Work with cross-data-center replication

Select a documentation version:

This page describes how to use cross-data-center replication by creating and working with secondary database clusters in Kubernetes.

For a conceptual overview of cross-data-center replication, see About cross-data-center-replication.

Before you begin

  • Confirm that you have reliable and low latency network connectivity between the primary and secondary data centers, which is crucial for cross-data-center replication to function effectively.
  • Install the latest version of AlloyDB Omni operator to deploy AlloyDB Omni on a Kubernetes cluster in the primary data center and a Kubernetes cluster in the secondary data center. Cross-data-center replication is supported in AlloyDB Omni operator version 1.5.0 and later.
  • Create an AlloyDB Omni database cluster on the Kubernetes cluster in the primary data center.
  • Make sure the primary and standby database servers in your primary database cluster have adequate Write-Ahead Logging (WAL) space that can accommodate the WAL files required for replication to the secondary cluster. Any data which isn't replicated yet to the secondary cluster is stored in the primary cluster as WAL files. Depending on the connection speed between the primary and secondary clusters, you might need additional disk space for this purpose.
  • Keep in mind the following information about secondary database high availability standbys (Preview): if the AlloyDB Omni operator version and the control plane agent version are 1.7.0 or later, then the secondary database cluster can support high availability (HA) standbys.

    Unlike HA standbys on the primary database cluster, HA standbys on the secondary database cluster are asynchronous until the secondary database cluster is promoted to a primary database. HA standbys automatically become synchronous or asynchronous if the database cluster role is changed during a switchover.

Create a secondary database cluster

To create an AlloyDB Omni secondary database cluster and enable replication from your primary database cluster, follow these steps:

  1. Verify that external connectivity is enabled on your AlloyDB Omni primary database cluster. If external connectivity is not enabled, add the following to the spec section of the database cluster manifest:

    ...
    spec:
      ...
      allowExternalIncomingTraffic: true
     
  2. To use cross-data-center replication with a primary database cluster that has high availability (HA) enabled, confirm that the replayReplicationSlotsOnStandbys field is enabled on the primary database cluster:

    ...
    spec:
      ...
      availability:
        ...
        replayReplicationSlotsOnStandbys: true
     

    Enabling this field, along with the logReplicationSlots explained in the following step, synchronizes the replication slot used by the secondary database cluster to all HA standbys. This configuration helps the new HA primary retain any Write-Ahead Logging (WAL) files not yet consumed by the secondary database cluster after a failover or switchover, allowing the secondary database cluster to resume replication without disruption.

  3. To enable replication on your primary database cluster, apply a manifest similar to the following to your Kubernetes cluster on the primary data center:

    apiVersion: v1
    kind: Secret
    metadata:
      name: ha-rep-pw-DB_CLUSTER_NAME
      namespace: DB_CLUSTER_NAMESPACE
    type: Opaque
    data:
      rep-user-pw: "ENCODED_PASSWORD"
    ---
    apiVersion: alloydbomni.dbadmin.goog/v1
    kind: Replication
    metadata:
      name: REPLICATION_NAME
      namespace: DB_CLUSTER_NAMESPACE
    spec:
      dbcluster:
        name: DB_CLUSTER_NAME
      upstream:
        password:
          name: ha-rep-pw-DB_CLUSTER_NAME
        logReplicationSlot: true

    Replace the following:

    • DB_CLUSTER_NAME: The name of the database cluster—for example, dbc-1.
    • ENCODED_PASSWORD: The password for the database user to be used for replication from secondary databases, encoded as a base64 string—for example, Q2hhbmdlTWUxMjM= for ChangeMe123. The default value is alloydbreplica.
    • REPLICATION_NAME: The name of the replication—for example, replication-1.
    • LOG_REPLICATION_SLOT: Log replication slot data to WAL files. To enable this option, set its value to true. The default value is false.

    We recommend that you enable the logReplicationSlot option with a primary database cluster that has high availability (HA) enabled to make sure that replication can continue working after a failover or switchover.

    Wait for the replication status to be ready.

