Move the startup probe configuration for the sidecar in the ObjectStore config

Signed-off-by: Tudor Golubenco <tudor@xata.io>

Author: tsg

api/v1/objectstore_types.go

@@ -22,6 +22,35 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
+// ProbeConfig holds configuration for probe timing and thresholds
+// This is a subset of the corev1.Probe type, with only the fields that we want to expose as configuration.
+type ProbeConfig struct {
+	// InitialDelaySeconds is the number of seconds after the container has started before startup probes are initiated.
+	// +kubebuilder:default:=0
+	// +optional
+	InitialDelaySeconds int32 `json:"initialDelaySeconds,omitempty"`
+
+	// TimeoutSeconds is the number of seconds after which the probe times out.
+	// +kubebuilder:default:=10
+	// +optional
+	TimeoutSeconds int32 `json:"timeoutSeconds,omitempty"`
+
+	// PeriodSeconds is how often (in seconds) to perform the probe.
+	// +kubebuilder:default:=10
+	// +optional
+	PeriodSeconds int32 `json:"periodSeconds,omitempty"`
+
+	// SuccessThreshold is the minimum consecutive successes for the probe to be considered successful.
+	// +kubebuilder:default:=1
+	// +optional
+	SuccessThreshold int32 `json:"successThreshold,omitempty"`
+
+	// FailureThreshold is the minimum consecutive failures for the probe to be considered failed.
+	// +kubebuilder:default:=10
+	// +optional
+	FailureThreshold int32 `json:"failureThreshold,omitempty"`
+}
+
 // InstanceSidecarConfiguration defines the configuration for the sidecar that runs in the instance pods.
 type InstanceSidecarConfiguration struct {
 	// The environment to be explicitly passed to the sidecar
@@ -37,6 +66,10 @@ type InstanceSidecarConfiguration struct {
 	// Resources define cpu/memory requests and limits for the sidecar that runs in the instance pods.
 	// +optional
 	Resources corev1.ResourceRequirements `json:"resources,omitempty"`
+
+	// StartupProbe defines the configuration for the startup probe of the sidecar container.
+	// +optional
+	StartupProbe *ProbeConfig `json:"startupProbe,omitempty"`
 }
 
 // ObjectStoreSpec defines the desired state of ObjectStore.

api/v1/zz_generated.deepcopy.go

@@ -577,6 +577,42 @@ spec:
                       The retentionCheckInterval defines the frequency at which the
                       system checks and enforces retention policies.
                     type: integer
+                  startupProbe:
+                    description: StartupProbe defines the configuration for the startup
+                      probe of the sidecar container.
+                    properties:
+                      failureThreshold:
+                        default: 10
+                        description: FailureThreshold is the minimum consecutive failures
+                          for the probe to be considered failed.
+                        format: int32
+                        type: integer
+                      initialDelaySeconds:
+                        default: 0
+                        description: InitialDelaySeconds is the number of seconds
+                          after the container has started before startup probes are
+                          initiated.
+                        format: int32
+                        type: integer
+                      periodSeconds:
+                        default: 10
+                        description: PeriodSeconds is how often (in seconds) to perform
+                          the probe.
+                        format: int32
+                        type: integer
+                      successThreshold:
+                        default: 1
+                        description: SuccessThreshold is the minimum consecutive successes
+                          for the probe to be considered successful.
+                        format: int32
+                        type: integer
+                      timeoutSeconds:
+                        default: 10
+                        description: TimeoutSeconds is the number of seconds after
+                          which the probe times out.
+                        format: int32
+                        type: integer
+                    type: object
                 type: object
               retentionPolicy:
                 description: |-

config/crd/bases/barmancloud.cnpg.io_objectstores.yaml

@@ -13,6 +13,12 @@ spec:
       limits:
         memory: "512Mi"
         cpu: "500m"
+    startupProbe:
+      initialDelaySeconds: 1
+      timeoutSeconds: 10
+      periodSeconds: 1
+      failureThreshold: 10
+      successThreshold: 1
   configuration:
     endpointCA:
       name: minio-server-tls

hack/examples/minio-store.yaml

@@ -1,7 +1,6 @@
 package config
 
 import (
-	"strconv"
 	"strings"
 
 	cnpgv1 "github.com/cloudnative-pg/cloudnative-pg/api/v1"
@@ -58,29 +57,6 @@ type PluginConfiguration struct {
 
 	ReplicaSourceBarmanObjectName string
 	ReplicaSourceServerName       string
-
-	// Probe configuration
-	StartupProbeConfig *ProbeConfig
-}
-
-// ProbeConfig holds configuration for Kubernetes probes
-type ProbeConfig struct {
-	InitialDelaySeconds int32
-	TimeoutSeconds      int32
-	PeriodSeconds       int32
-	FailureThreshold    int32
-	SuccessThreshold    int32
-}
-
-// DefaultProbeConfig returns the default probe configuration
-func DefaultProbeConfig() *ProbeConfig {
-	return &ProbeConfig{
-		InitialDelaySeconds: 0,
-		TimeoutSeconds:      10,
-		PeriodSeconds:       10,
-		FailureThreshold:    10,
-		SuccessThreshold:    1,
-	}
 }
 
