Merge 3e93fae514 into bbc3eab7e0

Makefile

@@ -76,6 +76,7 @@ $(GENERATED_CRDS): $(GENERATED)
 	@$(SED) -i -e 's/listKind: PostgresqlList/listKind: postgresqlList/' manifests/postgresql.crd.yaml
 	@hack/adjust_postgresql_crd.sh
 	@mv manifests/acid.zalan.do_operatorconfigurations.yaml manifests/operatorconfiguration.crd.yaml
+	@hack/adjust_operatorconfiguration_crd.sh
 	@mv manifests/acid.zalan.do_postgresteams.yaml manifests/postgresteam.crd.yaml
 	@cp manifests/postgresql.crd.yaml pkg/apis/acid.zalan.do/v1/postgresql.crd.yaml
 	@cp manifests/postgresql.crd.yaml charts/postgres-operator/crds/postgresqls.yaml

charts/postgres-operator/crds/operatorconfigurations.yaml

@@ -114,9 +114,15 @@ spec:
                     format: int32
                     minimum: 1
                     type: integer
+                  connection_pooler_pod_security_context:
+                    type: object
+                    x-kubernetes-preserve-unknown-fields: true
                   connection_pooler_schema:
                     default: pooler
                     type: string
+                  connection_pooler_security_context:
+                    type: object
+                    x-kubernetes-preserve-unknown-fields: true
                   connection_pooler_user:
                     default: pooler
                     type: string
@@ -664,8 +670,7 @@ spec:
                 type: object
               maintenance_windows:
                 items:
-                  pattern: ^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\
+                  pattern: '^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\ *$'
-                    *$
                   type: string
                 type: array
               major_version_upgrade:

charts/postgres-operator/values.yaml

@@ -475,6 +475,24 @@ configConnectionPooler:
   connection_pooler_default_memory_request: 100Mi
   connection_pooler_default_cpu_limit: "1"
   connection_pooler_default_memory_limit: 100Mi
+  # Override the pooler pod- and container-level securityContext. Useful for hardened
+  # images whose pgbouncer user/UID differs from the operator default (100/101) and which
+  # own /etc/pgbouncer, or to satisfy restricted Pod Security Standards. When unset the
+  # operator keeps its defaults (pod RunAsUser/RunAsGroup 100/101, container
+  # allowPrivilegeEscalation=false).
+  # connection_pooler_pod_security_context:
+  #   runAsUser: 100
+  #   runAsGroup: 101
+  #   runAsNonRoot: true
+  #   fsGroup: 101
+  #   seccompProfile:
+  #     type: RuntimeDefault
+  # connection_pooler_security_context:
+  #   allowPrivilegeEscalation: false
+  #   readOnlyRootFilesystem: true
+  #   capabilities:
+  #     drop:
+  #       - ALL
 
 configPatroni:
   # enable Patroni DCS failsafe_mode feature

docs/reference/operator_parameters.md

@@ -1107,3 +1107,48 @@ operator being able to provide some reasonable defaults.
   **connection_pooler_default_cpu_limit**
   **connection_pooler_default_memory_limit**
   Default resource configuration for connection pooler deployment.
+
+* **connection_pooler_pod_security_context**
+  Pod-level `securityContext` applied to the connection pooler deployment. The
+  value is a standard Kubernetes
+  [PodSecurityContext](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.31/#podsecuritycontext-v1-core)
+  object. Only fields you set are overridden; unset fields keep the operator
+  defaults (`runAsUser: 100`, `runAsGroup: 101`, and the cluster's `fsGroup`).
+  Settable **only** via the `OperatorConfiguration` CRD (not the ConfigMap).
+  Defaults to not set.
+
+* **connection_pooler_security_context**
+  Container-level `securityContext` applied to the pgbouncer container. The
+  value is a standard Kubernetes
+  [SecurityContext](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.31/#securitycontext-v1-core)
+  object. Only fields you set are overridden; when `allowPrivilegeEscalation`
+  is unset it defaults to `false`. Settable **only** via the
+  `OperatorConfiguration` CRD (not the ConfigMap). Defaults to not set.
+
+These two parameters are useful when running a hardened pooler image whose
+pgbouncer user/UID differs from the operator default (100/101), or to satisfy
+restricted [Pod Security Standards](https://kubernetes.io/docs/concepts/security/pod-security-standards/).
+Because they are object-valued they cannot be expressed in the flat operator
+ConfigMap; configure them through the `OperatorConfiguration` CRD:
+
+```yaml
+apiVersion: "acid.zalan.do/v1"
+kind: OperatorConfiguration
+metadata:
+  name: postgresql-operator-default-configuration
+configuration:
+  connection_pooler:
+    connection_pooler_pod_security_context:
+      runAsUser: 100
+      runAsGroup: 101
+      runAsNonRoot: true
+      fsGroup: 101
+      seccompProfile:
+        type: RuntimeDefault
+    connection_pooler_security_context:
+      allowPrivilegeEscalation: false
+      readOnlyRootFilesystem: true
+      capabilities:
+        drop:
+          - ALL
+```

