Allow marking releases stuck in a pending state as failed (#16)

Author: misberner

pkg/client/actionclient.go

@@ -62,6 +62,7 @@ type ActionInterface interface {
 	Get(name string, opts ...GetOption) (*release.Release, error)
 	Install(name, namespace string, chrt *chart.Chart, vals map[string]interface{}, opts ...InstallOption) (*release.Release, error)
 	Upgrade(name, namespace string, chrt *chart.Chart, vals map[string]interface{}, opts ...UpgradeOption) (*release.Release, error)
+	MarkFailed(release *release.Release, reason string) error
 	Uninstall(name string, opts ...UninstallOption) (*release.UninstallReleaseResponse, error)
 	Reconcile(rel *release.Release) error
 }
@@ -180,6 +181,14 @@ func (c *actionClient) Upgrade(name, namespace string, chrt *chart.Chart, vals m
 	return rel, nil
 }
 
+func (c *actionClient) MarkFailed(rel *release.Release, reason string) error {
+	infoCopy := *rel.Info
+	releaseCopy := *rel
+	releaseCopy.Info = &infoCopy
+	releaseCopy.SetStatus(release.StatusFailed, reason)
+	return c.conf.Releases.Update(&releaseCopy)
+}
+
 func (c *actionClient) Uninstall(name string, opts ...UninstallOption) (*release.UninstallReleaseResponse, error) {
 	uninstall := action.NewUninstall(c.conf)
 	for _, o := range opts {

pkg/reconciler/internal/conditions/conditions.go

@@ -41,6 +41,7 @@ const (
 	ReasonUpgradeError             = status.ConditionReason("UpgradeError")
 	ReasonReconcileError           = status.ConditionReason("ReconcileError")
 	ReasonUninstallError           = status.ConditionReason("UninstallError")
+	ReasonPendingError             = status.ConditionReason("PendingError")
 )
 
 func Initialized(stat corev1.ConditionStatus, reason status.ConditionReason, message interface{}) status.Condition {

pkg/reconciler/internal/fake/actionclient.go

@@ -48,17 +48,19 @@ func (hcg *fakeActionClientGetter) ActionClientFor(_ crclient.Object) (client.Ac
 }
 
 type ActionClient struct {
-	Gets       []GetCall
-	Installs   []InstallCall
-	Upgrades   []UpgradeCall
-	Uninstalls []UninstallCall
-	Reconciles []ReconcileCall
-
-	HandleGet       func() (*release.Release, error)
-	HandleInstall   func() (*release.Release, error)
-	HandleUpgrade   func() (*release.Release, error)
-	HandleUninstall func() (*release.UninstallReleaseResponse, error)
-	HandleReconcile func() error
+	Gets        []GetCall
+	Installs    []InstallCall
+	Upgrades    []UpgradeCall
+	MarkFaileds []MarkFailedCall
+	Uninstalls  []UninstallCall
+	Reconciles  []ReconcileCall
+
+	HandleGet        func() (*release.Release, error)
+	HandleInstall    func() (*release.Release, error)
+	HandleUpgrade    func() (*release.Release, error)
+	HandleMarkFailed func() error
+	HandleUninstall  func() (*release.UninstallReleaseResponse, error)
+	HandleReconcile  func() error
 }
 
 func NewActionClient() ActionClient {
@@ -109,6 +111,11 @@ type UpgradeCall struct {
 	Opts      []client.UpgradeOption
 }
 
+type MarkFailedCall struct {
+	Release *release.Release
+	Reason  string
+}
+
 type UninstallCall struct {
 	Name string
 	Opts []client.UninstallOption
@@ -133,6 +140,11 @@ func (c *ActionClient) Upgrade(name, namespace string, chrt *chart.Chart, vals m
 	return c.HandleUpgrade()
 }
 
+func (c *ActionClient) MarkFailed(rel *release.Release, reason string) error {
+	c.MarkFaileds = append(c.MarkFaileds, MarkFailedCall{rel, reason})
+	return c.HandleMarkFailed()
+}
+
 func (c *ActionClient) Uninstall(name string, opts ...client.UninstallOption) (*release.UninstallReleaseResponse, error) {
 	c.Uninstalls = append(c.Uninstalls, UninstallCall{name, opts})
 	return c.HandleUninstall()

pkg/reconciler/reconciler.go

@@ -78,6 +78,7 @@ type Reconciler struct {
 	skipDependentWatches    bool
 	maxConcurrentReconciles int
 	reconcilePeriod         time.Duration
+	markFailedAfter         time.Duration
 	maxHistory              int
 
 	annotSetupOnce       sync.Once
@@ -304,6 +305,18 @@ func WithMaxReleaseHistory(maxHistory int) Option {
 	}
 }
 
