add code to sync config maps

charts/postgres-operator/templates/clusterrole-postgres-pod.yaml

@@ -9,7 +9,7 @@ metadata:
     app.kubernetes.io/managed-by: {{ .Release.Service }}
     app.kubernetes.io/instance: {{ .Release.Name }}
 rules:
-# Patroni needs to watch and manage endpoints
+# Patroni needs to watch and manage config maps or endpoints
 {{- if toString .Values.configGeneral.kubernetes_use_configmaps | eq "true" }}
 - apiGroups:
   - ""
@@ -24,12 +24,6 @@ rules:
   - patch
   - update
   - watch
-- apiGroups:
-  - ""
-  resources:
-  - endpoints
-  verbs:
-  - get
 {{- else }}
 - apiGroups:
   - ""

charts/postgres-operator/templates/clusterrole.yaml

@@ -89,12 +89,6 @@ rules:
   - patch
   - update
   - watch
-- apiGroups:
-  - ""
-  resources:
-  - endpoints
-  verbs:
-  - get
 {{- else }}
 # to read configuration from ConfigMaps
 - apiGroups:

docs/quickstart.md

@@ -37,7 +37,7 @@ The Postgres Operator can be deployed in the following ways:
 * Kustomization
 * Helm chart
 
-### Manual deployment setup
+### Manual deployment setup on Kubernetes
 
 The Postgres Operator can be installed simply by applying yaml manifests. Note,
 we provide the `/manifests` directory as an example only; you should consider
@@ -71,6 +71,18 @@ manifest.
 ./run_operator_locally.sh
 ```
 
+### Manual deployment setup on OpenShift
+
+To install the Postgres Operator in OpenShift you have to change the config
+parameter `kubernetes_use_configmaps` to `"true"`. Otherwise, the operator
+and Patroni will store leader and config keys in `Endpoints` that are not
+supported in OpenShift. This requires also a slightly different set of rules
+for the `postgres-operator` and `postgres-pod` cluster roles.
+
+```bash
+oc create -f manifests/operator-service-account-rbac-openshift.yaml
+```
+
 ### Helm chart
 
 Alternatively, the operator can be installed by using the provided [Helm](https://helm.sh/)

e2e/tests/test_e2e.py

@@ -1759,6 +1759,8 @@ class EndToEndTestCase(unittest.TestCase):
             self.eventuallyEqual(lambda: len(k8s.api.custom_objects_api.list_namespaced_custom_object(
                 "acid.zalan.do", "v1", "default", "postgresqls", label_selector="cluster-name=acid-minimal-cluster")["items"]), 0, "Manifest not deleted")
 
+            self.eventuallyEqual(lambda: k8s.get_operator_state(), {"0": "idle"}, "Operator does not get in sync")
+
             # check if everything has been deleted
             self.eventuallyEqual(lambda: k8s.count_pods_with_label(cluster_label), 0, "Pods not deleted")
             self.eventuallyEqual(lambda: k8s.count_services_with_label(cluster_label), 0, "Service not deleted")

manifests/configmap.yaml

@@ -63,13 +63,13 @@ data:
   # etcd_host: ""
   external_traffic_policy: "Cluster"
   # gcp_credentials: ""
-  # kubernetes_use_configmaps: "false"
   # ignored_annotations: ""
   # infrastructure_roles_secret_name: "postgresql-infrastructure-roles"
   # infrastructure_roles_secrets: "secretname:monitoring-roles,userkey:user,passwordkey:password,rolekey:inrole"
   # inherited_annotations: owned-by
   # inherited_labels: application,environment
   # kube_iam_role: ""
+  # kubernetes_use_configmaps: "false"
