From 0c33ade1bca79ac1cada33608a596b04d005af2a Mon Sep 17 00:00:00 2001
From: Damien Ciabrini <dciabrin@redhat.com>
Date: Wed, 22 Mar 2023 16:16:20 +0000
Subject: [PATCH] galera: support for node selection and pod affinity

Add a spec option to drive the node selection during pod assignment,
as well as preferred pod antiafinity to separate galera pods when
possible.
---
 api/bases/mariadb.openstack.org_galeras.yaml  |  6 ++++++
 api/v1beta1/galera_types.go                   |  3 +++
 api/v1beta1/zz_generated.deepcopy.go          |  9 +++++++-
 .../bases/mariadb.openstack.org_galeras.yaml  |  6 ++++++
 pkg/statefulset.go                            | 21 +++++++++++++++++--
 5 files changed, 42 insertions(+), 3 deletions(-)

diff --git a/api/bases/mariadb.openstack.org_galeras.yaml b/api/bases/mariadb.openstack.org_galeras.yaml
index e7f8b004..2b7632f8 100644
--- a/api/bases/mariadb.openstack.org_galeras.yaml
+++ b/api/bases/mariadb.openstack.org_galeras.yaml
@@ -55,6 +55,12 @@ spec:
                 default: quay.io/tripleozedcentos9/openstack-mariadb:current-tripleo
                 description: Name of the galera container image to run
                 type: string
+              nodeSelector:
+                additionalProperties:
+                  type: string
+                description: NodeSelector to target subset of worker nodes running
+                  this service
+                type: object
               replicas:
                 default: 1
                 description: Size of the galera cluster deployment
diff --git a/api/v1beta1/galera_types.go b/api/v1beta1/galera_types.go
index 93786b99..be1e3afa 100644
--- a/api/v1beta1/galera_types.go
+++ b/api/v1beta1/galera_types.go
@@ -41,6 +41,9 @@ type GaleraSpec struct {
 	// Size of the galera cluster deployment
 	Replicas int32 `json:"replicas"`
 	// +kubebuilder:validation:Optional
+	// NodeSelector to target subset of worker nodes running this service
+	NodeSelector map[string]string `json:"nodeSelector,omitempty"`
+	// +kubebuilder:validation:Optional
 	// Adoption configuration
 	AdoptionRedirect AdoptionRedirectSpec `json:"adoptionRedirect"`
 }
diff --git a/api/v1beta1/zz_generated.deepcopy.go b/api/v1beta1/zz_generated.deepcopy.go
index 989b2a90..bc2e2bee 100644
--- a/api/v1beta1/zz_generated.deepcopy.go
+++ b/api/v1beta1/zz_generated.deepcopy.go
@@ -46,7 +46,7 @@ func (in *Galera) DeepCopyInto(out *Galera) {
 	*out = *in
 	out.TypeMeta = in.TypeMeta
 	in.ObjectMeta.DeepCopyInto(&out.ObjectMeta)
-	out.Spec = in.Spec
+	in.Spec.DeepCopyInto(&out.Spec)
 	in.Status.DeepCopyInto(&out.Status)
 }
 
@@ -118,6 +118,13 @@ func (in *GaleraList) DeepCopyObject() runtime.Object {
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *GaleraSpec) DeepCopyInto(out *GaleraSpec) {
 	*out = *in
+	if in.NodeSelector != nil {
+		in, out := &in.NodeSelector, &out.NodeSelector
+		*out = make(map[string]string, len(*in))
+		for key, val := range *in {
+			(*out)[key] = val
+		}
+	}
 	out.AdoptionRedirect = in.AdoptionRedirect
 }
 
diff --git a/config/crd/bases/mariadb.openstack.org_galeras.yaml b/config/crd/bases/mariadb.openstack.org_galeras.yaml
index e7f8b004..2b7632f8 100644
--- a/config/crd/bases/mariadb.openstack.org_galeras.yaml
+++ b/config/crd/bases/mariadb.openstack.org_galeras.yaml
@@ -55,6 +55,12 @@ spec:
                 default: quay.io/tripleozedcentos9/openstack-mariadb:current-tripleo
                 description: Name of the galera container image to run
                 type: string
+              nodeSelector:
+                additionalProperties:
+                  type: string
+                description: NodeSelector to target subset of worker nodes running
+                  this service
+                type: object
               replicas:
                 default: 1
                 description: Size of the galera cluster deployment
diff --git a/pkg/statefulset.go b/pkg/statefulset.go
index 7f3828e6..2333e69d 100644
--- a/pkg/statefulset.go
+++ b/pkg/statefulset.go
@@ -1,6 +1,8 @@
 package mariadb
 
 import (
+	common "github.com/openstack-k8s-operators/lib-common/modules/common"
+	"github.com/openstack-k8s-operators/lib-common/modules/common/affinity"
 	mariadbv1 "github.com/openstack-k8s-operators/mariadb-operator/api/v1beta1"
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
@@ -16,7 +18,7 @@ func StatefulSet(g *mariadbv1.Galera) *appsv1.StatefulSet {
 	runAsUser := int64(0)
 	storage := g.Spec.StorageClass
 	storageRequest := resource.MustParse(g.Spec.StorageRequest)
-	dep := &appsv1.StatefulSet{
+	sts := &appsv1.StatefulSet{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      name,
 			Namespace: g.Namespace,
@@ -259,5 +261,20 @@ func StatefulSet(g *mariadbv1.Galera) *appsv1.StatefulSet {
 			},
 		},
 	}
-	return dep
+
+	// If possible two pods of the same service should not
+	// run on the same worker node. If this is not possible
+	// the get still created on the same worker node.
+	sts.Spec.Template.Spec.Affinity = affinity.DistributePods(
+		common.AppSelector,
+		[]string{
+			name,
+		},
+		corev1.LabelHostname,
+	)
+	if g.Spec.NodeSelector != nil && len(g.Spec.NodeSelector) > 0 {
+		sts.Spec.Template.Spec.NodeSelector = g.Spec.NodeSelector
+	}
+
+	return sts
 }