Python wrapper for YARN Applications

yarn, dask, hadoop
pip install knit==0.2.4



Build Status Coverage Status

Note: This library has been superceded by Skein (, and is no longer maintained.

For deploying Dask on YARN, please see dask-yarn (, which has been rewritten to use Skein instead of Knit.

For user issues, please refer to either of those repositories

The knit library provides a Python interface to Scala for interacting with the YARN resource manager.

View the documentation for knit.


knit allows you to use python in conjunction with YARN, the most common resource manager for Hadoop systems. It provides to following high-level entry-points:

  • CondaCreator, a way to create zipped conda environments, so that they can be uploaded to HDFS and extracted for use in YARN containers
  • YARNAPI, an interface to the YARN resource manager to get application/container statuses, logs, and to kill running jobs
  • Knit, a YARN application runner, which generates an instance of a scala-based YARN client, and launches an application on YARN, which in turn runs commands in YARN containers
  • DaskYARNCluster, launches a Dask distributed cluster on YARN, one worker process per container.

The intent is to use knit from a cluster edge-node, i.e., with YARN configuration and the CLI available locally.


Install from conda-forge

> conda install -c conda-forge knit

or with pip

> pip install knit

If installing from source, you must first build the java library (requires java and maven)

> python install mvn

To run an arbitrary command on the yarn cluster

import knit
k = knit.Knit()
k.start('env')  # wait some time

To start a dask cluster on YARN

import dask_yarn
cluster = dask_yarn.DaskYARNCluster()
cluster.start(nworkers=4, memory=1024, cpus=2)