Add support for detached annotation

apis/metal3.io/v1alpha1/baremetalhost_types.go

@@ -39,6 +39,10 @@ const (
 	// an immediate requeue)
 	PausedAnnotation = "baremetalhost.metal3.io/paused"
 
+	// Detached is the annotation which stops provisioner management of the host
+	// unlike in the paused case, the host status may be updated
+	DetachedAnnotation = "baremetalhost.metal3.io/detached"
+
 	// StatusAnnotation is the annotation that keeps a copy of the Status of BMH
 	// This is particularly useful when we pivot BMH. If the status
 	// annotation is present and status is empty, BMO will reconstruct BMH Status
@@ -121,6 +125,10 @@ const (
 	// OperationalStatusDelayed is the status value for when the host
 	// deployment needs to be delayed to limit simultaneous hosts provisioning
 	OperationalStatusDelayed = "delayed"
+
+	// OperationalStatusDetached is the status value when the host is
+	// marked unmanaged via the detached annotation
+	OperationalStatusDetached OperationalStatus = "detached"
 )
 
 // ErrorType indicates the class of problem that has caused the Host resource
@@ -147,6 +155,9 @@ const (
 	// PowerManagementError is an error condition occurring when the
 	// controller is unable to modify the power state of the Host.
 	PowerManagementError ErrorType = "power management error"
+	// DetachError is an error condition occurring when the
+	// controller is unable to detatch the host from the provisioner
+	DetachError ErrorType = "detach error"
 )
 
 // ProvisioningState defines the states the provisioner will report
@@ -625,7 +636,7 @@ type BareMetalHostStatus struct {
 	// after modifying this file
 
 	// OperationalStatus holds the status of the host
-	// +kubebuilder:validation:Enum="";OK;discovered;error;delayed
+	// +kubebuilder:validation:Enum="";OK;discovered;error;delayed;detached
 	OperationalStatus OperationalStatus `json:"operationalStatus"`
 
 	// ErrorType indicates the type of failure encountered when the

config/crd/bases/metal3.io_baremetalhosts.yaml

@@ -596,6 +596,7 @@ spec:
                 - discovered
                 - error
                 - delayed
+                - detached
                 type: string
               poweredOn:
                 description: indicator for whether or not the host is powered on

config/render/capm3.yaml

@@ -594,6 +594,7 @@ spec:
                 - discovered
                 - error
                 - delayed
+                - detached
                 type: string
               poweredOn:
                 description: indicator for whether or not the host is powered on

controllers/metal3.io/baremetalhost_controller.go

@@ -322,6 +322,7 @@ func recordActionFailure(info *reconcileInfo, errorType metal3v1alpha1.ErrorType
 	setErrorMessage(info.host, errorType, errorMessage)
 
 	eventType := map[metal3v1alpha1.ErrorType]string{
+		metal3v1alpha1.DetachError:                  "DetachError",
 		metal3v1alpha1.ProvisionedRegistrationError: "ProvisionedRegistrationError",
 		metal3v1alpha1.RegistrationError:            "RegistrationError",
 		metal3v1alpha1.InspectionError:              "InspectionError",
@@ -539,6 +540,33 @@ func getCurrentImage(host *metal3v1alpha1.BareMetalHost) *metal3v1alpha1.Image {
 	return nil
 }
 
+// detachHost() detaches the host from the Provisioner
+func (r *BareMetalHostReconciler) detachHost(prov provisioner.Provisioner, info *reconcileInfo) actionResult {
+	provResult, err := prov.Detach()
+	if err != nil {
+		return actionError{errors.Wrap(err, "failed to detach")}
+	}
+	if provResult.ErrorMessage != "" {
+		return recordActionFailure(info, metal3v1alpha1.DetachError, provResult.ErrorMessage)
+	}
+	if provResult.Dirty {
+		if info.host.Status.ErrorType == metal3v1alpha1.DetachError && clearError(info.host) {
+			return actionUpdate{actionContinue{provResult.RequeueAfter}}
+		}
+		return actionContinue{provResult.RequeueAfter}
+	}
+	slowPoll := actionContinue{unmanagedRetryDelay}
+	if info.host.Status.ErrorType == metal3v1alpha1.DetachError {
+		clearError(info.host)
+		info.host.Status.ErrorCount = 0
+	}
+	if info.host.SetOperationalStatus(metal3v1alpha1.OperationalStatusDetached) {
+		info.log.Info("host is detached, removed from provisioner")
+		return actionUpdate{slowPoll}
+	}
+	return slowPoll
+}
+
 // Test the credentials by connecting to the management controller.
 func (r *BareMetalHostReconciler) registerHost(prov provisioner.Provisioner, info *reconcileInfo) actionResult {
 	info.log.Info("registering and validating access to management controller",

