Promote the replica site to a new primary¶

Let’s say the primary site with cluster1 is down. The client applications have automatically switched to the replica site. Now you need to reconfigure your setup to make cluster2 on the replica site a new primary and have it handle the load.

Check the replica before you promote it¶

Before you promote cluster2, confirm it applied everything cluster1 sent it, and that the cluster itself is healthy.

Check the replication status on cluster2:
```
mysql> SHOW REPLICA STATUS\G
```
Confirm Executed_Gtid_Set matches Retrieved_Gtid_Set. This shows cluster2 already applied every event it retrieved from cluster1, with nothing left queued.
Check the Galera status on every node of cluster2, not just one:
```
mysql> SHOW STATUS LIKE 'wsrep_cluster_status';
mysql> SHOW STATUS LIKE 'wsrep_local_state_comment';
mysql> SHOW STATUS LIKE 'wsrep_ready';
mysql> SHOW VARIABLES LIKE 'read_only';
```
Look for the following on every node:
- wsrep_cluster_status = Primary
- wsrep_local_state_comment = Synced
- wsrep_ready = ON
- read_only = ON
read_only = ON is expected here, since cluster2 is still running as the replica.

Promote the replica to primary¶

Modify the replication channel for cluster2 within the deploy/cr.yaml file:

Set the isSource value to true to make the replica site the source of the data.
Remove the sourcesList configuration.

Run the following command to apply a patch configuration to cluster2.

kubectl patch pxc cluster2 -n <namespace> --type=merge --patch '{
  "spec": {
    "pxc": {
      "replicationChannels": [
        {
          "name": "pxc1_to_pxc2",
          "isSource": true
        }
      ]
    }
  }}'

Now cluster2 acts as the primary site.

Check that the promotion is successful¶

Confirm that cluster2 is genuinely serving as the primary before you do anything irreversible to cluster1.

Check the Galera status on every node of cluster2, not just one:

mysql> SHOW STATUS LIKE 'wsrep_cluster_status';
mysql> SHOW STATUS LIKE 'wsrep_local_state_comment';
mysql> SHOW STATUS LIKE 'wsrep_ready';
mysql> SHOW VARIABLES LIKE 'read_only';

Look for the following on every node:

wsrep_cluster_status = Primary
wsrep_local_state_comment = Synced
wsrep_ready = ON
read_only = OFF

Then confirm that a write succeeds through the client-facing endpoint your application uses, not a direct connection to a single Pod:

mysql> INSERT INTO <your_table> VALUES (...);

Delete the previous primary¶

Important

Do this step only after you confirmed that the new primary is successfully promoted and correctly serves the write requests.

While the old primary site is unavailable, cluster1 no longer has up-to-date data. Once you’ve confirmed cluster2 is serving writes correctly, you can delete it. Refer to the Delete the database cluster tutorial for the steps how to do it.

Last update: September 30, 2026
Created: May 19, 2025