Allow forced detachment of a host from Ironic

Adds Force as DetachedAnnotationArgument to API.
Detach in all states if force is set to true.
Add force argument to provisioner's Detach and Delete functions.

Signed-off-by: CrystalChun <cchun@redhat.com>

Author: CrystalChun

apis/metal3.io/v1alpha1/baremetalhost_types.go

@@ -614,6 +614,11 @@ const (
 type DetachedAnnotationArguments struct {
 	// DeleteAction indicates the desired delete logic when the detached annotation is present
 	DeleteAction DetachedDeleteAction `json:"deleteAction,omitempty"`
+
+	// Force indicates if detaching should be forced regardless of the host's state
+	// +optional
+	// +kubebuilder:default:=false
+	Force bool `json:"force,omitempty"`
 }
 
 // Match compares the saved status information with the name and

internal/controller/metal3.io/baremetalhost_controller.go

@@ -538,7 +538,7 @@ func (r *BareMetalHostReconciler) actionPowerOffBeforeDeleting(prov provisioner.
 }
 
 // Manage deletion of the host.
-func (r *BareMetalHostReconciler) actionDeleting(prov provisioner.Provisioner, info *reconcileInfo) actionResult {
+func (r *BareMetalHostReconciler) actionDeleting(prov provisioner.Provisioner, info *reconcileInfo, force bool) actionResult {
 	info.log.Info(
 		"marked to be deleted",
 		"timestamp", info.host.DeletionTimestamp,
@@ -550,7 +550,7 @@ func (r *BareMetalHostReconciler) actionDeleting(prov provisioner.Provisioner, i
 		return deleteComplete{}
 	}
 
-	provResult, err := prov.Delete()
+	provResult, err := prov.Delete(force)
 	if err != nil {
 		return actionError{fmt.Errorf("failed to delete: %w", err)}
 	}
@@ -616,8 +616,8 @@ func hasCustomDeploy(host *metal3api.BareMetalHost) bool {
 }
 
 // detachHost() detaches the host from the Provisioner.
-func (r *BareMetalHostReconciler) detachHost(prov provisioner.Provisioner, info *reconcileInfo) actionResult {
-	provResult, err := prov.Detach()
+func (r *BareMetalHostReconciler) detachHost(prov provisioner.Provisioner, info *reconcileInfo, force bool) actionResult {
+	provResult, err := prov.Detach(force)
 	if err != nil {
 		return actionError{fmt.Errorf("failed to detach: %w", err)}
 	}

internal/controller/metal3.io/host_state_machine.go

@@ -316,9 +316,14 @@ func (hsm *hostStateMachine) checkDetachedHost(info *reconcileInfo) (result acti
 		// Only allow detaching hosts in Provisioned/ExternallyProvisioned/Ready/Available states
 		switch info.host.Status.Provisioning.State {
 		case metal3api.StateProvisioned, metal3api.StateExternallyProvisioned, metal3api.StateReady, metal3api.StateAvailable:
-			return hsm.Reconciler.detachHost(hsm.Provisioner, info)
+			return hsm.Reconciler.detachHost(hsm.Provisioner, info, false)
 		default:
-			info.log.Info("host cannot be detached yet, waiting for the current operation to finish", "provisioningState", info.host.Status.Provisioning.State)
+			info.log.Info("host not in allowed detaching state, checking for force annotation")
+			if hasForceDetachAnnotation(hsm.Host) {
+				info.log.Info("forcing detach of host", "host", info.host.Name, "annotation", hsm.Host.GetAnnotations()[metal3api.DetachedAnnotation])
+				return hsm.Reconciler.detachHost(hsm.Provisioner, info, true)
+			}
+			info.log.Info("host cannot be detached yet, waiting for the current operation to finish", "provisioningState", info.host.Status.Provisioning.State, "annotation", hsm.Host.GetAnnotations()[metal3api.DetachedAnnotation])
 		}
 	}
 	if info.host.Status.ErrorType == metal3api.DetachError {
@@ -339,6 +344,20 @@ func (hsm *hostStateMachine) checkDetachedHost(info *reconcileInfo) (result acti
 	return nil
 }
 
+func hasForceDetachAnnotation(host *metal3api.BareMetalHost) bool {
+	annotations := host.GetAnnotations()
+	if annotations != nil {
+		if val, ok := annotations[metal3api.DetachedAnnotation]; ok {
+			args := metal3api.DetachedAnnotationArguments{}
+			if err := json.Unmarshal([]byte(val), &args); err != nil {
+				return false
+			}
+			return args.Force
+		}
+	}
+	return false
+}
+
 func (hsm *hostStateMachine) ensureRegistered(info *reconcileInfo) (result actionResult) {
 	if !hsm.haveCreds {
 		// If we are in the process of deletion (which may start with
@@ -631,5 +650,5 @@ func (hsm *hostStateMachine) handlePoweringOffBeforeDelete(info *reconcileInfo)
 }
 
 func (hsm *hostStateMachine) handleDeleting(info *reconcileInfo) actionResult {
-	return hsm.Reconciler.actionDeleting(hsm.Provisioner, info)
+	return hsm.Reconciler.actionDeleting(hsm.Provisioner, info, false)
 }

internal/controller/metal3.io/host_state_machine_test.go

@@ -1378,11 +1378,11 @@ func (m *mockProvisioner) Deprovision(_ bool, _ metal3api.AutomatedCleaningMode)
 	return m.getNextResultByMethod("Deprovision"), err
 }
 