controllers/metal3.io/host_state_machine.go

@@ -183,6 +183,10 @@ func (hsm *hostStateMachine) ReconcileState(info *reconcileInfo) (actionRes acti
 		return actionComplete{}
 	}
 
+	if detachedResult := hsm.checkDetachedHost(info); detachedResult != nil {
+		return detachedResult
+	}
+
 	if registerResult := hsm.ensureRegistered(info); registerResult != nil {
 		hostRegistrationRequired.Inc()
 		return registerResult
@@ -216,7 +220,11 @@ func (hsm *hostStateMachine) checkInitiateDelete() bool {
 	default:
 		hsm.NextState = metal3v1alpha1.StateDeleting
 	case metal3v1alpha1.StateProvisioning, metal3v1alpha1.StateProvisioned:
-		hsm.NextState = metal3v1alpha1.StateDeprovisioning
+		if hsm.Host.OperationalStatus() == metal3v1alpha1.OperationalStatusDetached {
+			hsm.NextState = metal3v1alpha1.StateDeleting
+		} else {
+			hsm.NextState = metal3v1alpha1.StateDeprovisioning
+		}
 	case metal3v1alpha1.StateDeprovisioning:
 		// Allow state machine to run to continue deprovisioning.
 		return false
@@ -227,6 +235,46 @@ func (hsm *hostStateMachine) checkInitiateDelete() bool {
 	return true
 }
 
+// hasInspectAnnotation checks for existence of baremetalhost.metal3.io/detached
+func hasDetachedAnnotation(host *metal3v1alpha1.BareMetalHost) bool {
+	annotations := host.GetAnnotations()
+	if annotations != nil {
+		if _, ok := annotations[metal3v1alpha1.DetachedAnnotation]; ok {
+			return true
+		}
+	}
+	return false
+}
+
+func (hsm *hostStateMachine) checkDetachedHost(info *reconcileInfo) (result actionResult) {
+	// If the detached annotation is set we remove any host from the
+	// provisioner and take no further action
+	// Note this doesn't change the current state, only the OperationalStatus
+	if hasDetachedAnnotation(hsm.Host) {
+		// Only allow detaching hosts in Provisioned/ExternallyProvisioned states
+		switch info.host.Status.Provisioning.State {
+		case metal3v1alpha1.StateProvisioned, metal3v1alpha1.StateExternallyProvisioned:
+			return hsm.Reconciler.detachHost(hsm.Provisioner, info)
+		}
+	}
+	if info.host.Status.ErrorType == metal3v1alpha1.DetachError {
+		clearError(info.host)
+		hsm.Host.Status.ErrorCount = 0
+		info.log.Info("removed detach error")
+		return actionUpdate{}
+	}
+	if info.host.OperationalStatus() == metal3v1alpha1.OperationalStatusDetached {
+		newStatus := metal3v1alpha1.OperationalStatusOK
+		if info.host.Status.ErrorType != "" {
+			newStatus = metal3v1alpha1.OperationalStatusError
+		}
+		info.host.SetOperationalStatus(newStatus)
+		info.log.Info("removed detached status")
+		return actionUpdate{}
+	}
+	return nil
+}
+
 func (hsm *hostStateMachine) ensureRegistered(info *reconcileInfo) (result actionResult) {
 	if !hsm.haveCreds {
 		// If we are in the process of deletion (which may start with