   # log_s3_bucket: ""
   logical_backup_docker_image: "registry.opensource.zalan.do/acid/logical-backup:v1.7.1"
   # logical_backup_google_application_credentials: ""

manifests/operator-service-account-rbac-openshift.yaml

@@ -0,0 +1,283 @@
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: postgres-operator
+  namespace: default
+
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: postgres-operator
+rules:
+# all verbs allowed for custom operator resources
+- apiGroups:
+  - acid.zalan.do
+  resources:
+  - postgresqls
+  - postgresqls/status
+  - operatorconfigurations
+  verbs:
+  - create
+  - delete
+  - deletecollection
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# operator only reads PostgresTeams
+- apiGroups:
+  - acid.zalan.do
+  resources:
+  - postgresteams
+  verbs:
+  - get
+  - list
+  - watch
+# all verbs allowed for event streams (Zalando-internal feature)
+# - apiGroups:
+#   - zalando.org
+#   resources:
+#   - fabriceventstreams
+#   verbs:
+#   - create
+#   - delete
+#   - deletecollection
+#   - get
+#   - list
+#   - patch
+#   - update
+#   - watch
+# to create or get/update CRDs when starting up
+- apiGroups:
+  - apiextensions.k8s.io
+  resources:
+  - customresourcedefinitions
+  verbs:
+  - create
+  - get
+  - patch
+  - update
+# to read configuration and manage ConfigMaps used by Patroni
+- apiGroups:
+  - ""
+  resources:
+  - configmaps
+  verbs:
+  - create
+  - delete
+  - deletecollection
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# to send events to the CRs
+- apiGroups:
+  - ""
+  resources:
+  - events
+  verbs:
+  - create
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# to CRUD secrets for database access
+- apiGroups:
+  - ""
+  resources:
+  - secrets
+  verbs:
+  - create
+  - delete
+  - get
+  - update
+# to check nodes for node readiness label
+- apiGroups:
+  - ""
+  resources:
+  - nodes
+  verbs:
+  - get
+  - list
+  - watch
+# to read or delete existing PVCs. Creation via StatefulSet
+- apiGroups:
+  - ""
+  resources:
+  - persistentvolumeclaims
+  verbs:
+  - delete
+  - get
+  - list
+  - patch
+  - update
+ # to read existing PVs. Creation should be done via dynamic provisioning
+- apiGroups:
+  - ""
+  resources:
+  - persistentvolumes
+  verbs:
+  - get
+  - list
+  - update  # only for resizing AWS volumes
+# to watch Spilo pods and do rolling updates. Creation via StatefulSet
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - delete
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# to resize the filesystem in Spilo pods when increasing volume size
+- apiGroups:
+  - ""
+  resources:
+  - pods/exec
+  verbs:
+  - create
+# to CRUD services to point to Postgres cluster instances
+- apiGroups:
+  - ""
+  resources:
+  - services
+  verbs:
+  - create
+  - delete
+  - get
+  - patch
+  - update
+# to CRUD the StatefulSet which controls the Postgres cluster instances
+- apiGroups:
+  - apps
+  resources:
+  - statefulsets
+  - deployments
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+# to CRUD cron jobs for logical backups
+- apiGroups:
+  - batch
+  resources:
+  - cronjobs
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+# to get namespaces operator resources can run in
+- apiGroups:
+  - ""
+  resources:
+  - namespaces
