Use pod name as cluster check runner ID (#326)

DataDog · Jun 24, 2021 · 0a9f32b · 0a9f32b
1 parent 022a88e
commit 0a9f32b
Show file tree

Hide file tree

Showing 4 changed files with 38 additions and 11 deletions.
diff --git a/api/v1alpha1/const.go b/api/v1alpha1/const.go
@@ -78,6 +78,7 @@ const (
 	DDKubeStateMetricsCoreConfigMap              = "DD_KUBE_STATE_METRICS_CORE_CONFIGMAP_NAME"
 	DDClcRunnerEnabled                           = "DD_CLC_RUNNER_ENABLED"
 	DDClcRunnerHost                              = "DD_CLC_RUNNER_HOST"
+	DDClcRunnerID                                = "DD_CLC_RUNNER_ID"
 	DDExtraConfigProviders                       = "DD_EXTRA_CONFIG_PROVIDERS"
 	DDExtraListeners                             = "DD_EXTRA_LISTENERS"
 	DDHostname                                   = "DD_HOSTNAME"

diff --git a/controllers/datadogagent/clusterchecksrunner.go b/controllers/datadogagent/clusterchecksrunner.go
@@ -389,6 +389,14 @@ func getEnvVarsForClusterChecksRunner(dda *datadoghqv1alpha1.DatadogAgent) []cor
 				},
 			},
 		},
+		{
+			Name: datadoghqv1alpha1.DDClcRunnerID,
+			ValueFrom: &corev1.EnvVarSource{
+				FieldRef: &corev1.ObjectFieldSelector{
+					FieldPath: FieldPathMetaName,
+				},
+			},
+		},
 	}
 
 	if spec.ClusterName != "" {
@@ -491,22 +499,26 @@ func getClusterChecksRunnerCustomConfigConfigMapName(dda *datadoghqv1alpha1.Data
 }
 
 // getPodAffinity returns the pod anti affinity of the cluster check runner pods
-// the default anti affinity ensures we don't schedule multiple cluster check runners on the same node
+// the default anti affinity prefers scheduling the runners on different nodes if possible
+// for better checks stability in case of node failure.
 func getPodAffinity(affinity *corev1.Affinity) *corev1.Affinity {
 	if affinity != nil {
 		return affinity
 	}
 
 	return &corev1.Affinity{
 		PodAntiAffinity: &corev1.PodAntiAffinity{
-			RequiredDuringSchedulingIgnoredDuringExecution: []corev1.PodAffinityTerm{
+			PreferredDuringSchedulingIgnoredDuringExecution: []corev1.WeightedPodAffinityTerm{
 				{
-					LabelSelector: &metav1.LabelSelector{
-						MatchLabels: map[string]string{
-							datadoghqv1alpha1.AgentDeploymentComponentLabelKey: datadoghqv1alpha1.DefaultClusterChecksRunnerResourceSuffix,
+					Weight: 50,
+					PodAffinityTerm: corev1.PodAffinityTerm{
+						LabelSelector: &metav1.LabelSelector{
+							MatchLabels: map[string]string{
+								datadoghqv1alpha1.AgentDeploymentComponentLabelKey: datadoghqv1alpha1.DefaultClusterChecksRunnerResourceSuffix,
+							},
 						},
+						TopologyKey: "kubernetes.io/hostname",
 					},
-					TopologyKey: "kubernetes.io/hostname",
 				},
 			},
 		},

diff --git a/controllers/datadogagent/clusterchecksrunner_test.go b/controllers/datadogagent/clusterchecksrunner_test.go
@@ -180,6 +180,14 @@ func clusterChecksRunnerDefaultEnvVars() []corev1.EnvVar {
 				},
 			},
 		},
+		{
+			Name: "DD_CLC_RUNNER_ID",
+			ValueFrom: &corev1.EnvVarSource{
+				FieldRef: &corev1.ObjectFieldSelector{
+					FieldPath: "metadata.name",
+				},
+			},
+		},
 	}
 }
 
@@ -380,14 +388,17 @@ func Test_getPodAffinity(t *testing.T) {
 			affinity: nil,
 			want: &corev1.Affinity{
 				PodAntiAffinity: &corev1.PodAntiAffinity{
-					RequiredDuringSchedulingIgnoredDuringExecution: []corev1.PodAffinityTerm{
+					PreferredDuringSchedulingIgnoredDuringExecution: []corev1.WeightedPodAffinityTerm{
 						{
-							LabelSelector: &metav1.LabelSelector{
-								MatchLabels: map[string]string{
-									"agent.datadoghq.com/component": "cluster-checks-runner",
+							Weight: 50,
+							PodAffinityTerm: corev1.PodAffinityTerm{
+								LabelSelector: &metav1.LabelSelector{
+									MatchLabels: map[string]string{
+										"agent.datadoghq.com/component": "cluster-checks-runner",
+									},
 								},
+								TopologyKey: "kubernetes.io/hostname",
 							},
-							TopologyKey: "kubernetes.io/hostname",
 						},
 					},
 				},

diff --git a/controllers/datadogagent/const.go b/controllers/datadogagent/const.go
@@ -15,6 +15,9 @@ const (
 	// FieldPathStatusPodIP used as FieldPath to retrieve the pod ip
 	FieldPathStatusPodIP = "status.podIP"
 
+	// FieldPathMetaName used as FieldPath to retrieve the pod name
+	FieldPathMetaName = "metadata.name"
+
 	// kind names definition
 	extendedDaemonSetKind   = "ExtendedDaemonSet"
 	daemonSetKind           = "DaemonSet"