Merge pull request #8350 from patrickdillon/capi-restore-etcd

CORS-3471: standalone destroy bootstrap support with capi

Author: openshift-merge-bot[bot]

pkg/clusterapi/localcontrolplane.go

@@ -76,6 +76,11 @@ func (c *localControlPlane) Run(ctx context.Context) error {
 		BinaryAssetsDirectory:    c.BinDir,
 		ControlPlaneStartTimeout: 10 * time.Second,
 		ControlPlaneStopTimeout:  10 * time.Second,
+		ControlPlane: envtest.ControlPlane{
+			Etcd: &envtest.Etcd{
+				DataDir: c.BinDir,
+			},
+		},
 	}
 	var err error
 	c.Cfg, err = c.Env.Start()

pkg/clusterapi/system.go

@@ -17,8 +17,10 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/envtest"
 
+	"github.com/openshift/installer/cmd/openshift-install/command"
 	"github.com/openshift/installer/data"
-	"github.com/openshift/installer/pkg/asset/installconfig"
+	"github.com/openshift/installer/pkg/asset/cluster/metadata"
+	azic "github.com/openshift/installer/pkg/asset/installconfig/azure"
 	gcpic "github.com/openshift/installer/pkg/asset/installconfig/gcp"
 	powervsic "github.com/openshift/installer/pkg/asset/installconfig/powervs"
 	"github.com/openshift/installer/pkg/clusterapi/internal/process"
@@ -49,7 +51,7 @@ const (
 
 // Interface is the interface for the cluster-api system.
 type Interface interface {
-	Run(ctx context.Context, installConfig *installconfig.InstallConfig) error
+	Run(ctx context.Context) error
 	State() SystemState
 	Client() client.Client
 	Teardown()
@@ -78,7 +80,7 @@ type system struct {
 }
 
 // Run launches the cluster-api system.
-func (c *system) Run(ctx context.Context, installConfig *installconfig.InstallConfig) error {
+func (c *system) Run(ctx context.Context) error {
 	c.Lock()
 	defer c.Unlock()
 
@@ -121,9 +123,19 @@ func (c *system) Run(ctx context.Context, installConfig *installconfig.InstallCo
 		},
 	}
 
+	metadata, err := metadata.Load(command.RootOpts.Dir)
+	if err != nil {
+		return fmt.Errorf("failed to load metadata: %w", err)
+	}
+
+	platform := metadata.Platform()
+	if platform == "" {
+		return fmt.Errorf("no platform configured in metadata")
+	}
+
 	// Create the infrastructure controllers.
 	// Only add the controllers for the platform we are deploying to.
-	switch platform := installConfig.Config.Platform.Name(); platform {
+	switch platform {
 	case aws.Name:
 		controller := c.getInfrastructureController(
 			&AWS,
@@ -137,7 +149,7 @@ func (c *system) Run(ctx context.Context, installConfig *installconfig.InstallCo
 			},
 			map[string]string{},
 		)
-		if cfg := installConfig.Config.AWS; cfg != nil && len(cfg.ServiceEndpoints) > 0 {
+		if cfg := metadata.AWS; cfg != nil && len(cfg.ServiceEndpoints) > 0 {
 			endpoints := make([]string, 0, len(cfg.ServiceEndpoints))
 			// CAPA expects name=url pairs of service endpoints
 			for _, endpoint := range cfg.ServiceEndpoints {
@@ -147,9 +159,13 @@ func (c *system) Run(ctx context.Context, installConfig *installconfig.InstallCo
 		}
 		controllers = append(controllers, controller)
 	case azure.Name:
-		session, err := installConfig.Azure.Session()
+		cloudName := metadata.Azure.CloudName
+		if cloudName == "" {
+			cloudName = azure.PublicCloud
+		}
+		session, err := azic.GetSession(cloudName, metadata.Azure.ARMEndpoint)
 		if err != nil {
-			return fmt.Errorf("failed to create azure session: %w", err)
+			return fmt.Errorf("unable to retrieve azure session: %w", err)
 		}
 
 		controllers = append(controllers,

pkg/destroy/bootstrap/bootstrap.go

@@ -84,7 +84,7 @@ func Destroy(ctx context.Context, dir string) (err error) {
 		return fmt.Errorf("error getting infrastructure provider: %w", err)
 	}
 
