CloudNativePG 스위치오버·스위치백 가이드: 분산 클러스터 운영 런북
August 05, 2026
Swapnil Suryawanshi · 2026년 5월 26일
CloudNativePG 스위치오버와 스위치백은 분산 PostgreSQL 토폴로지에서 프라이머리 역할을 안전하게 전환하고 원래 구성으로 복귀하기 위한 핵심 운영 절차입니다. 이 런북에서는 CloudNativePG(CNPG)가 관리하는 두 EDB Postgres Advanced 18 클러스터를 대상으로 제어된 스위치오버(Switchover)와 스위치백(Switchback)을 단계별로 설명합니다. 본 글은 시리즈의 2부입니다. 아직 분산 토폴로지를 구성하지 않으셨다면 1부: Barman Cloud Plugin을 이용한 CloudNativePG PostgreSQL 레플리카 클러스터 배포부터 시작하세요.
CloudNativePG 스위치오버·스위치백 목표
목표는 분산 토폴로지 안에서 프라이머리 클러스터(cluster-primary)와 레플리카 클러스터(cluster-replica) 사이의 프라이머리 역할을 안전하게 교대(rotate)하는 것입니다. 이 과정은 CNPG 네이티브 승격(promotion)·강등(demotion) 워크플로를 활용해 무손실(zero data loss)을 보장합니다.
CloudNativePG 분산 토폴로지 환경
- 오퍼레이터 버전: 1.27.0
- 데이터베이스: EDB Postgres Advanced 18
- 백업: Barman-Plugin
- 프라이머리 클러스터:
cluster-primary(네임스페이스:primary) - 레플리카 클러스터:
cluster-replica(네임스페이스:replica)
1단계: 최초 스위치오버(프라이머리 → 레플리카)
1. 두 클러스터의 초기 상태 확인
작업을 시작하기 전에 프라이머리와 레플리카 클러스터의 상태(health)와 LSN을 확인합니다.
프라이머리:
user% kubectl cnp status cluster-primary -n primary
Cluster Summary
Name primary/cluster-primary
System ID: 7611448666720448538
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Primary instance: cluster-primary-1
Primary promotion time: 2026-02-27 07:48:28 +0000 UTC (310h54m54s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 185M
Current Write LSN: 0/9000060 (Timeline: 1 - WAL File: 000000010000000000000009)
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-primary
First Point of Recoverability: 2026-02-27 13:20:08 IST
Last Successful Backup: 2026-02-27 13:20:08 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 000000010000000000000008 @ 2026-02-27T07:54:14.417091Z
Last Failed WAL: -
Streaming Replication status
Replication Slots Enabled
Name Sent LSN Write LSN Flush LSN Replay LSN Write Lag Flush Lag Replay Lag State Sync State Sync Priority Replication Slot
---- -------- --------- --------- ---------- --------- --------- ---------- ----- ---------- ------------- ----------------
cluster-primary-2 0/9000060 0/9000060 0/9000060 0/9000060 00:00:00 00:00:00 00:00:00 streaming async 0 active
cluster-primary-3 0/9000060 0/9000060 0/9000060 0/9000060 00:00:00 00:00:00 00:00:00 streaming async 0 active
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-primary-1 0/9000060 Primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-2 0/9000060 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-3 0/9000060 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
레플리카:
user% kubectl cnp status cluster-replica -n replica
Replica Cluster Summary
Name replica/cluster-replica
System ID: 7611448666720448538
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Designated primary: cluster-replica-1
Source cluster: cluster-primary
Primary promotion time: 2026-02-27 07:52:58 +0000 UTC (310h50m34s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 104M
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-replica
First Point of Recoverability: -
Last Successful Backup: -
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 000000010000000000000008 @ 2026-02-27T07:54:22.112507Z
Last Failed WAL: -
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-replica-1 0/9000000 Designated primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-2 0/9000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-3 0/9000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
2. 프라이머리 클러스터 강등(Demote)
프라이머리의 replica 섹션을 변경하고 강등 토큰(demotion token)을 획득합니다.
