Drill-on-YARN Overview

Running Drill as a YARN application (Drill-on-YARN) enables Drill to work alongside other applications, such as Hadoop and Spark, in a YARN-managed cluster. If you are currently running Drill under Warden, you can upgrade Drill and continue to run Drill under Warden, or you can migrate Drill to run under YARN. See Migrate Drill to Run Under YARN for instructions.

YARN assigns resources, such as memory and CPU, to applications in the cluster and eliminates the manual steps associated with installation and resource allocation for stand-alone applications in a multi-tenant environment. YARN automatically deploys (localizes) the Drill software onto each Drill node and manages the Drill cluster. Drill becomes a long-running application with YARN. You can monitor the Drill-on-YARN cluster using the Application Master web UI.

Resource Usage

By design, Drill aggressively uses all of the resources available to run queries at optimal speed. When Drill runs under YARN, you must inform YARN of the resources that Drill needs. The resource settings are descriptive, not proscriptive. Drill does not limit itself to the YARN settings, instead the YARN settings inform YARN of the resources that Drill will consume so that YARN does not over-allocate those resources to other tasks.

The MapR version of YARN manages CPU, memory, and disks. YARN calls settings for memory and CPU “vcores." You configure Drill’s memory and then inform YARN of the Drill configuration. Drill uses all available disk I/O and CPU.


Several software components work together to run Drill as a YARN application. Drill, YARN, and the Drill-on-YARN application collectively provide the components required to run Drill under YARN.

The following table lists the software components with their descriptions:

Software Component Description
YARN Resource Manager The Resource Manager manages the set of applications running on the cluster. Each cluster must have one Resource Manager.
Node Manager The Node Manager manages the application tasks running on a particular node. Each node in a cluster must have one Node Manager.
Drill Drillbit A Drillbit is the Drill daemon software that YARN runs on each node in the cluster. SeeDrill Query Execution. The Drillbit process on the node acts as the application task.
Client A client, such as JDBC, ODBC, or SQLLine sends queries to a Drillbit. The client uses ZooKeeper to discover a Drillbit that the client treats as the Foreman. SeeDrill Clients.
Drill distribution archive The Drill distribution archive is a Drill distribution .tar.gz file included with the Drill installation. Drill-on-YARN uploads this archive to the distributed file system (DFS). YARN downloads (localizes) the file to each worker node.
ZooKeeper ZooKeeper ZooKeeper is the service that tracks the available set of Drillbit processes. The Foreman for a query uses ZooKeeper to identify the set of available drill nodes that can run the query. SeeDrill Query Execution.
Drill-on-YARN Application Application Master (AM) The Application Manager requests containers from the Resource Manager and launches Drillbits using those containers. The AM monitors Drillbits, detects failures, and restarts failed Drillbits. The AM also provides a web UI to manage the Drill cluster.
Drill-on-YARN client The Drill-on-YARN client is a command-line program that starts, stops, and monitors the Drill cluster. The client provides the information that YARN needs to start the Application Master. The client can run on any machine that has both the Drill and YARN client software. The client does not have to be part of the YARN cluster.
drill-on-yarn.conf The drill-on-yarn.conf configuration file provides the information that Drill-on-YARN needs to manage the Drill cluster. This file is separate from the configuration files for Drill itself.

Component Workflow

Running Drill as a YARN application is mostly an automated process carried out by the Drill and YARN components. After an administrator installs, configures, and launches Drill from the Drill-on-YARN client, YARN deploys (localizes) Drill on to designated nodes and starts the Drill process on each node.

The following diagram shows the workflow between the components in a cluster with the steps that Drill and YARN complete to deploy and run Drill as a YARN application: