Uploading Application Code
This is a design document, explaining how to upload application code to Spark nodes, via Databricks Connect.
Application code, such as model definitions and helper functions, should be defined within a folder which is isolated from the runtime program. I’ve found it helpful to organize all the application code within a folder named application. Within the application folder, you can separate your code into folders which are more specific. For example, I usually create a folder for models, another one for helper functions, and another for pipelines.
You’ll also need a file to install the Anaconda environment for that specific job, and another file for the Python package requirements.
├── myjob | ├── application | ├── models | ├── helpers | ├── pipelines ├── run.py ├── install.bat ├── requirements.txt
The file run.py contains the runtime program. This program has several responsibilities:
- Instantiate a connection to the Spark cluster
- Ship the application code to the Spark cluster
- Plan the Spark job
- Distribute the Spark job to the cluster nodes
Instantiate Cluster Connection
Before you execute the job, you’ll need to create a Spark session. The Spark session is the entry point for Pyspark. Keep in mind the Spark session will be connected to the cluster via Databricks Connect. To successfully instantiate the Spark session, you must have already configured Databricks connect on your machine. Let’s create a Spark session.
from pyspark.sql import SparkSession def main(): spark = SparkSession.builder.appName('temp').getOrCreate() if __name__ == '__main__': main()
Ship Application Code
So far our program is quite simple. The next thing we need to do is ship our application code to the Spark cluster. We will use SparkContext to accomplish this goal. The SparkContext API includes a function addPyFile, which can be used to add a
.zip dependency to all tasks executed within the context.
When you add a file to the SparkContext, it is shipped to the Databricks cluster and each of the worker nodes. We could add each individual Python file to the Spark context, or we could create a
.zip archive of the application code and ship that. The
.zip archive will be compressed, so in theory the upload time will be faster. Make sure to add
application.zip to the
.gitignore so it is not inadvertently added to source control.
import shutil from pyspark.sql import SparkSession def main(): spark = SparkSession.builder.appName('temp').getOrCreate() # upload the application code shutil.make_archive('application', 'zip', '.') spark.sparkContext.addPyFile('application.zip')
Now we can write our code which executes within the SparkSession! Since we have shipped the application code to the cluster, the cluster nodes each have a copy of the functions and modules in the application folder.