-	if err := provider.DestroyBootstrap(dir); err != nil {
+	if err := provider.DestroyBootstrap(ctx, dir); err != nil {
 		return fmt.Errorf("error destroying bootstrap resources %w", err)
 	}
 

pkg/infrastructure/aws/sdk/aws.go

@@ -303,7 +303,7 @@ func (a InfraProvider) Provision(ctx context.Context, dir string, parents asset.
 }
 
 // DestroyBootstrap destroys the temporary bootstrap resources.
-func (a InfraProvider) DestroyBootstrap(dir string) error {
+func (a InfraProvider) DestroyBootstrap(ctx context.Context, dir string) error {
 	// Unmarshall input from tf variables, so we can use it along with
 	// installConfig and other assets as the contractual input regardless of
 	// the implementation.
@@ -341,7 +341,7 @@ func (a InfraProvider) DestroyBootstrap(dir string) error {
 		Fn:   request.MakeAddToUserAgentHandler("OpenShift/4.x Destroyer", version.Raw),
 	})
 
-	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Minute)
+	ctx, cancel := context.WithTimeout(ctx, 30*time.Minute)
 	defer cancel()
 
 	logger := logrus.StandardLogger()

pkg/infrastructure/baremetal/baremetal.go

@@ -33,7 +33,7 @@ func (a Provider) Provision(ctx context.Context, dir string, parents asset.Paren
 }
 
 // DestroyBootstrap destroys the temporary bootstrap resources.
-func (a Provider) DestroyBootstrap(dir string) error {
+func (a Provider) DestroyBootstrap(ctx context.Context, dir string) error {
 	config, err := getConfig(dir)
 	if err != nil {
 		return fmt.Errorf("failed to get baremetal platform config: %w", err)

pkg/infrastructure/clusterapi/clusterapi.go

@@ -132,7 +132,7 @@ func (i *InfraProvider) Provision(ctx context.Context, dir string, parents asset
 	// Run the CAPI system.
 	timer.StartTimer(infrastructureStage)
 	capiSystem := clusterapi.System()
-	if err := capiSystem.Run(ctx, installConfig); err != nil {
+	if err := capiSystem.Run(ctx); err != nil {
 		return fileList, fmt.Errorf("failed to run cluster api system: %w", err)
 	}
 
@@ -362,49 +362,54 @@ func (i *InfraProvider) Provision(ctx context.Context, dir string, parents asset
 }
 
 // DestroyBootstrap destroys the temporary bootstrap resources.
-func (i *InfraProvider) DestroyBootstrap(dir string) error {
+func (i *InfraProvider) DestroyBootstrap(ctx context.Context, dir string) error {
 	metadata, err := metadata.Load(dir)
 	if err != nil {
 		return err
 	}
 
-	// TODO(padillon): start system if not running
-	if sys := clusterapi.System(); sys.State() == clusterapi.SystemStateRunning {
-		machineName := capiutils.GenerateBoostrapMachineName(metadata.InfraID)
-		machineNamespace := capiutils.Namespace
-		if err := sys.Client().Delete(context.TODO(), &clusterv1.Machine{
-			ObjectMeta: metav1.ObjectMeta{
-				Name:      machineName,
-				Namespace: machineNamespace,
-			},
-		}); err != nil {
-			return fmt.Errorf("failed to delete bootstrap machine: %w", err)
+	sys := clusterapi.System()
+	if sys.State() != clusterapi.SystemStateRunning {
+		if err := sys.Run(ctx); err != nil {
+			return fmt.Errorf("failed to run capi system: %w", err)
 		}
+	}
 
