Data Processing Stack Overflow Data Using Apache Spark on AWS EMR
An overview on how to process data in spark using DataBricks, add the script as a step in AWS EMR and output the data to Amazon Redshift
This article is part of the series and continuation of the previous post.
In the previous post, we saw how we can stream the data using Kinesis Firehose either using stackapi or using Kinesis Data Generator.
In this post, let’s see how we can decide on the key processing steps that need to be performed before we send the data to any analytics tool.
The key steps, I followed in this phase are as below :
Step 1 : Prototype in DataBricks
I always use DataBricks for prototyping as they are free and can be directly connected to AWS S3. Since EMR is charged by the hour, if you want to use spark for your projects DataBricks is the best way to go.
Key Steps in DataBricks
- Create a cluster
This is fairly straightforward, just create a cluster and attach to it any notebook.
2. Create a notebook and attach it to the cluster
3. Mount the S3 bucket to databricks
The good thing about databricks is that you can directly connect it to S3, but you need to mount the S3 bucket.
Below code mounts the s3 bucket so that you can access the contents of the S3 bucket
This command will show you the contents of the S3 bucket
%fs ls /mnt/stack-overflow-bucket/
After this we do some Exploratory data analysis in spark
Key Processing techniques done here are :
Check and Delete duplicate entries ,if any.
Convert the date into Salesforce compatible format (YYYY/MM/DD).
Step 2 : Create a Test EMR cluster
pyspark script that you developed in the DataBricks environment gives you the skeleton of the code .
However, you are not done yet!
You need to ensure that this code won’t fail when you run the script in aws. So for this purpose, I create a dummy EMR cluster and test my code in the iPython notebook.This step is covered in detail in this post.
Step 3:Test the Prototype code in pyspark
In the Jupyter notebook that you created, make sure to test out the code.
Key points to note here:
Kinesis Streams output the data into multiple files within a day. The format of the file depends on the prefix of the S3 file location set while creating the delivery stream.
Spark job runs on the next day and is expected to pick all the files of Yesterday’s date from the s3 bucket.
Function get_latest_filename creates a text string which matches the file name spark is supposed to pick up.
So in the Jupyter notebook, i am testing if spark is able to process the file without any errors
Step 4 & 5 :Convert the ipython notebook into python notebook and upload it in s3.
You can use the below commands to convert the ipython notebook to pyscript
pip install ipython
pip install nbconvert
ipython nbconvert — to script stack_processing.ipynb
After this is done ,upload this script to the S3 folder
Step 5 : Add the script as a step in EMR job using boto3
Before we do this, we need to configure the redshift cluster details and test if the functionality is working.
This will be covered in the Redshift post.