+// WithMarkFailedAfter specifies the duration after which the reconciler will mark a release in a pending (locked)
+// state as false in order to allow rolling forward.
+func WithMarkFailedAfter(duration time.Duration) Option {
+	return func(r *Reconciler) error {
+		if duration < 0 {
+			return errors.New("auto-rollback after duration must not be negative")
+		}
+		r.markFailedAfter = duration
+		return nil
+	}
+}
+
 // WithInstallAnnotations is an Option that configures Install annotations
 // to enable custom action.Install fields to be set based on the value of
 // annotations found in the custom resource watched by this reconciler.
@@ -553,6 +566,10 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (res ctrl.
 		)
 		return ctrl.Result{}, err
 	}
+	if state == statePending {
+		return r.handlePending(actionClient, rel, &u, log)
+	}
+
 	u.UpdateStatus(updater.EnsureCondition(conditions.Irreconcilable(corev1.ConditionFalse, "", "")))
 
 	for _, h := range r.preHooks {
@@ -630,6 +647,7 @@ const (
 	stateNeedsInstall helmReleaseState = "needs install"
 	stateNeedsUpgrade helmReleaseState = "needs upgrade"
 	stateUnchanged    helmReleaseState = "unchanged"
+	statePending      helmReleaseState = "pending"
 	stateError        helmReleaseState = "error"
 )
 
@@ -678,6 +696,10 @@ func (r *Reconciler) getReleaseState(client helmclient.ActionInterface, obj meta
 		return nil, stateNeedsInstall, nil
 	}
 
+	if currentRelease.Info != nil && currentRelease.Info.Status.IsPending() {
+		return currentRelease, statePending, nil
+	}
+
 	var opts []helmclient.UpgradeOption
 	if r.maxHistory > 0 {
 		opts = append(opts, func(u *action.Upgrade) error {
@@ -755,6 +777,35 @@ func (r *Reconciler) doUpgrade(actionClient helmclient.ActionInterface, u *updat
 	return rel, nil
 }
 
+func (r *Reconciler) handlePending(actionClient helmclient.ActionInterface, rel *release.Release, u *updater.Updater, log logr.Logger) (ctrl.Result, error) {
+	err := r.doHandlePending(actionClient, rel, log)
+	if err == nil {
+		err = errors.New("unknown error handling pending release")
+	}
+	u.UpdateStatus(
+		updater.EnsureCondition(conditions.Irreconcilable(corev1.ConditionTrue, conditions.ReasonPendingError, err)))
+	return ctrl.Result{}, err
+}
+
+func (r *Reconciler) doHandlePending(actionClient helmclient.ActionInterface, rel *release.Release, log logr.Logger) error {
+	if r.markFailedAfter <= 0 {
+		return errors.New("Release is in a pending (locked) state and cannot be modified. User intervention is required.")
+	}
+	if rel.Info == nil || rel.Info.LastDeployed.IsZero() {
+		return errors.New("Release is in a pending (locked) state and lacks 'last deployed' timestamp. User intervention is required.")
+	}
+	if pendingSince := time.Since(rel.Info.LastDeployed.Time); pendingSince < r.markFailedAfter {
+		return fmt.Errorf("Release is in a pending (locked) state and cannot currently be modified. Release will be marked failed to allow a roll-forward in %v.", r.markFailedAfter-pendingSince)
+	}
+
+	log.Info("Marking release as failed", "releaseName", rel.Name)
+	err := actionClient.MarkFailed(rel, fmt.Sprintf("operator marked pending (locked) release as failed after state did not change for %v", r.markFailedAfter))
+	if err != nil {
+		return fmt.Errorf("Failed to mark pending (locked) release as failed: %w", err)
+	}
+	return fmt.Errorf("marked release %s as failed to allow upgrade to succeed in next reconcile attempt", rel.Name)
+}
+
 func (r *Reconciler) reportOverrideEvents(obj runtime.Object) {
 	for k, v := range r.overrideValues {
 		r.eventRecorder.Eventf(obj, "Warning", "ValueOverridden",