-func (m *mockProvisioner) Delete() (result provisioner.Result, err error) {
+func (m *mockProvisioner) Delete(_ bool) (result provisioner.Result, err error) {
 	return m.getNextResultByMethod("Delete"), err
 }
 
-func (m *mockProvisioner) Detach() (result provisioner.Result, err error) {
+func (m *mockProvisioner) Detach(_ bool) (result provisioner.Result, err error) {
 	res := m.getNextResultByMethod("Detach")
 	return res, err
 }

pkg/provisioner/demo/demo.go

@@ -277,7 +277,7 @@ func (p *demoProvisioner) Deprovision(_ bool, _ metal3api.AutomatedCleaningMode)
 // Delete removes the host from the provisioning system. It may be
 // called multiple times, and should return true for its dirty flag
 // until the deprovisioning operation is completed.
-func (p *demoProvisioner) Delete() (result provisioner.Result, err error) {
+func (p *demoProvisioner) Delete(_ bool) (result provisioner.Result, err error) {
 	p.log.Info("deleting host")
 	return result, nil
 }
@@ -287,7 +287,7 @@ func (p *demoProvisioner) Delete() (result provisioner.Result, err error) {
 // for the target system.  It may be called multiple times,
 // and should return true for its dirty  flag until the
 // deletion operation is completed.
-func (p *demoProvisioner) Detach() (result provisioner.Result, err error) {
+func (p *demoProvisioner) Detach(_ bool) (result provisioner.Result, err error) {
 	p.log.Info("detaching host")
 	return result, nil
 }

pkg/provisioner/fixture/fixture.go

@@ -329,7 +329,7 @@ func (p *fixtureProvisioner) Deprovision(_ bool, _ metal3api.AutomatedCleaningMo
 // Delete removes the host from the provisioning system. It may be
 // called multiple times, and should return true for its dirty flag
 // until the deprovisioning operation is completed.
-func (p *fixtureProvisioner) Delete() (result provisioner.Result, err error) {
+func (p *fixtureProvisioner) Delete(_ bool) (result provisioner.Result, err error) {
 	p.log.Info("deleting host")
 
 	if !p.state.Deleted {
@@ -347,8 +347,8 @@ func (p *fixtureProvisioner) Delete() (result provisioner.Result, err error) {
 // for the target system.  It may be called multiple times,
 // and should return true for its dirty  flag until the
 // deletion operation is completed.
-func (p *fixtureProvisioner) Detach() (result provisioner.Result, err error) {
-	return p.Delete()
+func (p *fixtureProvisioner) Detach(force bool) (result provisioner.Result, err error) {
+	return p.Delete(force)
 }
 
 // PowerOn ensures the server is powered on independently of any image

pkg/provisioner/ironic/delete_test.go

@@ -217,9 +217,9 @@ func deleteTest(t *testing.T, detach bool) {
 
 			var result provisioner.Result
 			if detach {
-				result, err = prov.Detach()
+				result, err = prov.Detach(false)
 			} else {
-				result, err = prov.Delete()
+				result, err = prov.Delete(false)
 			}
 
 			assert.Equal(t, tc.expectedDirty, result.Dirty)

pkg/provisioner/ironic/ironic.go

@@ -1523,7 +1523,7 @@ func (p *ironicProvisioner) Deprovision(restartOnFailure bool, automatedCleaning
 // Delete removes the host from the provisioning system. It may be
 // called multiple times, and should return true for its dirty flag
 // until the deprovisioning operation is completed.
-func (p *ironicProvisioner) Delete() (result provisioner.Result, err error) {
+func (p *ironicProvisioner) Delete(_ bool) (result provisioner.Result, err error) {
 	ironicNode, err := p.getNode()
 	if err != nil {
 		if errors.Is(err, provisioner.ErrNeedsRegistration) {
@@ -1605,10 +1605,10 @@ func (p *ironicProvisioner) Delete() (result provisioner.Result, err error) {
 // for the target system.  It may be called multiple times,
 // and should return true for its dirty  flag until the
 // deletion operation is completed.
-func (p *ironicProvisioner) Detach() (result provisioner.Result, err error) {
+func (p *ironicProvisioner) Detach(force bool) (result provisioner.Result, err error) {
 	// Currently the same behavior as Delete()
 	p.log.Info("removing the node for detachment", "node", p.nodeID)
-	return p.Delete()
+	return p.Delete(force)
 }
 
 // softPowerOffUnsupportedError is returned when the BMC does not

pkg/provisioner/provisioner.go

@@ -182,14 +182,14 @@ type Provisioner interface {
 	// Delete removes the host from the provisioning system. It may be
 	// called multiple times, and should return true for its dirty
 	// flag until the deletion operation is completed.
-	Delete() (result Result, err error)
+	Delete(force bool) (result Result, err error)
 
 	// Detach removes the host from the provisioning system.
 	// Similar to Delete, but ensures non-interruptive behavior
 	// for the target system.  It may be called multiple times,
 	// and should return true for its dirty  flag until the
 	// deletion operation is completed.
-	Detach() (result Result, err error)
+	Detach(force bool) (result Result, err error)
 
 	// PowerOn ensures the server is powered on independently of any image
 	// provisioning operation.