Pod affinity

pkg/deployment/pods.go

@@ -285,7 +285,6 @@ func (d *Deployment) createReadinessProbe(apiObject *api.ArangoDeployment, group
 // ensurePods creates all Pods listed in member status
 func (d *Deployment) ensurePods(apiObject *api.ArangoDeployment) error {
 	kubecli := d.deps.KubeCli
-	owner := apiObject.AsOwner()
 
 	if err := apiObject.ForeachServerGroup(func(group api.ServerGroup, spec api.ServerGroupSpec, status *api.MemberStatusList) error {
 		for _, m := range *status {
@@ -305,7 +304,7 @@ func (d *Deployment) ensurePods(apiObject *api.ArangoDeployment) error {
 				if err != nil {
 					return maskAny(err)
 				}
-				if err := k8sutil.CreateArangodPod(kubecli, apiObject, role, m.ID, m.PersistentVolumeClaimName, apiObject.Spec.Image, apiObject.Spec.ImagePullPolicy, args, env, livenessProbe, readinessProbe, owner); err != nil {
+				if err := k8sutil.CreateArangodPod(kubecli, apiObject.Spec.IsDevelopment(), apiObject, role, m.ID, m.PersistentVolumeClaimName, apiObject.Spec.Image, apiObject.Spec.ImagePullPolicy, args, env, livenessProbe, readinessProbe); err != nil {
 					return maskAny(err)
 				}
 			} else if group.IsArangosync() {
@@ -315,7 +314,11 @@ func (d *Deployment) ensurePods(apiObject *api.ArangoDeployment) error {
 				if err != nil {
 					return maskAny(err)
 				}
-				if err := k8sutil.CreateArangoSyncPod(kubecli, apiObject, role, m.ID, apiObject.Spec.Sync.Image, apiObject.Spec.Sync.ImagePullPolicy, args, env, livenessProbe, owner); err != nil {
+				affinityWithRole := ""
+				if group == api.ServerGroupSyncWorkers {
+					affinityWithRole = api.ServerGroupDBServers.AsRole()
+				}
+				if err := k8sutil.CreateArangoSyncPod(kubecli, apiObject.Spec.IsDevelopment(), apiObject, role, m.ID, apiObject.Spec.Sync.Image, apiObject.Spec.Sync.ImagePullPolicy, args, env, livenessProbe, affinityWithRole); err != nil {
 					return maskAny(err)
 				}
 			}

pkg/util/k8sutil/affinity.go

@@ -0,0 +1,76 @@
+//
+// DISCLAIMER
+//
+// Copyright 2018 ArangoDB GmbH, Cologne, Germany
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// Copyright holder is ArangoDB GmbH, Cologne, Germany
+//
+// Author Ewout Prangsma
+//
+
+package k8sutil
+
+import (
+	"k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+// createAffinity creates pod anti-affinity for the given role.
+// role contains the name of the role to configure any-affinity with.
+// affinityWithRole contains the role to configure affinity with.
+func createAffinity(deploymentName, role string, required bool, affinityWithRole string) *v1.Affinity {
+	a := &v1.Affinity{
+		PodAntiAffinity: &v1.PodAntiAffinity{},
+	}
+	labels := LabelsForDeployment(deploymentName, role)
+	labelSelector := &metav1.LabelSelector{
+		MatchLabels: labels,
+	}
+	if required {
+		a.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution = append(a.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution, v1.PodAffinityTerm{
+			LabelSelector: labelSelector,
+			TopologyKey:   TopologyKeyHostname,
+		})
+	} else {
+		a.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution = append(a.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution, v1.WeightedPodAffinityTerm{
+			Weight: 1,
+			PodAffinityTerm: v1.PodAffinityTerm{
+				LabelSelector: labelSelector,
+				TopologyKey:   TopologyKeyHostname,
+			},
+		})
+	}
+	if affinityWithRole != "" {
+		a.PodAffinity = &v1.PodAffinity{}
+		labelSelector := &metav1.LabelSelector{
+			MatchLabels: LabelsForDeployment(deploymentName, affinityWithRole),
+		}
+		if required {
+			a.PodAffinity.RequiredDuringSchedulingIgnoredDuringExecution = append(a.PodAffinity.RequiredDuringSchedulingIgnoredDuringExecution, v1.PodAffinityTerm{
+				LabelSelector: labelSelector,
+				TopologyKey:   TopologyKeyHostname,
+			})
+		} else {
+			a.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution = append(a.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution, v1.WeightedPodAffinityTerm{
+				Weight: 1,
+				PodAffinityTerm: v1.PodAffinityTerm{
+					LabelSelector: labelSelector,
+					TopologyKey:   TopologyKeyHostname,
+				},
+			})
+		}
+	}
+	return a
+}