 // GetBarmanObjectKey gets the namespaced name of the barman object
@@ -190,50 +166,11 @@ func NewFromCluster(cluster *cnpgv1.Cluster) *PluginConfiguration {
 		// used for wal_restore in the designed primary of a replica cluster
 		ReplicaSourceServerName:       replicaSourceServerName,
 		ReplicaSourceBarmanObjectName: replicaSourceBarmanObjectName,
-		// probe configuration
-		StartupProbeConfig: parseProbeConfig(helper.Parameters),
 	}
 
 	return result
 }
 
-// parseProbeConfig parses probe configuration from plugin parameters
-func parseProbeConfig(parameters map[string]string) *ProbeConfig {
-	config := DefaultProbeConfig()
-
-	if val, ok := parameters["startupProbe.initialDelaySeconds"]; ok {
-		if parsed, err := strconv.ParseInt(val, 10, 32); err == nil {
-			config.InitialDelaySeconds = int32(parsed)
-		}
-	}
-
-	if val, ok := parameters["startupProbe.timeoutSeconds"]; ok {
-		if parsed, err := strconv.ParseInt(val, 10, 32); err == nil {
-			config.TimeoutSeconds = int32(parsed)
-		}
-	}
-
-	if val, ok := parameters["startupProbe.periodSeconds"]; ok {
-		if parsed, err := strconv.ParseInt(val, 10, 32); err == nil {
-			config.PeriodSeconds = int32(parsed)
-		}
-	}
-
-	if val, ok := parameters["startupProbe.failureThreshold"]; ok {
-		if parsed, err := strconv.ParseInt(val, 10, 32); err == nil {
-			config.FailureThreshold = int32(parsed)
-		}
-	}
-
-	if val, ok := parameters["startupProbe.successThreshold"]; ok {
-		if parsed, err := strconv.ParseInt(val, 10, 32); err == nil {
-			config.SuccessThreshold = int32(parsed)
-		}
-	}
-
-	return config
-}
-
 func getRecoveryParameters(cluster *cnpgv1.Cluster) map[string]string {
 	recoveryPluginConfiguration := getRecoverySourcePlugin(cluster)
 	if recoveryPluginConfiguration == nil {

internal/cnpgi/operator/config/config.go

@@ -17,6 +17,7 @@ import (
 	"k8s.io/utils/ptr"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
+	barmancloudv1 "github.com/cloudnative-pg/plugin-barman-cloud/api/v1"
 	"github.com/cloudnative-pg/plugin-barman-cloud/internal/cnpgi/metadata"
 	"github.com/cloudnative-pg/plugin-barman-cloud/internal/cnpgi/operator/config"
 )
@@ -125,19 +126,24 @@ func (impl LifecycleImplementation) reconcileJob(
 		return nil, err
 	}
 
+	startupProbe, err := impl.collectSidecarStartupProbeForRecoveryJob(ctx, pluginConfiguration)
+	if err != nil {
+		return nil, err
+	}
+
 	return reconcileJob(ctx, cluster, request, sidecarConfiguration{
 		env:          env,
 		certificates: certificates,
 		resources:    resources,
-		probeConfig:  pluginConfiguration.StartupProbeConfig,
+		startupProbe: startupProbe,
 	})
 }
 
 type sidecarConfiguration struct {
 	env          []corev1.EnvVar
 	certificates []corev1.VolumeProjection
 	resources    corev1.ResourceRequirements
-	probeConfig  *config.ProbeConfig
+	startupProbe *barmancloudv1.ProbeConfig
 }
 
 func reconcileJob(
@@ -219,11 +225,16 @@ func (impl LifecycleImplementation) reconcilePod(
 		return nil, err
 	}
 
+	startupProbe, err := impl.collectSidecarStartupProbeForInstancePod(ctx, pluginConfiguration)
+	if err != nil {
+		return nil, err
+	}
+
 	return reconcilePod(ctx, cluster, request, pluginConfiguration, sidecarConfiguration{
 		env:          env,
 		certificates: certificates,
 		resources:    resources,
-		probeConfig:  pluginConfiguration.StartupProbeConfig,
+		startupProbe: startupProbe,
 	})
 }
 