-		machineDeletionTimeout := 2 * time.Minute
-		logrus.Infof("Waiting up to %v for bootstrap machine deletion %s/%s...", machineDeletionTimeout, machineNamespace, machineName)
-		machineContext, cancel := context.WithTimeout(context.TODO(), machineDeletionTimeout)
-		wait.Until(func() {
-			err := sys.Client().Get(context.TODO(), client.ObjectKey{
-				Name:      machineName,
-				Namespace: machineNamespace,
-			}, &clusterv1.Machine{})
-			if err != nil {
-				if apierrors.IsNotFound(err) {
-					logrus.Debugf("Machine deleted: %s", machineName)
-					cancel()
-				} else {
-					logrus.Debugf("Error when deleting bootstrap machine: %s", err)
-				}
-			}
-		}, 2*time.Second, machineContext.Done())
+	machineName := capiutils.GenerateBoostrapMachineName(metadata.InfraID)
+	machineNamespace := capiutils.Namespace
+	if err := sys.Client().Delete(ctx, &clusterv1.Machine{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      machineName,
+			Namespace: machineNamespace,
+		},
+	}); err != nil {
+		return fmt.Errorf("failed to delete bootstrap machine: %w", err)
+	}
 
-		err = machineContext.Err()
-		if err != nil && !errors.Is(err, context.Canceled) {
-			logrus.Infof("Timeout deleting bootstrap machine: %s", err)
+	machineDeletionTimeout := 5 * time.Minute
+	logrus.Infof("Waiting up to %v for bootstrap machine deletion %s/%s...", machineDeletionTimeout, machineNamespace, machineName)
+	ctx, cancel := context.WithTimeout(ctx, machineDeletionTimeout)
+	wait.UntilWithContext(ctx, func(context.Context) {
+		err := sys.Client().Get(ctx, client.ObjectKey{
+			Name:      machineName,
+			Namespace: machineNamespace,
+		}, &clusterv1.Machine{})
+		if err != nil {
+			if apierrors.IsNotFound(err) {
+				logrus.Debugf("Machine deleted: %s", machineName)
+				cancel()
+			} else {
+				logrus.Debugf("Error when deleting bootstrap machine: %s", err)
+			}
 		}
+	}, 2*time.Second)
+
+	err = ctx.Err()
+	if err != nil && !errors.Is(err, context.Canceled) {
+		logrus.Warnf("Timeout deleting bootstrap machine: %s", err)
+	} else {
+		logrus.Infof("Finished destroying bootstrap resources")
 	}
-	logrus.Infof("Finished destroying bootstrap resources")
 	clusterapi.System().Teardown()
 
 	return nil

pkg/infrastructure/provider.go

@@ -18,7 +18,7 @@ type Provider interface {
 	Provision(ctx context.Context, dir string, parents asset.Parents) ([]*asset.File, error)
 
 	// DestroyBootstrap destroys the temporary bootstrap resources.
-	DestroyBootstrap(dir string) error
+	DestroyBootstrap(ctx context.Context, dir string) error
 
 	// ExtractHostAddresses extracts the IPs of the bootstrap and control plane machines.
 	ExtractHostAddresses(dir string, config *types.InstallConfig, ha *HostAddresses) error

pkg/terraform/terraform.go

@@ -81,7 +81,7 @@ func (p *Provider) Provision(_ context.Context, dir string, parents asset.Parent
 // DestroyBootstrap implements pkg/infrastructure/provider.DestroyBootstrap.
 // DestroyBootstrap iterates through each stage, and will run the destroy
 // command when defined on a stage.
-func (p *Provider) DestroyBootstrap(dir string) error {
+func (p *Provider) DestroyBootstrap(ctx context.Context, dir string) error {
 	varFiles := []string{tfVarsFileName, tfPlatformVarsFileName}
 	for _, stage := range p.stages {
 		varFiles = append(varFiles, stage.OutputsFilename())