프라이머리 클러스터 설정 변경:
변경 전:
replica:
primary: cluster-primary
source: cluster-primary
변경 후:
replica:
primary: cluster-replica
source: cluster-replica
토큰 획득 및 상태 확인:
user% kubectl get cluster cluster-primary -n primary \
-o jsonpath='{.status.demotionToken}'
eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjEiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAxMDAwMDAwMDAwMDAwMDAwQSIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTE0NDg2NjY3MjA0NDg1MzgiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC9BMDAwMDI4IiwidGltZU9mTGF0ZXN0Q2hlY2twb2ludCI6IlRodSBNYXIgMTIgMDY6NTA6NTEgMjAyNiIsIm9wZXJhdG9yVmVyc2lvbiI6IjEuMjcuMCJ9%
user% kubectl cnp status cluster-primary -n primary
Replica Cluster Summary
Name primary/cluster-primary
System ID: 7611448666720448538
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Designated primary: cluster-primary-1
Source cluster: cluster-replica
Primary promotion time: 2026-03-12 06:50:52 +0000 UTC (1m25s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 200M
Demotion token
Token eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjEiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAxMDAwMDAwMDAwMDAwMDAwQSIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTE0NDg2NjY3MjA0NDg1MzgiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC9BMDAwMDI4IiwidGltZU9mTGF0ZXN0Q2hlY2twb2ludCI6IlRodSBNYXIgMTIgMDY6NTA6NTEgMjAyNiIsIm9wZXJhdG9yVmVyc2lvbiI6IjEuMjcuMCJ9
Validity valid
Latest checkpoint's TimeLineID 1
Latest checkpoint's REDO WAL file 00000001000000000000000A
Latest checkpoint's REDO location 0/A000028
Database system identifier 7611448666720448538 (ok)
Time of latest checkpoint Thu Mar 12 06:50:51 2026
Version of the operator 1.27.0
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-primary
First Point of Recoverability: 2026-02-27 13:20:08 IST
Last Successful Backup: 2026-02-27 13:20:08 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 00000001000000000000000A @ 2026-03-12T06:52:15.544285Z
Last Failed WAL: -
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-primary-1 0/A0000A0 Designated primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-2 0/A0000A0 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-3 0/A0000A0 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
3. 레플리카 클러스터 승격(Promote)
promotionToken을 레플리카 클러스터 설정에 적용합니다.
레플리카 클러스터 설정 변경:
변경 전:
replica:
primary: cluster-primary
source: cluster-primary
변경 후:
replica:
primary: cluster-replica
promotionToken: eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjEiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAxMDAwMDAwMDAwMDAwMDAwQSIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTE0NDg2NjY3MjA0NDg1MzgiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC9BMDAwMDI4IiwidGltZU9mTGF0ZXN0Q2hlY2twb2ludCI6IlRodSBNYXIgMTIgMDY6NTA6NTEgMjAyNiIsIm9wZXJhdG9yVmVyc2lvbiI6IjEuMjcuMCJ9
source: cluster-primary
4. 첫 스위치오버/승격 후 상태 확인
역할이 올바르게 뒤바뀌었는지 확인합니다.