@@ -315,12 +326,13 @@ func reconcilePodSpec(
 	}
 
 	// Apply configurable probe settings if available
-	if config.probeConfig != nil {
-		baseProbe.InitialDelaySeconds = config.probeConfig.InitialDelaySeconds
-		baseProbe.TimeoutSeconds = config.probeConfig.TimeoutSeconds
-		baseProbe.PeriodSeconds = config.probeConfig.PeriodSeconds
-		baseProbe.FailureThreshold = config.probeConfig.FailureThreshold
-		baseProbe.SuccessThreshold = config.probeConfig.SuccessThreshold
+	if config.startupProbe != nil {
+		// Copy timing and threshold settings from user configuration
+		baseProbe.InitialDelaySeconds = config.startupProbe.InitialDelaySeconds
+		baseProbe.TimeoutSeconds = config.startupProbe.TimeoutSeconds
+		baseProbe.PeriodSeconds = config.startupProbe.PeriodSeconds
+		baseProbe.FailureThreshold = config.startupProbe.FailureThreshold
+		baseProbe.SuccessThreshold = config.startupProbe.SuccessThreshold
 	} else {
 		// Fallback to default values
 		baseProbe.FailureThreshold = 10

internal/cnpgi/operator/lifecycle.go

@@ -0,0 +1,59 @@
+package operator
+
+import (
+	"context"
+
+	barmancloudv1 "github.com/cloudnative-pg/plugin-barman-cloud/api/v1"
+	"github.com/cloudnative-pg/plugin-barman-cloud/internal/cnpgi/operator/config"
+)
+
+func (impl LifecycleImplementation) collectSidecarStartupProbeForRecoveryJob(
+	ctx context.Context,
+	configuration *config.PluginConfiguration,
+) (*barmancloudv1.ProbeConfig, error) {
+	if len(configuration.RecoveryBarmanObjectName) > 0 {
+		var barmanObjectStore barmancloudv1.ObjectStore
+		if err := impl.Client.Get(ctx, configuration.GetRecoveryBarmanObjectKey(), &barmanObjectStore); err != nil {
+			return nil, err
+		}
+
+		return barmanObjectStore.Spec.InstanceSidecarConfiguration.StartupProbe, nil
+	}
+
+	return nil, nil
+}
+
+func (impl LifecycleImplementation) collectSidecarStartupProbeForInstancePod(
+	ctx context.Context,
+	configuration *config.PluginConfiguration,
+) (*barmancloudv1.ProbeConfig, error) {
+	if len(configuration.BarmanObjectName) > 0 {
+		// On a replica cluster that also archives, the designated primary
+		// will use both the replica source object store and the object store
+		// of the cluster.
+		// In this case, we use the cluster object store for configuring
+		// the startup probe of the sidecar container.
+
+		var barmanObjectStore barmancloudv1.ObjectStore
+		if err := impl.Client.Get(ctx, configuration.GetBarmanObjectKey(), &barmanObjectStore); err != nil {
+			return nil, err
+		}
+
+		return barmanObjectStore.Spec.InstanceSidecarConfiguration.StartupProbe, nil
+	}
+
+	if len(configuration.RecoveryBarmanObjectName) > 0 {
+		// On a replica cluster that doesn't archive, the designated primary
+		// uses only the replica source object store.
+		// In this case, we use the replica source object store for configuring
+		// the startup probe of the sidecar container.
+		var barmanObjectStore barmancloudv1.ObjectStore
+		if err := impl.Client.Get(ctx, configuration.GetRecoveryBarmanObjectKey(), &barmanObjectStore); err != nil {
+			return nil, err
+		}
+
+		return barmanObjectStore.Spec.InstanceSidecarConfiguration.StartupProbe, nil
+	}
+
+	return nil, nil
+}

internal/cnpgi/operator/lifecycle_probes.go

@@ -10,6 +10,7 @@ import (
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
+	barmancloudv1 "github.com/cloudnative-pg/plugin-barman-cloud/api/v1"
 	"github.com/cloudnative-pg/plugin-barman-cloud/internal/cnpgi/operator/config"
 
 	. "github.com/onsi/ginkgo/v2"
@@ -173,8 +174,8 @@ var _ = Describe("LifecycleImplementation", func() {
 
 	Describe("reconcilePod", func() {
 		It("returns a patch for a valid pod with probe configuration", func(ctx SpecContext) {
-			// Configure plugin with custom probe settings
-			pluginConfiguration.StartupProbeConfig = &config.ProbeConfig{
+			// Configure sidecar with custom probe settings
+			startupProbeConfig := &barmancloudv1.ProbeConfig{
 				InitialDelaySeconds: 1,
 				TimeoutSeconds:      15,
 				PeriodSeconds:       2,
@@ -204,7 +205,7 @@ var _ = Describe("LifecycleImplementation", func() {
 			}
 
 			response, err := reconcilePod(ctx, cluster, request, pluginConfiguration, sidecarConfiguration{
-				probeConfig: pluginConfiguration.StartupProbeConfig,
+				startupProbe: startupProbeConfig,
 			})
 			Expect(err).NotTo(HaveOccurred())
 			Expect(response).NotTo(BeNil())