  4. To get the upstream connection information that is used to configure replication on the secondary database cluster, run the following command:

    kubectl get replication REPLICATION_NAME
    kubectl get replication REPLICATION_NAME -o json | jq .status.upstream

    Sample output looks similar to the following:

      {
        "host": "35.230.32.36",
        "password": {
          "name": "ha-rep-pw-dbc-1"
        },
        "port": 5432,
        "replicationSlotName": "dbc_1_replication_1",
        "username": "alloydbreplica"
      }
      
  5. Take a note of the output since you need it to enable replication on the secondary database cluster in the following step.
  6. Create an AlloyDB Omni cluster on your Kubernetes cluster in the secondary data center using a configuration that's identical to your primary database cluster.
  7. Verify that external connectivity is enabled on your AlloyDB Omni secondary database cluster
  8. If external connectivity isn't enabled, add the following to the spec section of its manifest:

    ...
    spec:
      ...
      allowExternalIncomingTraffic: true
  9. To enable replication on your secondary database cluster, apply a manifest similar to the following to your Kubernetes cluster on the secondary data center:
    apiVersion: v1
    kind: Secret
    metadata:
      name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME
      namespace: SECONDARY_DB_CLUSTER_NAMESPACE
    type: Opaque
    data:
      rep-user-pw: "ENCODED_PASSWORD"
    ---
    apiVersion: alloydbomni.dbadmin.goog/v1
    kind: Replication
    metadata:
      name: SECONDARY_REPLICATION_NAME
      namespace: SECONDARY_DB_CLUSTER_NAMESPACE
    spec:
      dbcluster:
        name: SECONDARY_DB_CLUSTER_NAME
      downstream:
        host: PRIMARY_HOST
        port: PRIMARY_PORT
        username: alloydbreplica
        password:
          name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME
        replicationSlotName: PRIMARY_REPLICATION_SLOT
        control: setup

    Replace the following:

    • SECONDARY_DB_CLUSTER_NAME: the name of the secondary database cluster—for example, dbc-2.
    • ENCODED_PASSWORD: the password for the database user to be used for replication from the primary database cluster, encoded as a base64 string—for example, Q2hhbmdlTWUxMjM= for ChangeMe123. The default value is alloydbreplica.
    • SECONDARY_REPLICATION_NAME: the name of the replication—for example, `replication-2`.
    • PRIMARY_HOST: the connection endpoint of the primary database cluster from the output in step 4 that the secondary database can access for replication.
    • PRIMARY_PORT: the connection port of the primary database cluster from the output in step 4 that the secondary database can access for replication.
    • PRIMARY_REPLICATION_SLOT: the name of the replication slot on the primary database cluster from the output in step 4 that the secondary database can use for replication.

View replication on the secondary database cluster

To view detailed information about an AlloyDB Omni secondary database cluster and its replication status, run the following commands:

kubectl get dbcluster SECONDARY_DB_CLUSTER_NAME
kubectl get replication SECONDARY_REPLICATION_NAME

When the secondary database cluster is successfully set up and has streaming replication from the primary database cluster, the replication status is both ready and healthy.

Promote a secondary database cluster

When you promote a secondary database cluster, you must first verify that the secondary database cluster applied all transactions received from the primary database cluster.

  1. Verify that the replication status of the secondary database cluster is both ready and healthy:
    kubectl get replication SECONDARY_REPLICATION_NAME
  2. Stop all writes to the primary database cluster. Execute a checkpoint on the primary database cluster, and then run the following query on your primary database cluster to check the replication lag of the secondary database. Confirm that the result shows a minimal lag.
    1. Execute a checkpoint to flush memory to disk.
      psql -h PRIMARY_HOST -U postgres -d postgres -c 'CHECKPOINT;'
    2. Check the replication lag. A lag value of 0 is optimal. If the lag is greater than 0, you can still promote the secondary database cluster. However, there is a risk of losing recent transactions that are already committed on the primary database cluster.
      psql -h PRIMARY_HOST -U postgres -d postgres -c 'SELECT application_name, pg_wal_lsn_diff(pg_current_wal_lsn(), replay_lsn) AS replay_lag FROM pg_stat_replication;'

  3. Promote a secondary database cluster to a primary database cluster by updating the control field of your secondary database cluster's replication manifest to promote. Then, apply the updated replication manifest on your Kubernetes cluster on the secondary data center.

    apiVersion: v1
    kind: Secret
    metadata:
      name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME
      namespace: SECONDARY_DB_CLUSTER_NAMESPACE
    type: Opaque
    data:
      rep-user-pw: "ENCODED_PASSWORD"
    ---
    apiVersion: alloydbomni.dbadmin.goog/v1
    kind: Replication
    metadata:
      name: SECONDARY_REPLICATION_NAME
      namespace: SECONDARY_DB_CLUSTER_NAMESPACE
    spec:
      dbcluster:
        name: SECONDARY_DB_CLUSTER_NAME
      downstream:
        host: PRIMARY_HOST
        port: PRIMARY_PORT
        username: alloydbreplica
        password:
          name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME
        replicationSlotName: PRIMARY_REPLICATION_SLOT
        control: promote

Perform a switchover

Before you perform a switchover, verify that the primary and secondary database clusters in both data centers are online and in a healthy state.

To maintain data consistency between your primary and secondary database clusters and perform a switchover, follow these steps:

  1. Verify that the replication status of the secondary database cluster is both ready and healthy:
    kubectl get replication SECONDARY_REPLICATION_NAME
  2. Stop all writes to the primary database cluster, and then do the following on your primary database cluster:
    1. Execute a checkpoint to flush memory to disk:
      psql -h PRIMARY_HOST -U postgres -d postgres -c 'CHECKPOINT;'
    2. Check the replication lag of the secondary database cluster and confirm that the result shows a lag value of 0:
      psql -h PRIMARY_HOST -U postgres -d postgres -c 'SELECT application_name, pg_wal_lsn_diff(pg_current_wal_lsn(), replay_lsn) AS replay_lag FROM pg_stat_replication;'
  3. To promote the secondary database cluster to a primary database cluster and convert the former primary database cluster into a secondary database cluster, follow these steps:

    1. To convert your AlloyDB Omni secondary database cluster into a primary database cluster, update its replication manifest on your Kubernetes cluster in the secondary data center as follows:

      apiVersion: v1
      kind: Secret
      metadata:
        name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME
        namespace: SECONDARY_DB_CLUSTER_NAMESPACE
      type: Opaque
      data:
        rep-user-pw: "ENCODED_PASSWORD"
      ---
      apiVersion: alloydbomni.dbadmin.goog/v1
      kind: Replication
      metadata:
        name: SECONDARY_REPLICATION_NAME
        namespace: SECONDARY_DB_CLUSTER_NAMESPACE
      spec:
        dbcluster:
          name: SECONDARY_DB_CLUSTER_NAME
        upstream:
          password:
            name: ha-rep-pw-SECONDARY_DB_CLUSTER_NAME

      Replace SECONDARY_DB_CLUSTER_NAME with the name of your secondary database cluster—for example, dbc-2.

      Wait for the replication status to be ready.

    2. To get the upstream connection information for replication, run the following command:

      kubectl get replication SECONDARY_REPLICATION_NAME
      kubectl get replication SECONDARY_REPLICATION_NAME -o json | jq .status.upstream

      Sample output looks similar to the following:

        {
          "host": "34.23.207.137",
          "password": {
            "name": "ha-rep-pw-dbc-2"
          },
          "port": 5432,
          "replicationSlotName": "dbc_2_replication_2",
          "username": "alloydbreplica"
        }
      
    3. To convert your AlloyDB Omni primary database cluster into a secondary database cluster, update its replication manifest on your Kubernetes cluster in the primary data center:

      apiVersion: v1
      kind: Secret
      metadata:
        name: ha-rep-pw-DB_CLUSTER_NAME
        namespace: DB_CLUSTER_NAMESPACE
      type: Opaque
      data:
        rep-user-pw: "ENCODED_PASSWORD"
      ---
      apiVersion: alloydbomni.dbadmin.goog/v1
      kind: Replication
      metadata:
        name: REPLICATION_NAME
        namespace: DB_CLUSTER_NAMESPACE
      spec:
        dbcluster:
          name: DB_CLUSTER_NAME
        downstream:
          host: SECONDARY_HOST
          port: SECONDARY_PORT
          username: alloydbreplica
          password:
            name: ha-rep-pw-DB_CLUSTER_NAME
          replicationSlotName: SECONDARY_REPLICATION_SLOT
          control: rewind

      Wait for the replication status to become ready and healthy.

    4. Verify the replication status:

      kubectl get replication REPLICATION_NAME

Perform unplanned disaster recovery

If your primary data center or primary database cluster becomes unexpectedly unavailable, you can perform an unplanned disaster recovery failover. To do this, promote the secondary database cluster, and then reconfigure the former primary database cluster after it comes back online:

  1. Promote your AlloyDB Omni secondary database cluster to a primary database cluster. Update the secondary database cluster's replication manifest on your Kubernetes cluster in the secondary data center as described in step 3a of Perform a switchover.
  2. After the original primary data center becomes available, convert the former primary database cluster into a secondary database cluster that replicates from the new primary. Follow steps 3b, 3c, and 3d in Perform a switchover, with a change in step 3c. In step 3c, set control: setup instead of control: rewind to reinitialize the cluster from the new primary. Don't use control: rewind because the former primary database cluster might contain changes that weren't synchronized before the failover.