pkg/util/k8sutil/constants.go

@@ -29,4 +29,5 @@ const (
 	// K8s constants
 	ClusterIPNone                      = "None"
 	TolerateUnreadyEndpointsAnnotation = "service.alpha.kubernetes.io/tolerate-unready-endpoints"
+	TopologyKeyHostname                = "kubernetes.io/hostname"
 )

pkg/util/k8sutil/pods.go

@@ -125,8 +125,8 @@ func newPod(deploymentName, ns, role, id string) v1.Pod {
 // CreateArangodPod creates a Pod that runs `arangod`.
 // If the pod already exists, nil is returned.
 // If another error occurs, that error is returned.
-func CreateArangodPod(kubecli kubernetes.Interface, deployment metav1.Object, role, id, pvcName, image string, imagePullPolicy v1.PullPolicy,
-	args []string, env map[string]string, livenessProbe *HTTPProbeConfig, readinessProbe *HTTPProbeConfig, owner metav1.OwnerReference) error {
+func CreateArangodPod(kubecli kubernetes.Interface, developmentMode bool, deployment APIObject, role, id, pvcName, image string, imagePullPolicy v1.PullPolicy,
+	args []string, env map[string]string, livenessProbe *HTTPProbeConfig, readinessProbe *HTTPProbeConfig) error {
 	// Prepare basic pod
 	p := newPod(deployment.GetName(), deployment.GetNamespace(), role, id)
 
@@ -157,7 +157,10 @@ func CreateArangodPod(kubecli kubernetes.Interface, deployment metav1.Object, ro
 		p.Spec.Volumes = append(p.Spec.Volumes, vol)
 	}
 
-	if err := createPod(kubecli, &p, deployment.GetNamespace(), owner); err != nil {
+	// Add (anti-)affinity
+	p.Spec.Affinity = createAffinity(deployment.GetName(), role, !developmentMode, "")
+
+	if err := createPod(kubecli, &p, deployment.GetNamespace(), deployment.AsOwner()); err != nil {
 		return maskAny(err)
 	}
 	return nil
@@ -166,16 +169,19 @@ func CreateArangodPod(kubecli kubernetes.Interface, deployment metav1.Object, ro
 // CreateArangoSyncPod creates a Pod that runs `arangosync`.
 // If the pod already exists, nil is returned.
 // If another error occurs, that error is returned.
-func CreateArangoSyncPod(kubecli kubernetes.Interface, deployment metav1.Object, role, id, image string, imagePullPolicy v1.PullPolicy,
-	args []string, env map[string]string, livenessProbe *HTTPProbeConfig, owner metav1.OwnerReference) error {
+func CreateArangoSyncPod(kubecli kubernetes.Interface, developmentMode bool, deployment APIObject, role, id, image string, imagePullPolicy v1.PullPolicy,
+	args []string, env map[string]string, livenessProbe *HTTPProbeConfig, affinityWithRole string) error {
 	// Prepare basic pod
 	p := newPod(deployment.GetName(), deployment.GetNamespace(), role, id)
 
 	// Add arangosync container
 	c := arangosyncContainer(p.GetName(), image, imagePullPolicy, args, env, livenessProbe)
 	p.Spec.Containers = append(p.Spec.Containers, c)
 
-	if err := createPod(kubecli, &p, deployment.GetNamespace(), owner); err != nil {
+	// Add (anti-)affinity
+	p.Spec.Affinity = createAffinity(deployment.GetName(), role, !developmentMode, affinityWithRole)
+
+	if err := createPod(kubecli, &p, deployment.GetNamespace(), deployment.AsOwner()); err != nil {
 		return maskAny(err)
 	}
 	return nil