YARN service management via ADCM

Elena Kostyuchenko

The ADCM UI provides actions to manage the YARN service and its components. For information on how to run service actions, refer to ADH service actions.

The actions available for the YARN service are listed in the table below.

For more information about YARN components, see YARN architecture.

Action Description

Add/Remove components

Running this action opens the component-host mapping interface where you can add, remove, and distribute YARN components

NodeManager Decommission/Recommission

Allows you to decommission/recommission NodeManager components. The decommission action can be used to temporarily take a NodeManager out of service or to safely delete it. When running this action, specify the hosts with NodeManagers to decommission in the Hosts in decommissioned mode list. For each item in this list, specify the FQDN of the host (Enter field name) and the timeout in seconds to start the decommission process for that host (Enter field value). By default, the decommission starts immediately. After all hosts are specified, review the configuration and run the action. To decommission a NodeManager gracefully, use the refreshNodes command with the -g option

Manage Ranger plugin

Enables or disables Ranger plugin for YARN

Add/modify node labels

Adds or changes node labels. You can use node labels to create groups of hosts, for example, to designate nodes to specific tasks or to manage nodes' resources between queues. When using this action, enter labels to be used in a cluster and their exclusivity (True or False). The dafault exclusivity is True

Add/modify node-to-labels mapping

Maps the node labels added using the Add/modify node labels action to the hosts. When using this action, enter the FQDNs of the hosts with NodeManager components and the labels you want to assign to them

Reload Scheduler config

Reloads access control lists, states, and scheduler specific properties. Use this action to apply changes made to Capacity Scheduler’s configuration file. Fair Scheduler doesn’t require reloading after its configuration has been altered

Remove node label

Removes node labels from the cluster. When using this action, enter the list of labels you want to remove separated by commas

Check

Runs service-specific tests to check the health of the service and its components

Start

Starts the service. When you run this action, the option Apply configs from ADCM is available. If it is set to true, all service configurations defined in ADCM settings will be applied on the service startup. Otherwise, the service starts without applying configurations from ADCM

Stop

Stops the service

Remove

Removes the service from the cluster. This action should be used to remove already installed services. Whereas the delete control can be used to remove a non-mapped service (a service which components have not been distributed among cluster hosts)

Restart

Restarts the service. When you run this action, the option Apply configs from ADCM is available. If it is set to true, all service configurations defined in ADCM settings will be applied during the service restart. Otherwise, the service restarts without applying configurations from ADCM ​.

The service supports the Rolling restart option that allows you to restart its NodeManager components one by one (or in batches) rather than all at once. With high availability enabled, this feature allows for a seamless restart of the YARN service, making it possible to preserve running YARN containers.

The Rolling restart option has the following parameters:

  • batch_size — the number of NodeManager components to be restarted in one iteration.

  • batch_delay — the delay in seconds between restarting batches of components.

  • health_checks — indicates whether to perform health checks on the restarted components.

  • max_failed_batches_number — the maximum number of NodeManager component batches allowed to fail during restart. Upon reaching this value, the Restart action fails.
YARN components actions
Action Description

Check

Verifies whether all the component instances in the cluster work correctly

Restart

Restarts all the component instances in the cluster

Start

Starts all the component instances in the cluster

Stop

Stops all the component instances in the cluster
Found a mistake? Seleсt text and press Ctrl+Enter to report it