Decommissioning Role Instances

You can remove a role instance such as a DataNode from a cluster while the cluster is running by decommissioning the role instance.

Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Limited Cluster Administrator , and Full Administrator)

When you decommission a role instance, Cloudera Manager performs a procedure so that you can safely retire a host without losing data. Role decommissioning applies to the following service roles:
  • HDFS DataNode
  • MapReduce TaskTracker
  • YARN NodeManager
  • HBase RegionServer
  • Kafka Broker

    Decommissioning only applies to the Kafka Broker role if Cruise Control is installed on the cluster. When a Kafka Broker role is decommissioned, Cruise Control moves the data and partitions from the selected broker role to other broker roles in the cluster. For more information about decommissioning Kafka brokers, see Scaling Kafka brokers.

Hosts with DataNodes and DataNode roles themselves can only be decommissioned if the resulting action leaves enough DataNodes commissioned to maintain the configured HDFS replication factor (by default 3). If you attempt to decommission a DataNode or a host with a DataNode in such situations, the decommission process will not complete and must be aborted.

A role will be decommissioned if its host is decommissioned.

To remove a DataNode from the cluster, you decommission the DataNode role as described here and then perform a few additional steps to remove the role. See the topic Delete a DataNode.

To decommission role instances:
  1. If you are decommissioning DataNodes, perform the steps in the topic Tuning HDFS Prior to Decommissioning DataNodes.
  2. Click the service instance that contains the role instance you want to decommission.
  3. Click the Instances tab.
  4. Check the checkboxes next to the role instances to decommission.
  5. Select Actions for Selected > Decommission, and then click Decommission again to start the process.
A Decommission Command pop-up displays that shows each step or decommission command as it is run. In the Details area, click to see the subcommands that are run. Depending on the role, the steps may include adding the host to an "exclusions list" and refreshing the NameNode, JobTracker, or NodeManager; stopping the Balancer (if it is running); and moving data blocks or regions. Roles that do not have specific decommission actions are stopped.

You can abort the decommission process by clicking the Abort button, but you must recommission and restart the role.

The Commission State facet in the Filters list displays Decommissioning while decommissioning is in progress, and Decommissioned when the decommissioning process has finished. When the process is complete, a is added in front of Decommission Command.