hack/adjust_operatorconfiguration_crd.sh

@@ -0,0 +1,22 @@
+#!/usr/bin/env bash
+
+# Hack to adjust the generated operatorconfiguration CRD YAML file and add
+# missing field settings which can not be expressed via kubebuilder markers.
+#
+# Injections:
+#
+#  * type: string and pattern for the maintenance_windows items.
+#    The MaintenanceWindow type marshals to/from a string (see marshal.go) but
+#    the field is declared `+kubebuilder:validation:Schemaless` /
+#    `+kubebuilder:validation:Type=array`, so controller-gen emits a bare
+#    `type: array` without the required `items` schema. A structural CRD must
+#    specify `items` for every array, so the generated CRD is rejected with:
+#      spec.validation.openAPIV3Schema.properties[configuration]
+#      .properties[maintenance_windows].items: Required value: must be specified
+
+file="${1:-"manifests/operatorconfiguration.crd.yaml"}"
+
+sed -i '/^[[:space:]]*maintenance_windows:$/{
+    # Capture the indentation
+    s/^\([[:space:]]*\)maintenance_windows:$/\1maintenance_windows:\n\1  items:\n\1    pattern: '\''^\\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\\d):([0-5]?\\d)|(2[0-3]|[01]?\\d):([0-5]?\\d))-((2[0-3]|[01]?\\d):([0-5]?\\d)|(2[0-3]|[01]?\\d):([0-5]?\\d))\\ *$'\''\n\1    type: string/
+}' "$file"

manifests/operatorconfiguration.crd.yaml

@@ -664,8 +664,7 @@ spec:
                 type: object
               maintenance_windows:
                 items:
-                  pattern: ^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\
+                  pattern: '^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\ *$'
-                    *$
                   type: string
                 type: array
               major_version_upgrade:

pkg/apis/acid.zalan.do/v1/operator_configuration_type.go

@@ -345,6 +345,14 @@ type ConnectionPoolerConfiguration struct {
 	DefaultCPULimit string `json:"connection_pooler_default_cpu_limit,omitempty"`
 	// +kubebuilder:validation:Pattern=`^(\d+(e\d+)?|\d+(\.\d+)?(e\d+)?[EPTGMK]i?)$`
 	DefaultMemoryLimit string `json:"connection_pooler_default_memory_limit,omitempty"`
+  // +kubebuilder:validation:XPreserveUnknownFields
+	// +kubebuilder:validation:Type=object
+	// +kubebuilder:validation:Schemaless
+	PodSecurityContext *v1.PodSecurityContext `json:"connection_pooler_pod_security_context,omitempty"`
+  // +kubebuilder:validation:XPreserveUnknownFields
+	// +kubebuilder:validation:Type=object
+	// +kubebuilder:validation:Schemaless
+	SecurityContext    *v1.SecurityContext    `json:"connection_pooler_security_context,omitempty"`
 }
 
 // OperatorLogicalBackupConfiguration defines configuration for logical backup

pkg/apis/acid.zalan.do/v1/operatorconfiguration.crd.yaml

@@ -664,8 +664,7 @@ spec:
                 type: object
               maintenance_windows:
                 items:
-                  pattern: ^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\
+                  pattern: '^\ *((Mon|Tue|Wed|Thu|Fri|Sat|Sun):(2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))-((2[0-3]|[01]?\d):([0-5]?\d)|(2[0-3]|[01]?\d):([0-5]?\d))\ *$'
-                    *$
                   type: string
                 type: array
               major_version_upgrade:

pkg/apis/acid.zalan.do/v1/zz_generated.deepcopy.go

@@ -142,6 +142,16 @@ func (in *ConnectionPoolerConfiguration) DeepCopyInto(out *ConnectionPoolerConfi
 		*out = new(int32)
 		**out = **in
 	}
+	if in.PodSecurityContext != nil {
+		in, out := &in.PodSecurityContext, &out.PodSecurityContext
+		*out = new(corev1.PodSecurityContext)
+		(*in).DeepCopyInto(*out)
+	}
+	if in.SecurityContext != nil {
+		in, out := &in.SecurityContext, &out.SecurityContext
+		*out = new(corev1.SecurityContext)
+		(*in).DeepCopyInto(*out)
+	}
 	return
 }
 

pkg/cluster/connection_pooler.go

@@ -29,6 +29,51 @@ import (
 var poolerRunAsUser = int64(100)
 var poolerRunAsGroup = int64(101)
 
+// generateConnectionPoolerPodSecurityContext returns the pod-level securityContext for the
+// connection pooler. When the operator configuration provides one it is used as the base
+// (deep-copied); otherwise an empty context is used. The historical defaults (RunAsUser/RunAsGroup
+// 100/101) and the spilo FSGroup are applied only for fields the configuration leaves unset, so
+// behavior is unchanged when no override is configured.
+func (c *Cluster) generateConnectionPoolerPodSecurityContext(spec *acidv1.PostgresSpec) *v1.PodSecurityContext {
+	var securityContext v1.PodSecurityContext
+	if c.OpConfig.ConnectionPooler.PodSecurityContext != nil {
+		securityContext = *c.OpConfig.ConnectionPooler.PodSecurityContext.DeepCopy()
+	}
+
+	if securityContext.RunAsUser == nil {
+		securityContext.RunAsUser = &poolerRunAsUser
+	}
+	if securityContext.RunAsGroup == nil {
+		securityContext.RunAsGroup = &poolerRunAsGroup
+	}
+
+	if securityContext.FSGroup == nil {
+		effectiveFSGroup := c.OpConfig.Resources.SpiloFSGroup
+		if spec.SpiloFSGroup != nil {
+			effectiveFSGroup = spec.SpiloFSGroup
+		}
+		if effectiveFSGroup != nil {
+			securityContext.FSGroup = effectiveFSGroup
+		}
+	}
+
+	return &securityContext
+}
+
+// generateConnectionPoolerContainerSecurityContext returns the container-level securityContext for
+// the pooler. A configured context is used as the base (deep-copied); AllowPrivilegeEscalation
+// defaults to false when left unset, preserving prior behavior.
+func (c *Cluster) generateConnectionPoolerContainerSecurityContext() *v1.SecurityContext {
+	var securityContext v1.SecurityContext
+	if c.OpConfig.ConnectionPooler.SecurityContext != nil {
+		securityContext = *c.OpConfig.ConnectionPooler.SecurityContext.DeepCopy()
+	}
+	if securityContext.AllowPrivilegeEscalation == nil {
+		securityContext.AllowPrivilegeEscalation = util.False()
+	}
+	return &securityContext
+}
+
 // ConnectionPoolerObjects K8s objects that are belong to connection pooler
 type ConnectionPoolerObjects struct {
 	AuthSecret  *v1.Secret
@@ -383,9 +428,7 @@ func (c *Cluster) generateConnectionPoolerPodTemplate(role PostgresRole) (
 				},
 			},
 		},
-		SecurityContext: &v1.SecurityContext{
+		SecurityContext: c.generateConnectionPoolerContainerSecurityContext(),
-			AllowPrivilegeEscalation: util.False(),
-		},
 	}
 
 	var poolerVolumes []v1.Volume
@@ -444,19 +487,7 @@ func (c *Cluster) generateConnectionPoolerPodTemplate(role PostgresRole) (
 	poolerContainer.Env = envVars
 	poolerContainer.VolumeMounts = volumeMounts
 	tolerationsSpec := tolerations(&spec.Tolerations, c.OpConfig.PodToleration)
-	securityContext := v1.PodSecurityContext{}
+	securityContext := c.generateConnectionPoolerPodSecurityContext(spec)
-
-	// determine the User, Group and FSGroup for the pooler pod
-	securityContext.RunAsUser = &poolerRunAsUser
-	securityContext.RunAsGroup = &poolerRunAsGroup
-
-	effectiveFSGroup := c.OpConfig.Resources.SpiloFSGroup
-	if spec.SpiloFSGroup != nil {
-		effectiveFSGroup = spec.SpiloFSGroup
-	}
-	if effectiveFSGroup != nil {
-		securityContext.FSGroup = effectiveFSGroup
-	}
 
 	podTemplate := &v1.PodTemplateSpec{
 		ObjectMeta: metav1.ObjectMeta{
@@ -469,7 +500,7 @@ func (c *Cluster) generateConnectionPoolerPodTemplate(role PostgresRole) (
 			Containers:                    []v1.Container{poolerContainer},
 			Tolerations:                   tolerationsSpec,
 			Volumes:                       poolerVolumes,
-			SecurityContext:               &securityContext,
+			SecurityContext:               securityContext,
 			ServiceAccountName:            c.OpConfig.PodServiceAccountName,
 		},
 	}

pkg/cluster/connection_pooler_test.go

@@ -1064,6 +1064,137 @@ func TestPoolerTLS(t *testing.T) {
 	assert.Contains(t, poolerContainer.Env, v1.EnvVar{Name: "CONNECTION_POOLER_CLIENT_CA_FILE", Value: "/tls/ca.crt"})
 }
 