A] 레플리카 클러스터가 프라이머리로 정상 승격됨:
user% kubectl cnp status cluster-replica -n replica
Cluster Summary
Name replica/cluster-replica
System ID: 7611448666720448538
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Primary instance: cluster-replica-1
Primary promotion time: 2026-02-27 07:52:58 +0000 UTC (311h6m25s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 169M
Current Write LSN: 0/B006E90 (Timeline: 2 - WAL File: 00000002000000000000000B)
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-replica
First Point of Recoverability: -
Last Successful Backup: -
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 00000002000000000000000A @ 2026-03-12T06:57:33.229487Z
Last Failed WAL: -
Streaming Replication status
Replication Slots Enabled
Name Sent LSN Write LSN Flush LSN Replay LSN Write Lag Flush Lag Replay Lag State Sync State Sync Priority Replication Slot
---- -------- --------- --------- ---------- --------- --------- ---------- ----- ---------- ------------- ----------------
cluster-replica-2 0/B006E90 0/B006E90 0/B006E90 0/B006E90 00:00:00 00:00:00 00:00:00 streaming async 0 active
cluster-replica-3 0/B006E90 0/B006E90 0/B006E90 0/B006E90 00:00:00 00:00:00 00:00:00 streaming async 0 active
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-replica-1 0/B006E90 Primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-2 0/B006E90 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-3 0/B006E90 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
B] 프라이머리 클러스터가 레플리카로 정상 강등됨:
user% kubectl cnp status cluster-primary -n primary
Replica Cluster Summary
Name primary/cluster-primary
System ID: 7611448666720448538
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Designated primary: cluster-primary-1
Source cluster: cluster-replica
Primary promotion time: 2026-03-12 06:50:52 +0000 UTC (9m32s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 216M
Demotion token
Token eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjEiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAxMDAwMDAwMDAwMDAwMDAwQSIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTE0NDg2NjY3MjA0NDg1MzgiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC9BMDAwMDI4IiwidGltZU9mTGF0ZXN0Q2hlY2twb2ludCI6IlRodSBNYXIgMTIgMDY6NTA6NTEgMjAyNiIsIm9wZXJhdG9yVmVyc2lvbiI6IjEuMjcuMCJ9
Validity valid
Latest checkpoint's TimeLineID 1
Latest checkpoint's REDO WAL file 00000001000000000000000A
Latest checkpoint's REDO location 0/A000028
Database system identifier 7611448666720448538 (ok)
Time of latest checkpoint Thu Mar 12 06:50:51 2026
Version of the operator 1.27.0
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-primary
First Point of Recoverability: 2026-02-27 13:20:08 IST
Last Successful Backup: 2026-02-27 13:20:08 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 00000002000000000000000A @ 2026-03-12T06:57:39.660756Z
Last Failed WAL: -
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-primary-1 0/B000000 Designated primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-2 0/B000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-3 0/B000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
5. 복제 확인
새 프라이머리에서 테이블을 생성하고, 레플리카에서 확인합니다.
승격된 프라이머리 (cluster-replica):
user% kubectl cnp psql cluster-replica -n replica
psql (18.1.0)
Type "help" for help.
postgres=# \dt
Did not find any tables.
postgres=# create table test (id int);
CREATE TABLE
postgres=# insert into test values (1);
INSERT 0 1
postgres=# select * from test;
id
----
1
(1 row)
강등된 레플리카 (cluster-primary):
user% kubectl cnp psql cluster-primary -n primary
psql (18.1.0)
Type "help" for help.
postgres=# \dt
Did not find any tables.
6. 현재 프라이머리의 신규 백업 수행
user% kubectl cnp backup cluster-replica -n replica --method=plugin --plugin-name=barman-cloud.cloudnative-pg.io
backup/cluster-replica-20260312133305 created
swapnilsuryawanshi@MAC-CR9L20YFN6 plugin % kubectl get backup -n replica
NAME AGE CLUSTER METHOD PHASE ERROR
cluster-replica-20260312133305 58s cluster-replica plugin completed
7. 백업 후 복제 확인
user% kubectl cnp psql cluster-primary -n primary
psql (18.1.0)
Type "help" for help.
postgres=# \dt
List of tables
Schema | Name | Type | Owner
--------+------+-------+----------
public | test | table | postgres
(1 row)
postgres=# select * from test;
id
----
1
(1 row)
2단계: 스위치백(레플리카 → 프라이머리)
8. 현재 프라이머리(cluster-replica)를 레플리카로 강등
설정을 변경하고 새 강등 토큰을 가져옵니다.
