cli: add "--skip-helm-wait" flag (#2061)

* cli: add "--skip-helm-wait" flag

This flag can be used to disable the atomic and wait flags during helm install.
This is useful when debugging a failing constellation init, since the user gains access to
the cluster even when one of the deployments is never in a ready state.

bootstrapper/internal/helm/helm.go

@@ -35,6 +35,8 @@ import (
 const (
 	// timeout is the maximum time given to the helm client.
 	timeout = 5 * time.Minute
+	// maximumRetryAttempts is the maximum number of attempts to retry a helm install.
+	maximumRetryAttempts = 3
 )
 
 // Client is used to install microservice during cluster initialization. It is a wrapper for a helm install action.
@@ -57,8 +59,6 @@ func New(log *logger.Logger) (*Client, error) {
 	action := action.NewInstall(actionConfig)
 	action.Namespace = constants.HelmNamespace
 	action.Timeout = timeout
-	action.Atomic = true
-	action.Wait = true
 
 	return &Client{
 		action,
@@ -69,6 +69,9 @@ func New(log *logger.Logger) (*Client, error) {
 // InstallConstellationServices installs the constellation-services chart. In the future this chart should bundle all microservices.
 func (h *Client) InstallConstellationServices(ctx context.Context, release helm.Release, extraVals map[string]any) error {
 	h.ReleaseName = release.ReleaseName
+	if err := h.setWaitMode(release.WaitMode); err != nil {
+		return err
+	}
 
 	mergedVals := helm.MergeMaps(release.Values, extraVals)
 
@@ -79,6 +82,9 @@ func (h *Client) InstallConstellationServices(ctx context.Context, release helm.
 func (h *Client) InstallCertManager(ctx context.Context, release helm.Release) error {
 	h.ReleaseName = release.ReleaseName
 	h.Timeout = 10 * time.Minute
+	if err := h.setWaitMode(release.WaitMode); err != nil {
+		return err
+	}
 
 	return h.install(ctx, release.Chart, release.Values)
 }
@@ -86,6 +92,9 @@ func (h *Client) InstallCertManager(ctx context.Context, release helm.Release) e
 // InstallOperators installs the Constellation Operators.
 func (h *Client) InstallOperators(ctx context.Context, release helm.Release, extraVals map[string]any) error {
 	h.ReleaseName = release.ReleaseName
+	if err := h.setWaitMode(release.WaitMode); err != nil {
+		return err
+	}
 
 	mergedVals := helm.MergeMaps(release.Values, extraVals)
 
@@ -95,6 +104,9 @@ func (h *Client) InstallOperators(ctx context.Context, release helm.Release, ext
 // InstallCilium sets up the cilium pod network.
 func (h *Client) InstallCilium(ctx context.Context, kubectl k8sapi.Client, release helm.Release, in k8sapi.SetupPodNetworkInput) error {
 	h.ReleaseName = release.ReleaseName
+	if err := h.setWaitMode(release.WaitMode); err != nil {
+		return err
+	}
 
 	timeoutS := int64(10)
 	// allow coredns to run on uninitialized nodes (required by cloud-controller-manager)
@@ -164,9 +176,22 @@ func (h *Client) installCiliumGCP(ctx context.Context, release helm.Release, nod
 
 // install tries to install the given chart and aborts after ~5 tries.
 // The function will wait 30 seconds before retrying a failed installation attempt.
-// After 10 minutes the retrier will be canceled and the function returns with an error.
+// After 3 tries, the retrier will be canceled and the function returns with an error.
 func (h *Client) install(ctx context.Context, chartRaw []byte, values map[string]any) error {
+	var retries int
 	retriable := func(err error) bool {
+		// abort after maximumRetryAttempts tries.
+		if retries >= maximumRetryAttempts {
+			return false
+		}
+		retries++
+		// only retry if atomic is set
+		// otherwise helm doesn't uninstall
+		// the release on failure
+		if !h.Atomic {
+			return false
+		}
+		// check if error is retriable
 		return wait.Interrupted(err) ||
 			strings.Contains(err.Error(), "connection refused")
 	}
@@ -185,14 +210,8 @@ func (h *Client) install(ctx context.Context, chartRaw []byte, values map[string
 	}
 	retrier := retry.NewIntervalRetrier(doer, 30*time.Second, retriable)
 
-	// Since we have no precise retry condition we want to stop retrying after 10 minutes.
-	// The helm library only reports a timeout error in the error cases we currently know.
-	// Other errors will not be retried.
-	newCtx, cancel := context.WithTimeout(ctx, 10*time.Minute)
-	defer cancel()
-
 	retryLoopStartTime := time.Now()
-	if err := retrier.Do(newCtx); err != nil {
+	if err := retrier.Do(ctx); err != nil {
 		return fmt.Errorf("helm install: %w", err)
 	}
 	retryLoopFinishDuration := time.Since(retryLoopStartTime)
@@ -201,6 +220,23 @@ func (h *Client) install(ctx context.Context, chartRaw []byte, values map[string
 	return nil
 }
 
+func (h *Client) setWaitMode(waitMode helm.WaitMode) error {
+	switch waitMode {
+	case helm.WaitModeNone:
+		h.Wait = false
+		h.Atomic = false
+	case helm.WaitModeWait:
+		h.Wait = true
+		h.Atomic = false
+	case helm.WaitModeAtomic:
+		h.Wait = true
+		h.Atomic = true
+	default:
+		return fmt.Errorf("unknown wait mode %q", waitMode)
+	}
+	return nil
+}
+
 // installDoer is a help struct to enable retrying helm's install action.
 type installDoer struct {
 	client *Client

cli/internal/cmd/BUILD.bazel

@@ -65,6 +65,7 @@ go_library(
         "//internal/config/migration",
         "//internal/constants",
         "//internal/crypto",
+        "//internal/deploy/helm",
         "//internal/file",
         "//internal/grpc/dialer",
         "//internal/grpc/grpclog",

cli/internal/cmd/init.go

@@ -43,6 +43,7 @@ import (
 	"github.com/edgelesssys/constellation/v2/internal/config"
 	"github.com/edgelesssys/constellation/v2/internal/constants"
 	"github.com/edgelesssys/constellation/v2/internal/crypto"
+	helmdeploy "github.com/edgelesssys/constellation/v2/internal/deploy/helm"
 	"github.com/edgelesssys/constellation/v2/internal/file"
 	"github.com/edgelesssys/constellation/v2/internal/grpc/dialer"
 	"github.com/edgelesssys/constellation/v2/internal/grpc/grpclog"
@@ -65,6 +66,7 @@ func NewInitCmd() *cobra.Command {
 	}
 	cmd.Flags().String("master-secret", "", "path to base64-encoded master secret")
 	cmd.Flags().Bool("conformance", false, "enable conformance mode")
+	cmd.Flags().Bool("skip-helm-wait", false, "install helm charts without waiting for deployments to be ready")
 	cmd.Flags().Bool("merge-kubeconfig", false, "merge Constellation kubeconfig file with default kubeconfig file in $HOME/.kube/config")
 	return cmd
 }
@@ -174,7 +176,7 @@ func (i *initCmd) initialize(cmd *cobra.Command, newDialer func(validator atls.V
 	}
 	helmLoader := helm.NewLoader(provider, k8sVersion)
 	i.log.Debugf("Created new Helm loader")
-	helmDeployments, err := helmLoader.Load(conf, flags.conformance, masterSecret.Key, masterSecret.Salt)
+	helmDeployments, err := helmLoader.Load(conf, flags.conformance, flags.helmWaitMode, masterSecret.Key, masterSecret.Salt)
 	i.log.Debugf("Loaded Helm deployments")
 	if err != nil {
 		return fmt.Errorf("loading Helm charts: %w", err)
@@ -409,6 +411,15 @@ func (i *initCmd) evalFlagArgs(cmd *cobra.Command) (initFlags, error) {
 		return initFlags{}, fmt.Errorf("parsing conformance flag: %w", err)
 	}
 	i.log.Debugf("Conformance flag is %t", conformance)
+	skipHelmWait, err := cmd.Flags().GetBool("skip-helm-wait")
+	if err != nil {
+		return initFlags{}, fmt.Errorf("parsing skip-helm-wait flag: %w", err)
+	}
+	helmWaitMode := helmdeploy.WaitModeAtomic
+	if skipHelmWait {
+		helmWaitMode = helmdeploy.WaitModeNone
+	}
+	i.log.Debugf("Helm wait flag is %t", skipHelmWait)
 	configPath, err := cmd.Flags().GetString("config")
 	if err != nil {
 		return initFlags{}, fmt.Errorf("parsing config path flag: %w", err)
@@ -429,6 +440,7 @@ func (i *initCmd) evalFlagArgs(cmd *cobra.Command) (initFlags, error) {
 	return initFlags{
 		configPath:       configPath,
 		conformance:      conformance,
+		helmWaitMode:     helmWaitMode,
 		masterSecretPath: masterSecretPath,
 		force:            force,
 		mergeConfigs:     mergeConfigs,
@@ -440,6 +452,7 @@ type initFlags struct {
 	configPath       string
 	masterSecretPath string
 	conformance      bool
+	helmWaitMode     helmdeploy.WaitMode
 	force            bool
 	mergeConfigs     bool
 }

cli/internal/helm/loader.go

@@ -101,24 +101,24 @@ func NewLoader(csp cloudprovider.Provider, k8sVersion versions.ValidK8sVersion)
 }
 
 // Load the embedded helm charts.
-func (i *ChartLoader) Load(config *config.Config, conformanceMode bool, masterSecret, salt []byte) ([]byte, error) {
-	ciliumRelease, err := i.loadRelease(ciliumInfo)
+func (i *ChartLoader) Load(config *config.Config, conformanceMode bool, helmWaitMode helm.WaitMode, masterSecret, salt []byte) ([]byte, error) {
+	ciliumRelease, err := i.loadRelease(ciliumInfo, helmWaitMode)
 	if err != nil {
 		return nil, fmt.Errorf("loading cilium: %w", err)
 	}
 	extendCiliumValues(ciliumRelease.Values, conformanceMode)
 
-	certManagerRelease, err := i.loadRelease(certManagerInfo)
+	certManagerRelease, err := i.loadRelease(certManagerInfo, helmWaitMode)
 	if err != nil {
 		return nil, fmt.Errorf("loading cert-manager: %w", err)
 	}
 
-	operatorRelease, err := i.loadRelease(constellationOperatorsInfo)
+	operatorRelease, err := i.loadRelease(constellationOperatorsInfo, helmWaitMode)
 	if err != nil {
 		return nil, fmt.Errorf("loading operators: %w", err)
 	}
 
-	conServicesRelease, err := i.loadRelease(constellationServicesInfo)
+	conServicesRelease, err := i.loadRelease(constellationServicesInfo, helmWaitMode)
 	if err != nil {
 		return nil, fmt.Errorf("loading constellation-services: %w", err)
 	}
@@ -136,7 +136,7 @@ func (i *ChartLoader) Load(config *config.Config, conformanceMode bool, masterSe
 }
 
 // loadRelease loads the embedded chart and values depending on the given info argument.
-func (i *ChartLoader) loadRelease(info chartInfo) (helm.Release, error) {
+func (i *ChartLoader) loadRelease(info chartInfo, helmWaitMode helm.WaitMode) (helm.Release, error) {
 	chart, err := loadChartsDir(helmFS, info.path)
 	if err != nil {
 		return helm.Release{}, fmt.Errorf("loading %s chart: %w", info.releaseName, err)
@@ -168,7 +168,7 @@ func (i *ChartLoader) loadRelease(info chartInfo) (helm.Release, error) {
 		return helm.Release{}, fmt.Errorf("packaging %s chart: %w", info.releaseName, err)
 	}
 
-	return helm.Release{Chart: chartRaw, Values: values, ReleaseName: info.releaseName}, nil
+	return helm.Release{Chart: chartRaw, Values: values, ReleaseName: info.releaseName, WaitMode: helmWaitMode}, nil
 }
 
 // loadCiliumValues is used to separate the marshalling step from the loading step.

cli/internal/helm/loader_test.go

@@ -39,7 +39,7 @@ func TestLoad(t *testing.T) {
 
 	config := &config.Config{Provider: config.ProviderConfig{GCP: &config.GCPConfig{}}}
 	chartLoader := ChartLoader{csp: config.GetProvider()}
-	release, err := chartLoader.Load(config, true, []byte("secret"), []byte("salt"))
+	release, err := chartLoader.Load(config, true, helm.WaitModeAtomic, []byte("secret"), []byte("salt"))
 	require.NoError(err)
 
 	var helmReleases helm.Releases

docs/docs/reference/cli.md

@@ -253,6 +253,7 @@ constellation init [flags]
   -h, --help                   help for init
       --master-secret string   path to base64-encoded master secret
       --merge-kubeconfig       merge Constellation kubeconfig file with default kubeconfig file in $HOME/.kube/config
+      --skip-helm-wait         install helm charts without waiting for deployments to be ready
 ```
 
 ### Options inherited from parent commands

internal/deploy/helm/helm.go

@@ -12,6 +12,7 @@ type Release struct {
 	Chart       []byte
 	Values      map[string]any
 	ReleaseName string
+	WaitMode    WaitMode
 }
 
 // Releases bundles all helm releases to be deployed to Constellation.
@@ -43,3 +44,16 @@ func MergeMaps(a, b map[string]any) map[string]any {
 	}
 	return out
 }
+
+// WaitMode specifies the wait mode for a helm release.
+type WaitMode string
+
+const (
+	// WaitModeNone specifies that the helm release should not wait for the resources to be ready.
+	WaitModeNone WaitMode = ""
+	// WaitModeWait specifies that the helm release should wait for the resources to be ready.
+	WaitModeWait WaitMode = "wait"
+	// WaitModeAtomic specifies that the helm release should
+	// wait for the resources to be ready and roll back atomically on failure.
+	WaitModeAtomic WaitMode = "atomic"
+)