To submit a PySpark job using SSHOperator in Airflow, we need three things:

  • an existing SSH connection to the Spark cluster
  • the location of the PySpark script (for example, an S3 location if we use EMR)
  • parameters used by PySpark and the script

The usage of the operator looks like this:

from airflow.contrib.operators.ssh_operator import SSHOperator

script = 's3://some_bucket/'
spark_parameters = '--executor-memory 100G'
# here we can use Airflow template to define the parameters used in the script
parameters = '--db {{ params.database_instance }}, --output_path {{ params.output_path }}' 

submit_pyspark_job = SSHOperator(
    command='set -a; PYSPARK_PYTHON=python3; /usr/bin/spark-submit --deploy-mode cluster %s %s %s' % (spark_parameters, script, parameters),
Older post

How to add a manual step to an Airflow DAG using the JiraOperator

How can you add a human action to an Airflow DAG?

Newer post

How to delay an Airflow DAG until a given hour using the DateTimeSensor

How to use the DateTimeSensor in Airflow