+  verbs:
+  - get
+# to define PDBs. Update happens via delete/create
+- apiGroups:
+  - policy
+  resources:
+  - poddisruptionbudgets
+  verbs:
+  - create
+  - delete
+  - get
+# to create ServiceAccounts in each namespace the operator watches
+- apiGroups:
+  - ""
+  resources:
+  - serviceaccounts
+  verbs:
+  - get
+  - create
+# to create role bindings to the postgres-pod service account
+- apiGroups:
+  - rbac.authorization.k8s.io
+  resources:
+  - rolebindings
+  verbs:
+  - get
+  - create
+# to grant privilege to run privileged pods (not needed by default)
+#- apiGroups:
+#  - extensions
+#  resources:
+#  - podsecuritypolicies
+#  resourceNames:
+#  - privileged
+#  verbs:
+#  - use
+
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: postgres-operator
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: postgres-operator
+subjects:
+- kind: ServiceAccount
+  name: postgres-operator
+  namespace: default
+
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: postgres-pod
+rules:
+# Patroni needs to watch and manage config maps
+- apiGroups:
+  - ""
+  resources:
+  - configmaps
+  verbs:
+  - create
+  - delete
+  - deletecollection
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# Patroni needs to watch pods
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get
+  - list
+  - patch
+  - update
+  - watch
+# to let Patroni create a headless service
+- apiGroups:
+  - ""
+  resources:
+  - services
+  verbs:
+  - create
+# to grant privilege to run privileged pods (not needed by default)
+#- apiGroups:
+#  - extensions
+#  resources:
+#  - podsecuritypolicies
+#  resourceNames:
+#  - privileged
+#  verbs:
+#  - use

pkg/cluster/cluster.go

@@ -59,6 +59,7 @@ type Config struct {
 type kubeResources struct {
 	Services            map[PostgresRole]*v1.Service
 	Endpoints           map[PostgresRole]*v1.Endpoints
+	ConfigMaps          map[PostgresRole]*v1.ConfigMap
 	Secrets             map[types.UID]*v1.Secret
 	Statefulset         *appsv1.StatefulSet
 	PodDisruptionBudget *policybeta1.PodDisruptionBudget
@@ -1484,22 +1485,29 @@ func (c *Cluster) GetCurrentProcess() Process {
 
 // GetStatus provides status of the cluster
 func (c *Cluster) GetStatus() *ClusterStatus {
-	return &ClusterStatus{
-		Cluster: c.Spec.ClusterName,
-		Team:    c.Spec.TeamID,
-		Status:  c.Status,
-		Spec:    c.Spec,
-
+	status := &ClusterStatus{
+		Cluster:             c.Spec.ClusterName,
+		Team:                c.Spec.TeamID,
+		Status:              c.Status,
+		Spec:                c.Spec,
 		MasterService:       c.GetServiceMaster(),
 		ReplicaService:      c.GetServiceReplica(),
-		MasterEndpoint:      c.GetEndpointMaster(),
-		ReplicaEndpoint:     c.GetEndpointReplica(),
 		StatefulSet:         c.GetStatefulSet(),
 		PodDisruptionBudget: c.GetPodDisruptionBudget(),
 		CurrentProcess:      c.GetCurrentProcess(),
 
 		Error: fmt.Errorf("error: %s", c.Error),
 	}
+
+	if c.patroniKubernetesUseConfigMaps() {
+		status.MasterEndpoint = c.GetEndpointMaster()
+		status.ReplicaEndpoint = c.GetEndpointReplica()
+	} else {
+		status.MasterConfigMap = c.GetConfigMapMaster()
+		status.ReplicaConfigMap = c.GetConfigMapReplica()
+	}
+
+	return status
 }
 
 // Switchover does a switchover (via Patroni) to a candidate pod
