create Pod Disruption Budget for DCA and CCR deployments

api/datadoghq/v2alpha1/datadogagent_types.go

@@ -1362,6 +1362,10 @@ type DatadogAgentComponentOverride struct {
 	// +optional
 	Replicas *int32 `json:"replicas,omitempty"`
 
+	// Set CreatePodDisruptionBudget to true to create a PodDisruptionBudget for this component.
+	// +optional
+	CreatePodDisruptionBudget *bool `json:"createPodDisruptionBudget,omitempty"`
+
 	// Set CreateRbac to false to prevent automatic creation of Role/ClusterRole for this component
 	// +optional
 	CreateRbac *bool `json:"createRbac,omitempty"`

config/crd/bases/v1/datadoghq.com_datadogagents.yaml

@@ -3567,6 +3567,9 @@ spec:
                           `security-agent`, `system-probe`, `trace-agent`, and `all`.
                           Configuration under `all` applies to all configured containers.
                         type: object
+                      createPodDisruptionBudget:
+                        description: Set CreatePodDisruptionBudget to true to create a PodDisruptionBudget for this component.
+                        type: boolean
                       createRbac:
                         description: Set CreateRbac to false to prevent automatic creation of Role/ClusterRole for this component
                         type: boolean

docs/configuration.v2alpha1.md

@@ -331,6 +331,7 @@ In the table, `spec.override.nodeAgent.image.name` and `spec.override.nodeAgent.
 | [key].containers.[key].securityContext.windowsOptions.hostProcess | HostProcess determines if a container should be run as a 'Host Process' container. All of a Pod's containers must have the same effective HostProcess value (it is not allowed to have a mix of HostProcess containers and non-HostProcess containers). In addition, if HostProcess is true then HostNetwork must also be set to true. |
 | [key].containers.[key].securityContext.windowsOptions.runAsUserName | The UserName in Windows to run the entrypoint of the container process. Defaults to the user specified in image metadata if unspecified. May also be set in PodSecurityContext. If set in both SecurityContext and PodSecurityContext, the value specified in SecurityContext takes precedence. |
 | [key].containers.[key].volumeMounts `[]object` | Specify additional volume mounts in the container. |
+| [key].createPodDisruptionBudget | Set CreatePodDisruptionBudget to true to create a PodDisruptionBudget for this component. |
 | [key].createRbac | Set CreateRbac to false to prevent automatic creation of Role/ClusterRole for this component |
 | [key].customConfigurations `map[string]object` | CustomConfiguration allows to specify custom configuration files for `datadog.yaml`, `datadog-cluster.yaml`, `security-agent.yaml`, and `system-probe.yaml`. The content is merged with configuration generated by the Datadog Operator, with priority given to custom configuration. WARNING: It is possible to override values set in the `DatadogAgent`. |
 | [key].customConfigurations.[key].configData | ConfigData corresponds to the configuration file content. |

internal/controller/datadogagent/component/clusteragent/utils.go

@@ -15,11 +15,16 @@ import (
 	"github.com/DataDog/datadog-operator/pkg/controller/utils/comparison"
 
 	corev1 "k8s.io/api/core/v1"
+	policyv1 "k8s.io/api/policy/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/intstr"
 	"k8s.io/apimachinery/pkg/version"
 )
 
+const (
+	pdbMinAvailableInstances = 1
+)
+
 // GetClusterAgentService returns the Cluster-Agent service
 func GetClusterAgentService(dda metav1.Object) *corev1.Service {
 	labels := object.GetDefaultLabels(dda, v2alpha1.DefaultClusterAgentResourceSuffix, GetClusterAgentVersion(dda))
@@ -53,6 +58,27 @@ func GetClusterAgentService(dda metav1.Object) *corev1.Service {
 	return service
 }
 
+func GetClusterAgentPodDisruptionBudget(dda metav1.Object) *policyv1.PodDisruptionBudget {
+	// labels and annotations
+	minAvailableStr := intstr.FromInt(pdbMinAvailableInstances)
+	matchLabels := map[string]string{
+		apicommon.AgentDeploymentNameLabelKey:      dda.GetName(),
+		apicommon.AgentDeploymentComponentLabelKey: v2alpha1.DefaultClusterAgentResourceSuffix}
+	pdb := &policyv1.PodDisruptionBudget{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "datadog-cluster-agent-pdb",
+			Namespace: dda.GetNamespace(),
+		},
+		Spec: policyv1.PodDisruptionBudgetSpec{
+			MinAvailable: &minAvailableStr,
+			Selector: &metav1.LabelSelector{
+				MatchLabels: matchLabels,
+			},
+		},
+	}
+	return pdb
+}
+
 // GetMetricsServerServiceName returns the external metrics provider service name
 func GetMetricsServerServiceName(dda metav1.Object) string {
 	return fmt.Sprintf("%s-%s", dda.GetName(), v2alpha1.DefaultMetricsServerResourceSuffix)

