Working with PySpark jobs

Apache Spark is a distributed processing framework for unstructured and semi-structured data and a part of the Hadoop project ecosystem.

In this section, we provide a simple example that demonstrates how to use PySpark, the Spark interface for Python, in Yandex Data Processing. In the example, we use PySpark to count the number of times each word is seen in a short text.

Getting startedGetting started

1. Create a service account with the dataproc.agent and dataproc.provisioner roles.

2. In Object Storage, create buckets and configure access to them:

   1. Create a bucket for the input data and grant the cluster service account READ permissions for this bucket.
   2. Create a bucket for the processing output and grant the cluster service account READ and WRITE permissions for this bucket.

3. Create a Yandex Data Processing cluster with the following settings:

   • Services:
     • HDFS
     • SPARK
     • YARN
   • Service account: Select the service account you previously created.
   • Bucket name: Select a bucket to hold the processing results.

Create a PySpark jobCreate a PySpark job

1. Upload a file for processing:

   1. Copy and save the following to a file named text.txt:

      text.txt

      she sells sea shells on the sea shore
      the shells that she sells are sea shells I am sure
      so if she sells sea shells on the sea shore
      I am sure that the shells are sea shore shells
      

   2. Upload the text.txt file to the source data bucket.

2. Upload a file with the Python code of the analysis program:

   1. Copy and save to the word_count.py file:

      word_count.py

      import sys
      from pyspark import SparkConf, SparkContext
      
      
      def main():
      
          if len(sys.argv) != 3:
              print('Usage job.py <input_directory> <output_directory>')
              sys.exit(1)
      
          in_dir = sys.argv[1]
          out_dir = sys.argv[2]
      
          conf = SparkConf().setAppName("Word count - PySpark")
          sc = SparkContext(conf=conf)
      
          text_file = sc.textFile(in_dir)
          counts = text_file.flatMap(lambda line: line.split(" ")) \
              .map(lambda word: (word, 1)) \
              .reduceByKey(lambda a, b: a + b)
      
          if out_dir.startswith('s3a://'):
              counts.saveAsTextFile(out_dir) 
          else:
              default_fs = sc._jsc.hadoopConfiguration().get('fs.defaultFS')
              counts.saveAsTextFile(default_fs + out_dir)
      
      
      if __name__ == "__main__":
          main()
      

   2. Upload the word_count.py file to the source data bucket.

3. Create a PySpark job with the following parameters:

   • Main python file: s3a://<input_data_bucket_name>/word_count.py

   • Arguments:

     • s3a://<input_data_bucket_name>/text.txt
     • s3a://<output_bucket_name>/<output_directory>

4. Wait for the job status to change to Done.

5. Download from the bucket and review the files with the results from the bucket:

   part-00000

   ('sea', 6)
   ('are', 2)
   ('am', 2)
   ('sure', 2)
   

   part-00001

   ('she', 3)
   ('sells', 3)
   ('shells', 6)
   ('on', 2)
   ('the', 4)
   ('shore', 3)
   ('that', 2)
   ('I', 2)
   ('so', 1)
   ('if', 1)
   

Delete the resources you createdDelete the resources you created

Some resources are not free of charge. To avoid paying for them, delete the resources you no longer need:

1. Delete the cluster.
2. Delete buckets.
3. Delete the service account.