@@ -1579,10 +1587,11 @@ func (c *Cluster) deletePatroniClusterObjects() error {
 	}
 
 	if c.patroniKubernetesUseConfigMaps() {
-		actionsList = append(actionsList, c.deletePatroniClusterServices, c.deletePatroniClusterConfigMaps)
+		actionsList = append(actionsList, c.deletePatroniClusterConfigMaps)
 	} else {
 		actionsList = append(actionsList, c.deletePatroniClusterEndpoints)
 	}
+	actionsList = append(actionsList, c.deletePatroniClusterServices)
 
 	c.logger.Debugf("removing leftover Patroni objects (endpoints / services and configmaps)")
 	for _, deleter := range actionsList {

pkg/cluster/k8sres.go

@@ -76,13 +76,12 @@ func (c *Cluster) statefulSetName() string {
 	return c.Name
 }
 
-func (c *Cluster) endpointName(role PostgresRole) string {
-	name := c.Name
-	if role == Replica {
-		name = name + "-repl"
-	}
+func (c *Cluster) configMapName(role PostgresRole) string {
+	return c.serviceName(role)
+}
 
-	return name
+func (c *Cluster) endpointName(role PostgresRole) string {
+	return c.serviceName(role)
 }
 
 func (c *Cluster) serviceName(role PostgresRole) string {
@@ -1821,6 +1820,16 @@ func (c *Cluster) generateEndpoint(role PostgresRole, subsets []v1.EndpointSubse
 	return endpoints
 }
 
+func (c *Cluster) generateConfigMap(role PostgresRole) *v1.ConfigMap {
+	return &v1.ConfigMap{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      c.configMapName(role),
+			Namespace: c.Namespace,
+			Labels:    c.roleLabelsSet(true, role),
+		},
+	}
+}
+
 func (c *Cluster) generateCloneEnvironment(description *acidv1.CloneDescription) []v1.EnvVar {
 	result := make([]v1.EnvVar, 0)
 

pkg/cluster/resources.go

@@ -35,8 +35,14 @@ func (c *Cluster) listResources() error {
 		c.logger.Infof("found secret: %q (uid: %q) namesapce: %s", util.NameFromMeta(obj.ObjectMeta), obj.UID, obj.ObjectMeta.Namespace)
 	}
 
-	for role, endpoint := range c.Endpoints {
-		c.logger.Infof("found %s endpoint: %q (uid: %q)", role, util.NameFromMeta(endpoint.ObjectMeta), endpoint.UID)
+	if c.patroniKubernetesUseConfigMaps() {
+		for role, configMap := range c.ConfigMaps {
+			c.logger.Infof("found %s config map: %q (uid: %q)", role, util.NameFromMeta(configMap.ObjectMeta), configMap.UID)
+		}
+	} else {
+		for role, endpoint := range c.Endpoints {
+			c.logger.Infof("found %s endpoint: %q (uid: %q)", role, util.NameFromMeta(endpoint.ObjectMeta), endpoint.UID)
+		}
 	}
 
 	for role, service := range c.Services {
@@ -402,6 +408,20 @@ func (c *Cluster) generateEndpointSubsets(role PostgresRole) []v1.EndpointSubset
 	return result
 }
 
+func (c *Cluster) createConfigMap(role PostgresRole) (*v1.ConfigMap, error) {
+	c.setProcessName("creating config map")
+	configMapSpec := c.generateConfigMap(role)
+
+	configMap, err := c.KubeClient.ConfigMaps(configMapSpec.Namespace).Create(context.TODO(), configMapSpec, metav1.CreateOptions{})
+	if err != nil {
+		return nil, fmt.Errorf("could not create %s config map: %v", role, err)
+	}
+
+	c.ConfigMaps[role] = configMap
+
+	return configMap, nil
+}
+
 func (c *Cluster) createPodDisruptionBudget() (*policybeta1.PodDisruptionBudget, error) {
 	podDisruptionBudgetSpec := c.generatePodDisruptionBudget()
 	podDisruptionBudget, err := c.KubeClient.
