Add FailureDomain to M3Cluster and Machine

Signed-off-by: Steven Fitzpatrick <[email protected]>
metal3-io · Nov 30, 2022 · a1c0cfa · a1c0cfa
1 parent 496bfb4
commit a1c0cfa
Show file tree

Hide file tree

Showing 8 changed files with 85 additions and 7 deletions.
diff --git a/api/v1alpha5/zz_generated.conversion.go b/api/v1alpha5/zz_generated.conversion.go
diff --git a/api/v1beta1/metal3cluster_types.go b/api/v1beta1/metal3cluster_types.go
@@ -76,6 +76,11 @@ type Metal3ClusterStatus struct {
 	// +optional
 	FailureMessage *string `json:"failureMessage,omitempty"`
 
+	// FailureDomains specifies the failure domains available in the cluster.
+	// This will be used by Cluster API to try and spread the machines across failure domains
+	// +optional
+	FailureDomains clusterv1.FailureDomains `json:"failureDomains,omitempty"`
+
 	// Ready denotes that the Metal3 cluster (infrastructure) is ready. In
 	// Baremetal case, it does not mean anything for now as no infrastructure
 	// steps need to be performed. Required by Cluster API. Set to True by the

diff --git a/api/v1beta1/metal3machine_types.go b/api/v1beta1/metal3machine_types.go
@@ -76,6 +76,10 @@ type Metal3MachineSpec struct {
 	// +kubebuilder:validation:Enum:=metadata;disabled
 	// +optional
 	AutomatedCleaningMode *string `json:"automatedCleaningMode,omitempty"`
+
+	// FailureDomain is the failure domain the machine will be created in.
+	// +optional
+	FailureDomain *string `json:"failureDomain,omitempty"`
 }
 
 // Metal3MachineStatus defines the observed state of Metal3Machine.

diff --git a/api/v1beta1/zz_generated.deepcopy.go b/api/v1beta1/zz_generated.deepcopy.go
diff --git a/baremetal/metal3machine_manager.go b/baremetal/metal3machine_manager.go
@@ -71,6 +71,9 @@ const (
 	ProviderIDPrefix = "metal3://"
 	// ProviderLabelPrefix is a label prefix for ProviderID.
 	ProviderLabelPrefix = "metal3.io/uuid"
+	// FailureDomainLabelPrefix is a label prefix for FailureDomains
+	// TODO: Should this be customizable by an operator?
+	FailureDomainLabelPrefix = "infrastructure.cluster.x-k8s.io/failure-domain"
 )
 
 var (
@@ -889,6 +892,14 @@ func (m *MachineManager) chooseHost(ctx context.Context) (*bmov1alpha1.BareMetal
 		}
 		reqs = append(reqs, *r)
 	}
+	if m.Metal3Machine.Spec.FailureDomain != nil {
+		r, err := labels.NewRequirement(FailureDomainLabelPrefix, selection.Equals, []string{*m.Metal3Machine.Spec.FailureDomain})
+		if err != nil {
+			m.Log.Error(err, "Failed to create FailureDomain MatchLabel requirement, not choosing host")
+			return nil, nil, err
+		}
+		reqs = append(reqs, *r)
+	}
 	labelSelector = labelSelector.Add(reqs...)
 
 	availableHosts := []*bmov1alpha1.BareMetalHost{}

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3clusters.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3clusters.yaml
@@ -229,6 +229,27 @@ spec:
                   - type
                   type: object
                 type: array
+              failureDomains:
+                additionalProperties:
+                  description: FailureDomainSpec is the Schema for Cluster API failure
+                    domains. It allows controllers to understand how many failure
+                    domains a cluster can optionally span across.
+                  properties:
+                    attributes:
+                      additionalProperties:
+                        type: string
+                      description: Attributes is a free form map of attributes an
+                        infrastructure provider might use or require.
+                      type: object
+                    controlPlane:
+                      description: ControlPlane determines if this failure domain
+                        is suitable for use by control plane machines.
+                      type: boolean
+                  type: object
+                description: FailureDomains specifies the list of unique failure domains
+                  available in the cluster. The list will be used by Cluster API to
+                  try and spread the machines across failure domains
+                type: object
               failureMessage:
                 description: FailureMessage indicates that there is a fatal problem
                   reconciling the state, and will be set to a descriptive error message.

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3machines.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3machines.yaml
@@ -468,6 +468,10 @@ spec:
                     type: string
                 type: object
                 x-kubernetes-map-type: atomic
+              failureDomain:
+                description: FailureDomain is the failure domain the machine will
+                  be created in.
+                type: string
               hostSelector:
                 description: HostSelector specifies matching criteria for labels on
                   BareMetalHosts. This is used to limit the set of BareMetalHost objects

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3machinetemplates.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_metal3machinetemplates.yaml
@@ -323,6 +323,10 @@ spec:
                             type: string
                         type: object
                         x-kubernetes-map-type: atomic
+                      failureDomain:
+                        description: FailureDomain is the failure domain the machine
+                          will be created in.
+                        type: string
                       hostSelector:
                         description: HostSelector specifies matching criteria for
                           labels on BareMetalHosts. This is used to limit the set