Standby cluster promotion by changing manifest (#2472)

* Standby cluster promotion by changing manifest * Updated the documentation --------- Co-authored-by: Senthilnathan M <snathanm@vmware.com>
2026-10-08 16:33:26 +02:00 · 2024-01-04 12:33:50 +01:00
parent bbba15f9bf
commit dad5b132ec
5 changed files with 212 additions and 25 deletions
@@ -2,6 +2,7 @@ package cluster

 import (
 	"bytes"
+	"fmt"
 	"io/ioutil"
 	"net/http"
 	"testing"
@@ -480,6 +481,140 @@ func TestCheckAndSetGlobalPostgreSQLConfiguration(t *testing.T) {
 	}
 }

+func TestSyncStandbyClusterConfiguration(t *testing.T) {
+	client, _ := newFakeK8sSyncClient()
+	clusterName := "acid-standby-cluster"
+	applicationLabel := "spilo"
+	namespace := "default"
+
+	ctrl := gomock.NewController(t)
+	defer ctrl.Finish()
+
+	pg := acidv1.Postgresql{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      clusterName,
+			Namespace: namespace,
+		},
+		Spec: acidv1.PostgresSpec{
+			NumberOfInstances: int32(1),
+			Volume: acidv1.Volume{
+				Size: "1Gi",
+			},
+		},
+	}
+
+	var cluster = New(
+		Config{
+			OpConfig: config.Config{
+				PatroniAPICheckInterval: time.Duration(1),
+				PatroniAPICheckTimeout:  time.Duration(5),
+				PodManagementPolicy:     "ordered_ready",
+				Resources: config.Resources{
+					ClusterLabels:         map[string]string{"application": applicationLabel},
+					ClusterNameLabel:      "cluster-name",
+					DefaultCPURequest:     "300m",
+					DefaultCPULimit:       "300m",
+					DefaultMemoryRequest:  "300Mi",
+					DefaultMemoryLimit:    "300Mi",
+					MinInstances:          int32(-1),
+					MaxInstances:          int32(-1),
+					PodRoleLabel:          "spilo-role",
+					ResourceCheckInterval: time.Duration(3),
+					ResourceCheckTimeout:  time.Duration(10),
+				},
+			},
+		}, client, pg, logger, eventRecorder)
+
+	cluster.Name = clusterName
+	cluster.Namespace = namespace
+
+	// mocking a config after getConfig is called
+	mockClient := mocks.NewMockHTTPClient(ctrl)
+	configJson := `{"ttl": 20}`
+	r := ioutil.NopCloser(bytes.NewReader([]byte(configJson)))
+	response := http.Response{
+		StatusCode: 200,
+		Body:       r,
+	}
+	mockClient.EXPECT().Get(gomock.Any()).Return(&response, nil).AnyTimes()
+
+	// mocking a config after setConfig is called
+	standbyJson := `{"standby_cluster":{"create_replica_methods":["bootstrap_standby_with_wale","basebackup_fast_xlog"],"restore_command":"envdir \"/run/etc/wal-e.d/env-standby\" /scripts/restore_command.sh \"%f\" \"%p\""}}`
+	r = ioutil.NopCloser(bytes.NewReader([]byte(standbyJson)))
+	response = http.Response{
+		StatusCode: 200,
+		Body:       r,
+	}
+	mockClient.EXPECT().Do(gomock.Any()).Return(&response, nil).AnyTimes()
+	p := patroni.New(patroniLogger, mockClient)
+	cluster.patroni = p
+
+	mockPod := newMockPod("192.168.100.1")
+	mockPod.Name = fmt.Sprintf("%s-0", clusterName)
+	mockPod.Namespace = namespace
+	podLabels := map[string]string{
+		"cluster-name": clusterName,
+		"application":  applicationLabel,
+		"spilo-role":   "master",
+	}
+	mockPod.Labels = podLabels
+	client.PodsGetter.Pods(namespace).Create(context.TODO(), mockPod, metav1.CreateOptions{})
+
+	// create a statefulset
+	sts, err := cluster.createStatefulSet()
+	assert.NoError(t, err)
+
+	// check that pods do not have a STANDBY_* environment variable
+	assert.NotContains(t, sts.Spec.Template.Spec.Containers[0].Env, v1.EnvVar{Name: "STANDBY_METHOD", Value: "STANDBY_WITH_WALE"})
+
+	// add standby section
+	cluster.Spec.StandbyCluster = &acidv1.StandbyDescription{
+		S3WalPath: "s3://custom/path/to/bucket/",
+	}
+	cluster.syncStatefulSet()
+	updatedSts := cluster.Statefulset
+
+	// check that pods do not have a STANDBY_* environment variable
+	assert.Contains(t, updatedSts.Spec.Template.Spec.Containers[0].Env, v1.EnvVar{Name: "STANDBY_METHOD", Value: "STANDBY_WITH_WALE"})
+
+	// this should update the Patroni config
+	err = cluster.syncStandbyClusterConfiguration()
+	assert.NoError(t, err)
+
+	configJson = `{"standby_cluster":{"create_replica_methods":["bootstrap_standby_with_wale","basebackup_fast_xlog"],"restore_command":"envdir \"/run/etc/wal-e.d/env-standby\" /scripts/restore_command.sh \"%f\" \"%p\""}, "ttl": 20}`
+	r = ioutil.NopCloser(bytes.NewReader([]byte(configJson)))
+	response = http.Response{
+		StatusCode: 200,
+		Body:       r,
+	}
+	mockClient.EXPECT().Get(gomock.Any()).Return(&response, nil).AnyTimes()
+
+	pods, err := cluster.listPods()
+	assert.NoError(t, err)
+
+	_, _, err = cluster.patroni.GetConfig(&pods[0])
+	assert.NoError(t, err)
+	// ToDo extend GetConfig to return standy_cluster setting to compare
+	/*
+		defaultStandbyParameters := map[string]interface{}{
+			"create_replica_methods": []string{"bootstrap_standby_with_wale", "basebackup_fast_xlog"},
+			"restore_command":        "envdir \"/run/etc/wal-e.d/env-standby\" /scripts/restore_command.sh \"%f\" \"%p\"",
+		}
+		assert.True(t, reflect.DeepEqual(defaultStandbyParameters, standbyCluster))
+	*/
+	// remove standby section
+	cluster.Spec.StandbyCluster = &acidv1.StandbyDescription{}
+	cluster.syncStatefulSet()
+	updatedSts2 := cluster.Statefulset
+
+	// check that pods do not have a STANDBY_* environment variable
+	assert.NotContains(t, updatedSts2.Spec.Template.Spec.Containers[0].Env, v1.EnvVar{Name: "STANDBY_METHOD", Value: "STANDBY_WITH_WALE"})
+
+	// this should update the Patroni config again
+	err = cluster.syncStandbyClusterConfiguration()
+	assert.NoError(t, err)
+}
+
 func TestUpdateSecret(t *testing.T) {
 	testName := "test syncing secrets"
 	client, _ := newFakeK8sSyncSecretsClient()