Automated Multi-Region PostgreSQL Disaster Recovery with CloudNativePG, AWS S3, and ArgoCD

Building high-availability PostgreSQL architectures across geographic regions requires reliable Write-Ahead Log (WAL) shipping, point-in-time recovery (PITR) capabilities, and declarative lifecycle management. CloudNativePG provides native Kubernetes abstractions for PostgreSQL, leveraging Barman Cloud for object storage synchronization. When combined with AWS S3 cross-region replication and ArgoCD, teams can implement an enterprise-grade GitOps DR pattern achieving Near-Zero Recovery Point Objective (RPO) and low Recovery Time Objective (RTO).

1. Multi-Region Architectural Topology

The architecture consists of a Primary Cluster located in us-east-1 (Region A) and a Designated Replica Cluster running in us-west-2 (Region B). Continuous WAL archives and base backups stream to an encrypted AWS S3 bucket in us-east-1, which cross-region replicates to an S3 bucket in us-west-2.

ArgoCD manages both Kubernetes clusters declaratively via GitOps, ensuring environment consistency and controlling promote/failover workflows through configuration changes.

2. AWS IAM and S3 Cross-Region Setup

Provision AWS IAM roles using IAM Roles for Service Accounts (IRSA) to grant CloudNativePG pods permission to upload and retrieve WALs without static credentials.

# IAM Trust Policy for CloudNativePG Operator ServiceAccount
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "Federated": "arn:aws:iam::123456789012:oidc-provider/oidc.eks.us-east-1.amazonaws.com/id/EXAMPLE123456"
      },
      "Action": "sts:AssumeRoleWithWebIdentity",
      "Condition": {
        "StringEquals": {
          "oidc.eks.us-east-1.amazonaws.com/id/EXAMPLE123456:sub": "system:serviceaccount:postgres-system:cnpg-main"
        }
      }
    }
  ]
}

3. Primary Cluster Manifest (Region A: us-east-1)

The primary cluster writes WAL logs continuously to the primary S3 bucket using Barman Cloud Object Store integration natively supported by CloudNativePG.

apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: postgres-primary
  namespace: database
spec:
  instances: 3
  primaryUpdateStrategy: Unsupervised
  storage:
    size: 100Gi
    storageClass: gp3-encrypted
  walStorage:
    size: 50Gi
    storageClass: gp3-encrypted
  backup:
    barmanObjectStore:
      destinationPath: s3://company-postgres-backups-us-east-1/primary-db
      endpointURL: https://s3.us-east-1.amazonaws.com
      s3Credentials:
        inheritFromIAMRole: true
      wal:
        compression: gzip
        maxParallel: 4
      data:
        compression: gzip
        jobs: 2
  resources:
    requests:
      cpu: "2"
      memory: 4Gi
    limits:
      cpu: "4"
      memory: 8Gi

4. Designated Replica Cluster Manifest (Region B: us-west-2)

The replica cluster in us-west-2 runs in continuous standby mode, fetching WAL files from the local replicated S3 bucket. It maintains near-real-time synchronization without opening direct cross-region network links between database pods.

apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: postgres-replica
  namespace: database
spec:
  instances: 3
  replica:
    enabled: true
    source: primary-s3-source
  externalClusters:
    - name: primary-s3-source
      barmanObjectStore:
        destinationPath: s3://company-postgres-backups-us-west-2/primary-db
        endpointURL: https://s3.us-west-2.amazonaws.com
        s3Credentials:
          inheritFromIAMRole: true
        wal:
          maxParallel: 4
  storage:
    size: 100Gi
    storageClass: gp3-encrypted
  resources:
    requests:
      cpu: "2"
      memory: 4Gi
    limits:
      cpu: "4"
      memory: 8Gi

5. ArgoCD Application Management & Automated Failover

ArgoCD targets both primary and backup Kubernetes clusters using ApplicationSet resources. To execute an orchestrated disaster recovery failover:

• Step 1: Disable production traffic to us-east-1 via ExternalDNS / Route53 DNS switch.
• Step 2: In Git, update the replica cluster spec to disable standby mode by removing spec.replica.enabled: true or setting it to false.
• Step 3: Push the change to the repository. ArgoCD syncs the target state to us-west-2, promoting the standby PostgreSQL cluster to a fully functional primary.

Operational Note: Ensure S3 Cross-Region Replication latency stays below 10 seconds to satisfy enterprise RPO requirements during emergency promotion scenarios.