+// poolerSecurityContextOpConfig builds a minimal OpConfig usable for exercising the
+// connection pooler securityContext behavior. Callers tweak the returned config's
+// ConnectionPooler / Resources before passing it to syncedPoolerDeployment.
+func poolerSecurityContextOpConfig() config.Config {
+	return config.Config{
+		PodManagementPolicy: "ordered_ready",
+		ProtectedRoles:      []string{"admin"},
+		Auth: config.Auth{
+			SuperUsername:       superUserName,
+			ReplicationUsername: replicationUserName,
+		},
+		Resources: config.Resources{
+			ClusterLabels:        map[string]string{"application": "spilo"},
+			ClusterNameLabel:     "cluster-name",
+			DefaultCPURequest:    "300m",
+			DefaultCPULimit:      "300m",
+			DefaultMemoryRequest: "300Mi",
+			DefaultMemoryLimit:   "300Mi",
+			PodRoleLabel:         "spilo-role",
+		},
+		ConnectionPooler: config.ConnectionPooler{
+			ConnectionPoolerDefaultCPURequest:    "100m",
+			ConnectionPoolerDefaultCPULimit:      "100m",
+			ConnectionPoolerDefaultMemoryRequest: "100Mi",
+			ConnectionPoolerDefaultMemoryLimit:   "100Mi",
+		},
+		PodServiceAccountName: "postgres-pod",
+	}
+}
+
+// syncedPoolerDeployment provisions a pooler with the given OpConfig and returns the
+// resulting master pooler Deployment.
+func syncedPoolerDeployment(t *testing.T, opConfig config.Config) *appsv1.Deployment {
+	t.Helper()
+	client, _ := newFakeK8sPoolerTestClient()
+	clusterName := "acid-test-cluster"
+	namespace := "default"
+
+	pg := acidv1.Postgresql{
+		ObjectMeta: metav1.ObjectMeta{Name: clusterName, Namespace: namespace},
+		Spec: acidv1.PostgresSpec{
+			TeamID: "myapp", NumberOfInstances: 1,
+			EnableConnectionPooler: util.True(),
+			Resources: &acidv1.Resources{
+				ResourceRequests: acidv1.ResourceDescription{CPU: k8sutil.StringToPointer("1"), Memory: k8sutil.StringToPointer("10")},
+				ResourceLimits:   acidv1.ResourceDescription{CPU: k8sutil.StringToPointer("1"), Memory: k8sutil.StringToPointer("10")},
+			},
+			Volume: acidv1.Volume{Size: "1G"},
+		},
+	}
+
+	cluster := New(Config{OpConfig: opConfig}, client, pg, logger, eventRecorder)
+
+	_, err := cluster.createStatefulSet()
+	assert.NoError(t, err)
+
+	cluster.ConnectionPooler = map[PostgresRole]*ConnectionPoolerObjects{
+		Master: {
+			Name:        cluster.connectionPoolerName(Master),
+			ClusterName: clusterName,
+			Namespace:   namespace,
+			Role:        Master,
+		},
+	}
+
+	_, err = cluster.syncConnectionPoolerWorker(nil, &pg, Master)
+	assert.NoError(t, err)
+
+	deploy, err := client.Deployments(namespace).Get(context.TODO(), cluster.connectionPoolerName(Master), metav1.GetOptions{})
+	assert.NoError(t, err)
+	return deploy
+}
+
+func TestConnectionPoolerDefaultSecurityContext(t *testing.T) {
+	deploy := syncedPoolerDeployment(t, poolerSecurityContextOpConfig())
+
+	podSc := deploy.Spec.Template.Spec.SecurityContext
+	assert.NotNil(t, podSc, "pod securityContext should be set")
+	assert.Equal(t, int64(100), *podSc.RunAsUser, "defaults to RunAsUser 100 for backward compatibility")
+	assert.Equal(t, int64(101), *podSc.RunAsGroup, "defaults to RunAsGroup 101 for backward compatibility")
+
+	containerSc := deploy.Spec.Template.Spec.Containers[constants.ConnectionPoolerContainer].SecurityContext
+	assert.NotNil(t, containerSc, "container securityContext should be set")
+	assert.Equal(t, false, *containerSc.AllowPrivilegeEscalation, "defaults AllowPrivilegeEscalation to false")
+}
+
+func TestConnectionPoolerPodSecurityContextOverride(t *testing.T) {
+	runAsUser := int64(1000001)
+	runAsGroup := int64(1000001)
+	spiloFSGroup := int64(103)
+
+	opConfig := poolerSecurityContextOpConfig()
+	// FSGroup comes from the spilo config and must still apply when the override leaves it nil.
+	opConfig.Resources.SpiloFSGroup = &spiloFSGroup
+	opConfig.ConnectionPooler.PodSecurityContext = &v1.PodSecurityContext{
+		RunAsUser:    &runAsUser,
+		RunAsGroup:   &runAsGroup,
+		RunAsNonRoot: util.True(),
+		SeccompProfile: &v1.SeccompProfile{
+			Type: v1.SeccompProfileTypeRuntimeDefault,
+		},
+	}
+
+	deploy := syncedPoolerDeployment(t, opConfig)
+	podSc := deploy.Spec.Template.Spec.SecurityContext
+
+	assert.Equal(t, runAsUser, *podSc.RunAsUser, "honors configured RunAsUser")
+	assert.Equal(t, runAsGroup, *podSc.RunAsGroup, "honors configured RunAsGroup")
+	assert.Equal(t, true, *podSc.RunAsNonRoot, "honors configured RunAsNonRoot")
+	assert.Equal(t, v1.SeccompProfileTypeRuntimeDefault, podSc.SeccompProfile.Type, "honors configured SeccompProfile")
+	assert.Equal(t, spiloFSGroup, *podSc.FSGroup, "still applies spilo FSGroup when override leaves it nil")
+}
+
+func TestConnectionPoolerContainerSecurityContextOverride(t *testing.T) {
+	opConfig := poolerSecurityContextOpConfig()
+	opConfig.ConnectionPooler.SecurityContext = &v1.SecurityContext{
+		AllowPrivilegeEscalation: util.False(),
+		ReadOnlyRootFilesystem:   util.True(),
+		Capabilities: &v1.Capabilities{
+			Drop: []v1.Capability{"ALL"},
+		},
+	}
+
+	deploy := syncedPoolerDeployment(t, opConfig)
+	containerSc := deploy.Spec.Template.Spec.Containers[constants.ConnectionPoolerContainer].SecurityContext
+
+	assert.Equal(t, true, *containerSc.ReadOnlyRootFilesystem, "honors configured ReadOnlyRootFilesystem")
+	assert.Equal(t, false, *containerSc.AllowPrivilegeEscalation, "keeps AllowPrivilegeEscalation false")
+	assert.Equal(t, []v1.Capability{"ALL"}, containerSc.Capabilities.Drop, "honors dropped capabilities")
+}
+
 func TestConnectionPoolerServiceSpec(t *testing.T) {
 	testName := "Test connection pooler service spec generation"
 	var cluster = New(

