespace          string
-	Name               string
-	Image              string
-	Replicas           int
-	ServiceAccountName string
-	SchedulerName      string
+	Namespace           string
+	Name                string
+	Image               string
+	Replicas            int
+	ProvisioningTimeout time.Duration
+	ServiceAccountName  string
+	SchedulerName       string
 
 	// Qemu
 	Qemu struct {