레플리카 클러스터 설정 변경:
변경 전:
replica:
primary: cluster-replica
promotionToken: eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjEiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAxMDAwMDAwMDAwMDAwMDAwQSIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTYyNzEyOTAzMDUyODIwNzMiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC9BMDAwMDI4IiwidGltZU9mTGF0ZXN0Q2hlY2twb2ludCI6IlRodSBNYXIgMTIgMDc6NTE6MjMgMjAyNiIsIm9wZXJhdG9yVmVyc2lvbiI6IjEuMjcuMCJ9
source: cluster-primary
변경 후:
replica:
primary: cluster-primary
source: cluster-primary
토큰 획득 및 상태:
user% kubectl get cluster cluster-replica -n replica -o jsonpath='{.status.demotionToken}'
eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjIiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAyMDAwMDAwMDAwMDAwMDAxMCIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTYyNzEyOTAzMDUyODIwNzMiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC8xMDAwMDAyOCIsInRpbWVPZkxhdGVzdENoZWNrcG9pbnQiOiJUaHUgTWFyIDEyIDA4OjEwOjA4IDIwMjYiLCJvcGVyYXRvclZlcnNpb24iOiIxLjI3LjAifQ==%
user% kubectl cnp status cluster-replica -n replica
Replica Cluster Summary
Name replica/cluster-replica
System ID: 7616271290305282073
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Designated primary: cluster-replica-1
Source cluster: cluster-primary
Primary promotion time: 2026-03-12 08:10:09 +0000 UTC (14s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 248M
Demotion token
Token eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjIiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAyMDAwMDAwMDAwMDAwMDAxMCIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTYyNzEyOTAzMDUyODIwNzMiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC8xMDAwMDAyOCIsInRpbWVPZkxhdGVzdENoZWNrcG9pbnQiOiJUaHUgTWFyIDEyIDA4OjEwOjA4IDIwMjYiLCJvcGVyYXRvclZlcnNpb24iOiIxLjI3LjAifQ==
Validity valid
Latest checkpoint's TimeLineID 2
Latest checkpoint's REDO WAL file 000000020000000000000010
Latest checkpoint's REDO location 0/10000028
Database system identifier 7616271290305282073 (ok)
Time of latest checkpoint Thu Mar 12 08:10:08 2026
Version of the operator 1.27.0
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-replica
First Point of Recoverability: 2026-03-12 13:33:10 IST
Last Successful Backup: 2026-03-12 13:33:10 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 00000002.history @ 2026-03-12T08:10:15.149343Z
Last Failed WAL: -
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-replica-1 0/100000A0 Designated primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-2 0/100000A0 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-3 0/100000A0 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
9. 현재 레플리카(cluster-primary)를 프라이머리로 승격
원래 프라이머리에 승격 토큰을 다시 적용합니다.
프라이머리 클러스터 설정 변경:
변경 전:
replica:
primary: cluster-replica
source: cluster-replica
변경 후:
replica:
primary: cluster-primary
promotionToken: eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjIiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAyMDAwMDAwMDAwMDAwMDAxMCIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTYyNzEyOTAzMDUyODIwNzMiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC8xMDAwMDAyOCIsInRpbWVPZkxhdGVzdENoZWNrcG9pbnQiOiJUaHUgTWFyIDEyIDA4OjEwOjA4IDIwMjYiLCJvcGVyYXRvclZlcnNpb24iOiIxLjI3LjAifQ==
source: cluster-replica
10. 두 클러스터의 상태 확인
최종 스위치백 상태를 검증합니다.
