Title
Create new category
Edit page index title
Edit category
Edit link
Remove Kudu Masters (Ambari Multi-Master)
This page describes how to remove a Kudu master from an Ambari-managed multi-master deployment.
Ambari Service Component Management
Stop and Delete the Kudu Master Service
Before making configuration changes, ensure that the master you want to remove is properly stopped in Ambari:
Navigate to the Hosts page in the Ambari Web UI.
Click the host containing the Kudu Master component (for example,
kudu3.acceldata.dvl).In the Components section, locate the Kudu Master component.
Stop the component if it is still running.
Delete the component using the Delete option from the component drop-down menu.
Master Removal Command Execution
Run the kudu master remove command to remove the master from the cluster configuration.
Example:
This command removes kudu3.acceldata.dvl from the cluster configuration. The remaining masters (kudu1.acceldata.dvl and kudu2.acceldata.dvl) now form the new master quorum.
Sample output:
WAL and Data Directory Cleanup
Remove the Write-Ahead Log (WAL) and data directories from the removed master to prevent it from rejoining the cluster.
This step is especially important if a corrupted master is being removed. Allowing a corrupted master to rejoin the cluster can cause data loss.
Service Restart Sequence
Ambari automatically updates the master_addresses and tserver_master_addrs configuration parameters for the remaining masters and tablet servers.
Restart the remaining masters one by one:
In Ambari, go to Kudu > Summary.
Restart the
kudu1.acceldata.dvlmaster first.Wait for it to show as healthy.
Restart the
kudu2.acceldata.dvlmaster.
Restart all tablet servers:
Restart all tablet servers at once, or use Ambari’s Rolling Restart option.
In Ambari, go to Kudu > Summary.
Select all tablet servers.
Click Restart to apply the updated master configuration.
Verification Steps
Verify the cluster state by accessing the master web interface.
Open the web UI for each remaining master (
http://<hostname>:8051/masters).Verify that only the remaining masters are listed.
Confirm that one master is shown as LEADER and the other as FOLLOWER.
Run a Cluster Health Check
Run the following command to ensure that all tablet servers are properly connected to the updated master configuration: