cancel
Showing results for 
Search instead for 
Did you mean: 
Data Engineering
Join discussions on data engineering best practices, architectures, and optimization strategies within the Databricks Community. Exchange insights and solutions with fellow data engineers.
cancel
Showing results for 
Search instead for 
Did you mean: 

SparkContext lost when running %sh script.py

madrhr
New Contributor III

I need to execute a .py file in Databricks from a notebook (with arguments which for simplicity i exclude here). For this i am using:

%sh script.py

script.py:

from pyspark import SparkContext

def main():
    sc = SparkContext.getOrCreate()
    print(sc)

if __name__ == "__main__":
    main()

However, i need SparkContext in .py file and its suggested to use SparkContext.getOrCreate() but i get the exception that i need to set a master url. 

pyspark.errors.exceptions.base.PySparkRuntimeError: [MASTER_URL_NOT_SET] A master URL must be set in your configuration.

But even if i set the master url, i get another exception. Now whats really weird is that if i run the same .py script directly in Databricks using the little play button it works. It also works if i open a web terminal of the cluster und execute my .py script in this bash shell. So using both approaches it works and i get the SparkContext. However this is obvious not very useful. In the %sh shell and in the web shell, user is root, same working directory and the python env is also not the problem.

The cluster i am using is a single node NC24ads_A100, so only a driver node and no additional worker nodes. I running DBR 14.2 ML and Spark 3.5.0.

Would be very happy to know whats so special about %sh or where my problem is or whats a workaround to execute .py files from a databricks notebooks with arguments and while staying/getting SparkContext.

1 ACCEPTED SOLUTION

Accepted Solutions

madrhr
New Contributor III

I got it eventually working with a combination of:

  • from databricks.sdk.runtime import *
  • spark.sparkContext.addPyFile("/path/to/your/file")
  • sys.path.append("path/to/your")
 
 
 

View solution in original post

3 REPLIES 3

Yeshwanth
Honored Contributor
Honored Contributor

@madrhr

I think this occurs because one session is initiated within the Python script (.py file), while in the Databricks notebook, we have a pre-configured Spark session. It is important to note that we cannot use more than one Spark session per notebook, and each session should be unique.

madrhr
New Contributor III

Thanks for you answer. Thats also how i understand it. But is there a way to inject or connect to the pre-configured Spark session from within the Python script (.py file)?

madrhr
New Contributor III

I got it eventually working with a combination of:

  • from databricks.sdk.runtime import *
  • spark.sparkContext.addPyFile("/path/to/your/file")
  • sys.path.append("path/to/your")
 
 
 
Join 100K+ Data Experts: Register Now & Grow with Us!

Excited to expand your horizons with us? Click here to Register and begin your journey to success!

Already a member? Login and join your local regional user group! If there isn’t one near you, fill out this form and we’ll create one for you to join!