@@ -589,11 +609,21 @@ func (c *Cluster) GetEndpointMaster() *v1.Endpoints {
 	return c.Endpoints[Master]
 }
 
-// GetEndpointReplica returns cluster's kubernetes master Endpoint
+// GetEndpointReplica returns cluster's kubernetes replica Endpoint
 func (c *Cluster) GetEndpointReplica() *v1.Endpoints {
 	return c.Endpoints[Replica]
 }
 
+// GetConfigMapMaster returns cluster's kubernetes master ConfigMap
+func (c *Cluster) GetConfigMapMaster() *v1.ConfigMap {
+	return c.ConfigMaps[Master]
+}
+
+// GetConfigMapReplica returns cluster's kubernetes replica ConfigMap
+func (c *Cluster) GetConfigMapReplica() *v1.ConfigMap {
+	return c.ConfigMaps[Replica]
+}
+
 // GetStatefulSet returns cluster's kubernetes StatefulSet
 func (c *Cluster) GetStatefulSet() *appsv1.StatefulSet {
 	return c.Statefulset

pkg/cluster/sync.go

@@ -144,7 +144,11 @@ func (c *Cluster) syncServices() error {
 	for _, role := range []PostgresRole{Master, Replica} {
 		c.logger.Debugf("syncing %s service", role)
 
-		if !c.patroniKubernetesUseConfigMaps() {
+		if c.patroniKubernetesUseConfigMaps() {
+			if err := c.syncConfigMap(role); err != nil {
+				return fmt.Errorf("could not sync %s config map: %v", role, err)
+			}
+		} else {
 			if err := c.syncEndpoint(role); err != nil {
 				return fmt.Errorf("could not sync %s endpoint: %v", role, err)
 			}
@@ -234,6 +238,40 @@ func (c *Cluster) syncEndpoint(role PostgresRole) error {
 	return nil
 }
 
+func (c *Cluster) syncConfigMap(role PostgresRole) error {
+	var (
+		cm  *v1.ConfigMap
+		err error
+	)
+	c.setProcessName("syncing %s config map", role)
+
+	if cm, err = c.KubeClient.ConfigMaps(c.Namespace).Get(context.TODO(), c.configMapName(role), metav1.GetOptions{}); err == nil {
+		// TODO: No syncing of config map here, is this covered completely by updateService?
+		c.ConfigMaps[role] = cm
+		return nil
+	}
+	if !k8sutil.ResourceNotFound(err) {
+		return fmt.Errorf("could not get %s config map: %v", role, err)
+	}
+	// no existing config map, create new one
+	c.ConfigMaps[role] = nil
+	c.logger.Infof("could not find the cluster's %s config map", role)
+
+	if cm, err = c.createConfigMap(role); err == nil {
+		c.logger.Infof("created missing %s config map %q", role, util.NameFromMeta(cm.ObjectMeta))
+	} else {
+		if !k8sutil.ResourceAlreadyExists(err) {
+			return fmt.Errorf("could not create missing %s config map: %v", role, err)
+		}
+		c.logger.Infof("%s config map %q already exists", role, util.NameFromMeta(cm.ObjectMeta))
+		if cm, err = c.KubeClient.ConfigMaps(c.Namespace).Get(context.TODO(), c.configMapName(role), metav1.GetOptions{}); err != nil {
+			return fmt.Errorf("could not fetch existing %s config map: %v", role, err)
+		}
+	}
+	c.ConfigMaps[role] = cm
+	return nil
+}
+
 func (c *Cluster) syncPodDisruptionBudget(isUpdate bool) error {
 	var (
 		pdb *policybeta1.PodDisruptionBudget

pkg/cluster/types.go

@@ -63,6 +63,8 @@ type ClusterStatus struct {
 	ReplicaService      *v1.Service
 	MasterEndpoint      *v1.Endpoints
 	ReplicaEndpoint     *v1.Endpoints
+	MasterConfigMap     *v1.ConfigMap
+	ReplicaConfigMap    *v1.ConfigMap
 	StatefulSet         *appsv1.StatefulSet
 	PodDisruptionBudget *policybeta1.PodDisruptionBudget
 

pkg/controller/postgresql.go

@@ -544,7 +544,8 @@ func (c *Controller) postgresqlCheck(obj interface{}) *acidv1.Postgresql {
   Ensures the pod service account and role bindings exists in a namespace
   before a PG cluster is created there so that a user does not have to deploy
   these credentials manually.  StatefulSets require the service account to
-  create pods; Patroni requires relevant RBAC bindings to access endpoints.
+  create pods; Patroni requires relevant RBAC bindings to access endpoints
+  or config maps.
 
   The operator does not sync accounts/role bindings after creation.
 */