Shrink a Dedicated Kafka Cluster in Confluent Cloud

This topic shows you how to shrink an existing Dedicated Apache Kafka® cluster. Dedicated clusters are provisioned and billed in terms of Confluent Unit for Kafka (CKU). Reducing the number of CKUs for your cluster enables you to reduce a cluster to fit its utilization and help reduce costs.

When you reduce the size of a Dedicated Kafka cluster, Confluent rebalances the cluster to help ensure a balanced load across the remaining brokers in the cluster and then removes the brokers.

You can reduce the capacity of a cluster one CKU at a time.

Prerequisites:

You can shrink the number of CKUs in the Confluent Cloud Console. When you shrink using the console, you get a preview of the adjusted load of the cluster.

  1. Select your Confluent Cloud cluster.

    Confluent Cluster panel
  2. Click the Cluster Settings, the Capacity tab and on the Adjustable limits pane, select Adjust capacity.

    Confluent Cluster capacity
  3. View the current usage on your cluster and determine if your cluster load is low enough to support a shrink operation. In general, shrinking a highly loaded cluster will result in application performance issues. For more information, see Cluster Load Metric for Dedicated Clusters in Confluent Cloud, and Kafka cluster quotas.

  4. Use the slider to reduce the number of CKUs for the cluster. You can reduce the capacity 1 CKU at a time. You will see a preview of the reduced capacity before it is applied. If you are satisfied with the changes, click Apply changes. For details on how to expand a cluster, see Expand a Dedicated Kafka Cluster in Confluent Cloud.

    Expand Confluent Cluster capacity
  5. Confirm your changes and click Continue.

    Confirm cluster expansion

Important

If the shrink operation fails, the cluster will expand to its original capacity, and you will be notified when the expansion operation is complete.

See Dedicated cluster provisioning and resizing time for guidance on how long shrinking may take and variables that affect that timing.