internal/controller/datadogagent/component/clusterchecksrunner/default.go

@@ -11,7 +11,9 @@
 
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
+	policyv1 "k8s.io/api/policy/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
 
 	apicommon "github.com/DataDog/datadog-operator/api/datadoghq/common"
 	"github.com/DataDog/datadog-operator/api/datadoghq/v2alpha1"
@@ -21,6 +23,10 @@
 	"github.com/DataDog/datadog-operator/pkg/defaulting"
 )
 
+const (
+	pdMaxUnavailableInstances = 1
+)
+
 // GetClusterChecksRunnerName return the Cluster-Checks-Runner name based on the DatadogAgent name
 func GetClusterChecksRunnerName(dda metav1.Object) string {
 	return fmt.Sprintf("%s-%s", dda.GetName(), v2alpha1.DefaultClusterChecksRunnerResourceSuffix)
@@ -82,6 +88,27 @@
 	return template
 }
 
+func GetClusterChecksRunnerPodDisruptionBudget(dda metav1.Object) *policyv1.PodDisruptionBudget {
+	maxUnavailableStr := intstr.FromInt(pdMaxUnavailableInstances)
+	matchLabels := map[string]string{
+		apicommon.AgentDeploymentNameLabelKey:      dda.GetName(),
+		apicommon.AgentDeploymentComponentLabelKey: v2alpha1.DefaultClusterChecksRunnerResourceSuffix}
+	pdb := &policyv1.PodDisruptionBudget{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "datadog-cluster-checks-runner-pdb",
+			Namespace: dda.GetNamespace(),
+		},
+		Spec: policyv1.PodDisruptionBudgetSpec{
+			MaxUnavailable: &maxUnavailableStr,
+			Selector: &metav1.LabelSelector{
+				MatchLabels: matchLabels,
+			},
+		},
+	}
+	fmt.Println("in GetClusterAgentPodDisruptionBudget, ", dda.GetNamespace())
+	return pdb
+}
+
 // getDefaultServiceAccountName return the default Cluster-Agent ServiceAccountName
 func getDefaultServiceAccountName(dda metav1.Object) string {
 	return fmt.Sprintf("%s-%s", dda.GetName(), v2alpha1.DefaultClusterChecksRunnerResourceSuffix)

internal/controller/datadogagent/controller_reconcile_v2.go

@@ -32,7 +32,7 @@
 
 func (r *Reconciler) internalReconcileV2(ctx context.Context, request reconcile.Request) (reconcile.Result, error) {
 	reqLogger := r.log.WithValues("datadogagent", request.NamespacedName)
-	reqLogger.Info("Reconciling DatadogAgent")
+	reqLogger.Info("Reconciling DatadogAgent1")
 
 	// Fetch the DatadogAgent instance
 	instance := &datadoghqv2alpha1.DatadogAgent{}
@@ -46,10 +46,12 @@
 			return result, nil
 		}
 		// Error reading the object - requeue the request.
+		reqLogger.Error(err, "unable to fetch DatadogAgent")
 		return result, err
 	}
 
 	if instance.Spec.Global == nil || instance.Spec.Global.Credentials == nil {
+		reqLogger.Info("credentials not configured in the DatadogAgent, can't reconcile")
 		return result, fmt.Errorf("credentials not configured in the DatadogAgent, can't reconcile")
 	}
 
@@ -74,6 +76,7 @@
 	}*/
 
 	if result, err = r.handleFinalizer(reqLogger, instance, r.finalizeDadV2); utils.ShouldReturn(result, err) {
+		reqLogger.V(1).Info("finalizer error", "error", err)
 		return result, err
 	}
 
@@ -88,16 +91,19 @@
 	// Set default values for GlobalConfig and Features
 	instanceCopy := instance.DeepCopy()
 	datadoghqv2alpha1.DefaultDatadogAgent(instanceCopy)
-
+	reqLogger.Info("reconcile instance")
 	return r.reconcileInstanceV2(ctx, reqLogger, instanceCopy)
 }
 
 func (r *Reconciler) reconcileInstanceV2(ctx context.Context, logger logr.Logger, instance *datadoghqv2alpha1.DatadogAgent) (reconcile.Result, error) {
 	var result reconcile.Result
 	newStatus := instance.Status.DeepCopy()
 	now := metav1.NewTime(time.Now())
-
+	// logger = logger.WithValues("datadogagent", instance.Name, "namespace", instance.Namespace)
+	fmt.Println("ReconcileInstanceV2")
+	logger.Info("ReconcileInstanceV2")
 	features, requiredComponents := feature.BuildFeatures(instance, reconcilerOptionsToFeatureOptions(&r.options, logger))
+	logger.Info("ReconcileInstanceV2", "features", features)
 	// update list of enabled features for metrics forwarder
 	r.updateMetricsForwardersFeatures(instance, features)
 
@@ -117,7 +123,9 @@
 	var errs []error
 
 	// Set up dependencies required by enabled features
 	for _, feat := range features {
+		fmt.Println("Dependency ManageDependencies", "featureID", feat.ID())
+		logger.Info("Dependency ManageDependencies", "featureID", feat.ID())
 		logger.V(1).Info("Dependency ManageDependencies", "featureID", feat.ID())
 		if featErr := feat.ManageDependencies(resourceManagers, requiredComponents); featErr != nil {
 			errs = append(errs, featErr)

internal/controller/datadogagent/feature/enabledefault/feature.go

@@ -67,6 +67,7 @@
 	logger                  logr.Logger
 	disableNonResourceRules bool
 	otelAgentEnabled        bool
+	PodDisruptionBudget     bool
 
 	customConfigAnnotationKey   string
 	customConfigAnnotationValue string
@@ -123,7 +124,6 @@
 		if dda.Spec.Global.DisableNonResourceRules != nil && *dda.Spec.Global.DisableNonResourceRules {
 			f.disableNonResourceRules = true
 		}
-
 		if dda.Spec.Global.Credentials != nil {
 			creds := dda.Spec.Global.Credentials
 
@@ -208,7 +208,6 @@
 			},
 		}
 	}
-
 }
 
 // ManageDependencies allows a feature to manage its dependencies.
@@ -250,6 +249,8 @@
 	}
 
 	if components.ClusterAgent.IsEnabled() {
+		f.logger.Info("Cluster Agent is enabled")
+		fmt.Println("Cluster Agent is enabled")
 		if err := f.clusterAgentDependencies(managers, components.ClusterAgent); err != nil {
 			errs = append(errs, err)
 		}

internal/controller/datadogagent/override/dependencies.go

@@ -14,6 +14,8 @@
 	"k8s.io/apimachinery/pkg/util/errors"
 
 	"github.com/DataDog/datadog-operator/api/datadoghq/v2alpha1"
+	componentdca "github.com/DataDog/datadog-operator/internal/controller/datadogagent/component/clusteragent"
+	componentccr "github.com/DataDog/datadog-operator/internal/controller/datadogagent/component/clusterchecksrunner"
 	"github.com/DataDog/datadog-operator/internal/controller/datadogagent/feature"
 	"github.com/DataDog/datadog-operator/internal/controller/datadogagent/object"
 	"github.com/DataDog/datadog-operator/internal/controller/datadogagent/object/configmap"
@@ -42,6 +44,28 @@
 		// Handle custom check files
 		checksdCMName := fmt.Sprintf(extraChecksdConfigMapName, strings.ToLower((string(component))))
 		errs = append(errs, overrideExtraConfigs(logger, manager, override.ExtraChecksd, namespace, checksdCMName, false)...)
+
+		if override.CreatePodDisruptionBudget != nil {
+			fmt.Println("override.CreatePodDisruptionBudget is not nil for", component)
+			if component == v2alpha1.ClusterAgentComponentName {
+				pdb := componentdca.GetClusterAgentPodDisruptionBudget(dda)
+				if err := manager.Store().AddOrUpdate(kubernetes.PodDisruptionBudgetsKind, pdb); err != nil {
+					fmt.Println("error with created pod disruption budget")
+					errs = append(errs, err)
+				} else {
+					fmt.Println("created pod disruption budget")
+				}
+
+			} else if component == v2alpha1.ClusterChecksRunnerComponentName {
+				pdb := componentccr.GetClusterChecksRunnerPodDisruptionBudget(dda)
+				if err := manager.Store().AddOrUpdate(kubernetes.PodDisruptionBudgetsKind, pdb); err != nil {
+					fmt.Println("error with created pod disruption budget")
+					errs = append(errs, err)
+				} else {
+					fmt.Println("created pod disruption budget")
+				}
+			}
+		}
 	}
 
 	return errs