Improving the default PDB implementation.

config/crds/v1/all-crds.yaml

@@ -4661,9 +4661,17 @@ spec:
                 type: array
               podDisruptionBudget:
                 description: |-
-                  PodDisruptionBudget provides access to the default Pod disruption budget for the Elasticsearch cluster.
-                  The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
-                  In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  PodDisruptionBudget provides access to the default Pod disruption budget(s) for the Elasticsearch cluster.
+                  The behavior depends on the license level.
+                  With a Basic license:
+                    The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
+                    In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  With an Enterprise license:
+                    The default budget is split into multiple budgets, each targeting a specific node role type allowing additional disruptions
+                    for certain roles according to the health status of the cluster.
+                      Example:
+                        All data roles (excluding frozen): allows disruptions only when the cluster is green.
+                        All other roles: allows disruptions only when the cluster is yellow or green.
                   To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML).
                 properties:
                   metadata:

config/crds/v1/resources/elasticsearch.k8s.elastic.co_elasticsearches.yaml

@@ -9239,9 +9239,17 @@ spec:
                 type: array
               podDisruptionBudget:
                 description: |-
-                  PodDisruptionBudget provides access to the default Pod disruption budget for the Elasticsearch cluster.
-                  The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
-                  In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  PodDisruptionBudget provides access to the default Pod disruption budget(s) for the Elasticsearch cluster.
+                  The behavior depends on the license level.
+                  With a Basic license:
+                    The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
+                    In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  With an Enterprise license:
+                    The default budget is split into multiple budgets, each targeting a specific node role type allowing additional disruptions
+                    for certain roles according to the health status of the cluster.
+                      Example:
+                        All data roles (excluding frozen): allows disruptions only when the cluster is green.
+                        All other roles: allows disruptions only when the cluster is yellow or green.
                   To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML).
                 properties:
                   metadata:

deploy/eck-operator/charts/eck-operator-crds/templates/all-crds.yaml

@@ -4703,9 +4703,17 @@ spec:
                 type: array
               podDisruptionBudget:
                 description: |-
-                  PodDisruptionBudget provides access to the default Pod disruption budget for the Elasticsearch cluster.
-                  The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
-                  In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  PodDisruptionBudget provides access to the default Pod disruption budget(s) for the Elasticsearch cluster.
+                  The behavior depends on the license level.
+                  With a Basic license:
+                    The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
+                    In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+                  With an Enterprise license:
+                    The default budget is split into multiple budgets, each targeting a specific node role type allowing additional disruptions
+                    for certain roles according to the health status of the cluster.
+                      Example:
+                        All data roles (excluding frozen): allows disruptions only when the cluster is green.
+                        All other roles: allows disruptions only when the cluster is yellow or green.
                   To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML).
                 properties:
                   metadata:

docs/reference/api-docs.md

@@ -10,4 +10,4 @@ applies_to:
 
 # {{eck}} API Reference (moved)
 
-This page has moved [here](./api-reference/index.md).
+This page has moved [here](./api-reference/index.md).

docs/reference/api-reference/main.md

@@ -1093,7 +1093,7 @@ ElasticsearchSpec holds the specification of an Elasticsearch cluster.
 | *`transport`* __[TransportConfig](#transportconfig)__ | Transport holds transport layer settings for Elasticsearch. |
 | *`nodeSets`* __[NodeSet](#nodeset) array__ | NodeSets allow specifying groups of Elasticsearch nodes sharing the same configuration and Pod templates. |
 | *`updateStrategy`* __[UpdateStrategy](#updatestrategy)__ | UpdateStrategy specifies how updates to the cluster should be performed. |
-| *`podDisruptionBudget`* __[PodDisruptionBudgetTemplate](#poddisruptionbudgettemplate)__ | PodDisruptionBudget provides access to the default Pod disruption budget for the Elasticsearch cluster.<br>The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.<br>In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.<br>To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML). |
+| *`podDisruptionBudget`* __[PodDisruptionBudgetTemplate](#poddisruptionbudgettemplate)__ | PodDisruptionBudget provides access to the default Pod disruption budget(s) for the Elasticsearch cluster.<br>The behavior depends on the license level.<br>With a Basic license:<br>  The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.<br>  In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.<br>With an Enterprise license:<br>  The default budget is split into multiple budgets, each targeting a specific node role type allowing additional disruptions<br>  for certain roles according to the health status of the cluster.<br>    Example:<br>      All data roles (excluding frozen): allows disruptions only when the cluster is green.<br>      All other roles: allows disruptions only when the cluster is yellow or green.<br>To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML). |
 | *`auth`* __[Auth](#auth)__ | Auth contains user authentication and authorization security settings for Elasticsearch. |
 | *`secureSettings`* __[SecretSource](#secretsource) array__ | SecureSettings is a list of references to Kubernetes secrets containing sensitive configuration options for Elasticsearch. |
 | *`serviceAccountName`* __string__ | ServiceAccountName is used to check access from the current resource to a resource (for ex. a remote Elasticsearch cluster) in a different namespace.<br>Can only be used if ECK is enforcing RBAC on references. |