A] cluster-primary가 프라이머리로 정상 재승격됨:
user% kubectl cnp status cluster-primary -n primary
Cluster Summary
Name primary/cluster-primary
System ID: 7616271290305282073
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Primary instance: cluster-primary-1
Primary promotion time: 2026-03-12 07:51:23 +0000 UTC (25m20s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 344M
Current Write LSN: 0/11001F18 (Timeline: 3 - WAL File: 000000030000000000000011)
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-primary
First Point of Recoverability: 2026-03-12 13:15:19 IST
Last Successful Backup: 2026-03-12 13:15:19 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 000000030000000000000010 @ 2026-03-12T08:16:05.880203Z
Last Failed WAL: -
Streaming Replication status
Replication Slots Enabled
Name Sent LSN Write LSN Flush LSN Replay LSN Write Lag Flush Lag Replay Lag State Sync State Sync Priority Replication Slot
---- -------- --------- --------- ---------- --------- --------- ---------- ----- ---------- ------------- ----------------
cluster-primary-2 0/11001F18 0/11001F18 0/11001F18 0/11001F18 00:00:00 00:00:00 00:00:00 streaming async 0 active
cluster-primary-3 0/11001F18 0/11001F18 0/11001F18 0/11001F18 00:00:00 00:00:00 00:00:00 streaming async 0 active
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-primary-1 0/11001F18 Primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-2 0/11001F18 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-primary-3 0/11001F18 Standby (async) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
B] cluster-replica가 레플리카로 정상 재강등됨:
user% kubectl cnp status cluster-replica -n replica
Replica Cluster Summary
Name replica/cluster-replica
System ID: 7616271290305282073
PostgreSQL Image: docker.enterprisedb.com/k8s/edb-postgres-advanced:18-standard-ubi9
Designated primary: cluster-replica-1
Source cluster: cluster-primary
Primary promotion time: 2026-03-12 08:10:09 +0000 UTC (8m7s)
Status: Cluster in healthy state
Instances: 3
Ready instances: 3
Size: 264M
Demotion token
Token eyJsYXRlc3RDaGVja3BvaW50VGltZWxpbmVJRCI6IjIiLCJyZWRvV2FsRmlsZSI6IjAwMDAwMDAyMDAwMDAwMDAwMDAwMDAxMCIsImRhdGFiYXNlU3lzdGVtSWRlbnRpZmllciI6Ijc2MTYyNzEyOTAzMDUyODIwNzMiLCJsYXRlc3RDaGVja3BvaW50UkVET0xvY2F0aW9uIjoiMC8xMDAwMDAyOCIsInRpbWVPZkxhdGVzdENoZWNrcG9pbnQiOiJUaHUgTWFyIDEyIDA4OjEwOjA4IDIwMjYiLCJvcGVyYXRvclZlcnNpb24iOiIxLjI3LjAifQ==
Validity valid
Latest checkpoint's TimeLineID 2
Latest checkpoint's REDO WAL file 000000020000000000000010
Latest checkpoint's REDO location 0/10000028
Database system identifier 7616271290305282073 (ok)
Time of latest checkpoint Thu Mar 12 08:10:08 2026
Version of the operator 1.27.0
Continuous Backup status (Barman Cloud Plugin)
ObjectStore / Server name: s3-store/cluster-replica
First Point of Recoverability: 2026-03-12 13:33:10 IST
Last Successful Backup: 2026-03-12 13:33:10 IST
Last Failed Backup: -
Working WAL archiving: OK
WALs waiting to be archived: 0
Last Archived WAL: 000000030000000000000010 @ 2026-03-12T08:16:08.35668Z
Last Failed WAL: -
Instances status
Name Current LSN Replication role Status QoS Manager Version Node
---- ----------- ---------------- ------ --- --------------- ----
cluster-replica-1 0/11000000 Designated primary OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-2 0/11000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
cluster-replica-3 0/11000000 Standby (in Replica Cluster) OK BestEffort 1.27.0 cnp-1.27.0-control-plane
Plugins status
Name Version Status Reported Operator Capabilities
---- ------- ------ ------------------------------
barman-cloud.cloudnative-pg.io 0.11.0 N/A Reconciler Hooks, Lifecycle Service
11. 최종 복제 상태 검증
데이터 일관성을 확인하고, 원래 프라이머리에서 쓰기/아카이브 기능을 테스트합니다.
프라이머리 (cluster-primary):
user% kubectl cnp psql cluster-primary -n primary
psql (18.1.0)
Type "help" for help.
postgres=# \dt
List of tables
Schema | Name | Type | Owner
--------+------+-------+----------
public | test | table | postgres
(1 row)
postgres=# select * from test;
id
----
1
2
(2 rows)
postgres=# insert into test values (3);
INSERT 0 1
postgres=# checkpoint;
CHECKPOINT
postgres=# select * from pg_switch_wal();
pg_switch_wal
---------------
0/11002180
(1 row)
레플리카 (cluster-replica):
user% kubectl cnp psql cluster-replica -n replica
psql (18.1.0)
Type "help" for help.
postgres=# \dt
List of tables
Schema | Name | Type | Owner
--------+------+-------+----------
public | test | table | postgres
(1 row)
postgres=# select * from test;
id
----
1
2
3
(3 rows)
CloudNativePG 스위치오버·스위치백 운영 요약
이 런북은 CNPG 네이티브 강등 토큰(demotion token)·승격 토큰(promotion token) 워크플로를 사용해, 두 CloudNativePG 클러스터 사이에서 완전한 무손실(zero-data-loss) 스위치오버·스위치백 사이클을 시연했습니다.
구조화된 2단계 접근법, 즉 프라이머리 역할을 먼저 cluster-replica로 교대했다가 다시 cluster-primary로 되돌리는 과정을 통해 다음을 확인했습니다.
- 강등 토큰은 승격이 시작되기 전에 기존 프라이머리를 안전하게 펜싱(fencing)합니다.
- 승격 토큰은 새 프라이머리가 이전 프라이머리가 멈춘 지점을 정확히 이어받도록 보장하여, 타임라인 전환 전반에 걸쳐 WAL 연속성을 유지합니다.
- 각 페이즈 동안 기록된 모든 데이터는 스탠바이 측에 정확히 복제되어 검증되었습니다.
이 패턴은 WAL 아카이빙과 백업을 위한 Barman Cloud Plugin과 완전히 호환되므로, 쿠버네티스 상의 프로덕션 분산 토폴로지에 적합합니다.
CloudNativePG 스위치오버·스위치백 FAQ
CloudNativePG 스위치오버와 스위치백의 차이는 무엇인가요?
스위치오버는 프라이머리 역할을 cluster-primary에서 cluster-replica로 전환하는 과정입니다. 스위치백은 전환된 역할을 다시 원래 구성으로 되돌려 cluster-primary를 프라이머리로 복구하는 과정입니다.
CloudNativePG에서 demotion token은 어떤 역할을 하나요?
강등 토큰(demotion token)은 승격을 시작하기 전에 기존 프라이머리를 안전하게 펜싱하는 데 사용됩니다. 이후 이 토큰을 새 프라이머리의 promotionToken으로 적용해 역할 전환을 진행합니다.
promotion token은 왜 필요한가요?
승격 토큰(promotion token)은 새 프라이머리가 이전 프라이머리가 중단된 지점을 정확히 이어받도록 합니다. 이를 통해 타임라인 전환 과정에서 WAL 연속성을 유지합니다.
스위치오버 후 복제 상태는 어떻게 확인하나요?
kubectl cnp status로 두 클러스터의 역할과 상태, LSN을 확인합니다. 새 프라이머리에서 테스트 데이터를 생성한 뒤 레플리카에서 같은 데이터가 조회되는지도 검증합니다.
스위치백 후에는 무엇을 최종 검증해야 하나요?
cluster-primary가 프라이머리로 재승격되고 cluster-replica가 레플리카로 재강등되었는지 확인해야 합니다. 이어서 데이터 일관성, 쓰기 작업, 체크포인트, WAL 전환 및 복제 결과를 검증합니다.
EDB Postgres AI와 CloudNativePG, 더 알아보기
쿠버네티스 기반 PostgreSQL 운영(HA·DR·마이그레이션)이나 PoC가 궁금하시다면 EDB Korea로 문의해 주세요.
메일: salesinquiry@enterprisedb.com
원문: Switchover and Switchback of CloudNativePG Replica Clusters in a Distributed Topology (K8s) – Part 2 (EDB Blog)