pkg/controller/operator_config.go

@@ -285,5 +285,10 @@ func (c *Controller) importConfigurationFromCRD(fromCRD *acidv1.OperatorConfigur
 		fromCRD.ConnectionPooler.MaxDBConnections,
 		k8sutil.Int32ToPointer(constants.ConnectionPoolerMaxDBConnections))
 
+	// Security contexts are nil unless explicitly configured; the cluster package falls back
+	// to the historical pooler defaults when they are not set.
+	result.ConnectionPooler.PodSecurityContext = fromCRD.ConnectionPooler.PodSecurityContext
+	result.ConnectionPooler.SecurityContext = fromCRD.ConnectionPooler.SecurityContext
+
 	return result
 }

pkg/util/config/config.go

@@ -165,6 +165,13 @@ type ConnectionPooler struct {
 	ConnectionPoolerDefaultMemoryRequest string `name:"connection_pooler_default_memory_request"`
 	ConnectionPoolerDefaultCPULimit      string `name:"connection_pooler_default_cpu_limit"`
 	ConnectionPoolerDefaultMemoryLimit   string `name:"connection_pooler_default_memory_limit"`
+	// PodSecurityContext and SecurityContext let operators override the pooler
+	// pod- and container-level security contexts. They are populated only from the
+	// OperatorConfiguration CRD (name:"-" excludes them from the ConfigMap decoder,
+	// mirroring LivenessProbe). When nil, the operator falls back to the historical
+	// defaults (pod RunAsUser/RunAsGroup 100/101, container AllowPrivilegeEscalation=false).
+	PodSecurityContext *v1.PodSecurityContext `name:"-"`
+	SecurityContext    *v1.SecurityContext    `name:"-"`
 }
 
 // Config describes operator config