pkg/apis/elasticsearch/v1/elasticsearch_config.go

@@ -16,6 +16,7 @@ import (
 type NodeRole string
 
 const (
+	CoordinatingRole        NodeRole = ""
 	DataColdRole            NodeRole = "data_cold"
 	DataContentRole         NodeRole = "data_content"
 	DataFrozenRole          NodeRole = "data_frozen"
@@ -129,6 +130,8 @@ func (n *Node) IsConfiguredWithRole(role NodeRole) bool {
 		return ptr.Deref(n.Transform, n.IsConfiguredWithRole(DataRole))
 	case VotingOnlyRole:
 		return ptr.Deref(n.VotingOnly, false)
+	case CoordinatingRole:
+		return n.Roles != nil && len(n.Roles) == 0
 	}
 
 	// This point should never be reached. The default is to assume that a node has all roles except voting_only.

pkg/apis/elasticsearch/v1/elasticsearch_types.go

@@ -103,9 +103,17 @@ type ElasticsearchSpec struct {
 	// +kubebuilder:validation:Optional
 	UpdateStrategy UpdateStrategy `json:"updateStrategy,omitempty"`
 
-	// PodDisruptionBudget provides access to the default Pod disruption budget for the Elasticsearch cluster.
-	// The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
-	// In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+	// PodDisruptionBudget provides access to the default Pod disruption budget(s) for the Elasticsearch cluster.
+	// The behavior depends on the license level.
+	// With a Basic license:
+	//   The default budget doesn't allow any Pod to be removed in case the cluster is not green or if there is only one node of type `data` or `master`.
+	//   In all other cases the default PodDisruptionBudget sets `minUnavailable` equal to the total number of nodes minus 1.
+	// With an Enterprise license:
+	//   The default budget is split into multiple budgets, each targeting a specific node role type allowing additional disruptions
+	//   for certain roles according to the health status of the cluster.
+	//     Example:
+	//       All data roles (excluding frozen): allows disruptions only when the cluster is green.
+	//       All other roles: allows disruptions only when the cluster is yellow or green.
 	// To disable, set `PodDisruptionBudget` to the empty value (`{}` in YAML).
 	// +kubebuilder:validation:Optional
 	PodDisruptionBudget *commonv1.PodDisruptionBudgetTemplate `json:"podDisruptionBudget,omitempty"`

pkg/apis/elasticsearch/v1/name.go

@@ -200,3 +200,13 @@ func StackConfigAdditionalSecretName(esName string, secretName string) string {
 	secretNameHash := hash.HashObject(secretName)
 	return ESNamer.Suffix(esName, "scp", secretNameHash)
 }
+
+// PodDisruptionBudgetNameForRole returns the name of the PodDisruptionBudget for a given Elasticsearch cluster name and role.
+func PodDisruptionBudgetNameForRole(esName string, role string) string {
+	name := DefaultPodDisruptionBudget(esName) + "-" + role
+	// For coordinating nodes (no roles), append "coordinating" to the name
+	if role == "" {
+		name += "coordinating"
+	}
+	return name
+}

pkg/controller/common/statefulset/fixtures.go

@@ -16,16 +16,24 @@ import (
 )
 
 type TestSset struct {
-	Namespace       string
-	Name            string
-	ClusterName     string
-	Version         string
-	Replicas        int32
-	Master          bool
-	Data            bool
-	Ingest          bool
-	Status          appsv1.StatefulSetStatus
-	ResourceVersion string
+	Namespace           string
+	Name                string
+	ClusterName         string
+	Version             string
+	Replicas            int32
+	Master              bool
+	Data                bool
+	Ingest              bool
+	ML                  bool
+	Transform           bool
+	RemoteClusterClient bool
+	DataHot             bool
+	DataWarm            bool
+	DataCold            bool
+	DataContent         bool
+	DataFrozen          bool
+	Status              appsv1.StatefulSetStatus
+	ResourceVersion     string
 }
 
 func (t TestSset) Pods() []client.Object {
@@ -54,6 +62,14 @@ func (t TestSset) Build() appsv1.StatefulSet {
 	label.NodeTypesMasterLabelName.Set(t.Master, labels)
 	label.NodeTypesDataLabelName.Set(t.Data, labels)
 	label.NodeTypesIngestLabelName.Set(t.Ingest, labels)
+	label.NodeTypesMLLabelName.Set(t.ML, labels)
+	label.NodeTypesTransformLabelName.Set(t.Transform, labels)
+	label.NodeTypesRemoteClusterClientLabelName.Set(t.RemoteClusterClient, labels)
+	label.NodeTypesDataHotLabelName.Set(t.DataHot, labels)
+	label.NodeTypesDataWarmLabelName.Set(t.DataWarm, labels)
+	label.NodeTypesDataColdLabelName.Set(t.DataCold, labels)
+	label.NodeTypesDataContentLabelName.Set(t.DataContent, labels)
+	label.NodeTypesDataFrozenLabelName.Set(t.DataFrozen, labels)
 	statefulSet := appsv1.StatefulSet{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      t.Name,
@@ -86,19 +102,27 @@ func (t TestSset) BuildPtr() *appsv1.StatefulSet {
 }
 
 type TestPod struct {
-	Namespace       string
-	Name            string
-	ClusterName     string
-	StatefulSetName string
-	Version         string
-	Revision        string
-	Master          bool
-	Data            bool
-	Ingest          bool
-	Ready           bool
-	RestartCount    int32
-	Phase           corev1.PodPhase
-	ResourceVersion string
+	Namespace           string
+	Name                string
+	ClusterName         string
+	StatefulSetName     string
+	Version             string
+	Revision            string
+	Master              bool
+	Data                bool
+	Ingest              bool
+	ML                  bool
+	Transform           bool
+	RemoteClusterClient bool
+	DataHot             bool
+	DataWarm            bool
+	DataCold            bool
+	DataContent         bool
+	DataFrozen          bool
+	Ready               bool
+	RestartCount        int32
+	Phase               corev1.PodPhase
+	ResourceVersion     string
 }
 
 func (t TestPod) Build() corev1.Pod {
@@ -111,6 +135,14 @@ func (t TestPod) Build() corev1.Pod {
 	label.NodeTypesMasterLabelName.Set(t.Master, labels)
 	label.NodeTypesDataLabelName.Set(t.Data, labels)
 	label.NodeTypesIngestLabelName.Set(t.Ingest, labels)
+	label.NodeTypesMLLabelName.Set(t.ML, labels)
+	label.NodeTypesTransformLabelName.Set(t.Transform, labels)
+	label.NodeTypesRemoteClusterClientLabelName.Set(t.RemoteClusterClient, labels)
+	label.NodeTypesDataHotLabelName.Set(t.DataHot, labels)
+	label.NodeTypesDataWarmLabelName.Set(t.DataWarm, labels)
+	label.NodeTypesDataColdLabelName.Set(t.DataCold, labels)
+	label.NodeTypesDataContentLabelName.Set(t.DataContent, labels)
+	label.NodeTypesDataFrozenLabelName.Set(t.DataFrozen, labels)
 
 	status := corev1.PodStatus{
 		// assume Running by default

pkg/controller/elasticsearch/driver/nodes.go

@@ -133,7 +133,7 @@ func (d *defaultDriver) reconcileNodeSpecs(
 	}
 
 	// Update PDB to account for new replicas.
-	if err := pdb.Reconcile(ctx, d.Client, d.ES, actualStatefulSets, meta); err != nil {
+	if err := pdb.Reconcile(ctx, d.Client, d.ES, actualStatefulSets, expectedResources, meta); err != nil {
 		return results.WithError(err)
 	}
 

pkg/controller/elasticsearch/elasticsearch_controller.go

@@ -13,6 +13,7 @@ import (
 	"go.elastic.co/apm/v2"
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
+	policyv1 "k8s.io/api/policy/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/tools/record"
@@ -108,6 +109,12 @@ func addWatches(mgr manager.Manager, c controller.Controller, r *ReconcileElasti
 		return err
 	}
 
+	// Watch PodDisruptionBudgets
+	if err := c.Watch(
+		source.Kind(mgr.GetCache(), &policyv1.PodDisruptionBudget{}, handler.TypedEnqueueRequestForOwner[*policyv1.PodDisruptionBudget](mgr.GetScheme(), mgr.GetRESTMapper(), &esv1.Elasticsearch{}, handler.OnlyControllerOwner()))); err != nil {
+		return err
+	}
+
 	// Watch owned and soft-owned secrets
 	if err := c.Watch(source.Kind(mgr.GetCache(), &corev1.Secret{}, r.dynamicWatches.Secrets)); err != nil {
 		return err

pkg/controller/elasticsearch/label/label.go

@@ -85,7 +85,7 @@ func IsMasterNodeSet(statefulSet appsv1.StatefulSet) bool {
 
 // IsDataNodeSet returns true if the given StatefulSet specifies data nodes.
 func IsDataNodeSet(statefulSet appsv1.StatefulSet) bool {
-	return NodeTypesDataLabelName.HasValue(true, statefulSet.Spec.Template.Labels)
+	return NodeTypesDataLabelName.HasValue(true, statefulSet.Spec.Template.Labels) || NodeTypesDataHotLabelName.HasValue(true, statefulSet.Spec.Template.Labels) || NodeTypesDataColdLabelName.HasValue(true, statefulSet.Spec.Template.Labels) || NodeTypesDataContentLabelName.HasValue(true, statefulSet.Spec.Template.Labels) || NodeTypesDataWarmLabelName.HasValue(true, statefulSet.Spec.Template.Labels)
 }
 
 // IsIngestNodeSet returns true if the given StatefulSet specifies ingest nodes.