Parallelizing Command-Line Tools With the Pipe Transformer

To help you use single-node tools on massive data sets, Databricks Runtime HLS includes a utility called the Pipe Transformer to process Spark DataFrames with command-line programs.


The Pipe Transformer is currently in Beta. In particular, the name of the transformer entry point (currently db_genomics in Python and DBGenomics in Java/Scala) will change.

Python usage


The Pipe Transformer requires Python 3.

Consider a minimal case with a DataFrame containing a single column of strings. You can use the Pipe Transformer to reverse each of the strings in the input DataFrame using the rev Linux command:

import db_genomics as dbg
import pyspark.sql.expressions

# Create a text-only DataFrame
df = spark.createDataFrame([['foo'], ['bar'], ['baz']], ['value'])
display(dbg.transform('pipe', df, cmd='["rev"]', inputformatter='text', outputformatter='text'))

The options in this example demonstrate how to control the basic behavior of the transformer:

  • cmd is a JSON-encoded array that contains the command to invoke the program
  • inputformatter defines how each input row should be passed to the program
  • outputformatter defines how the program output should be converted into a new DataFrame

The input DataFrame can come from any Spark data source — Delta Lake, Parquet, VCF, BGEN, and so on.

Integrating with bioinformatics tools

To integrate with tools for genomic data, you can configure the Pipe Transformer to write each partition of the input DataFrame as VCF by choosing vcf as the input and output formatter.

import db_genomics as dbg

df ="com.databricks.vcf")\

dbg.transform('pipe', df, cmd='["grep", "-v", "#INFO"]',
              inputformatter='vcf', in_vcfheader='infer', outputformatter='vcf')

When you use the VCF input formatter, you must specify a method to determine the VCF header. The simplest option is infer, which instructs the Pipe Transformer to derive a VCF header from the DataFrame schema.

Scala usage

You can also invoke the Pipe Transformer from Scala. You specify options as a Map[String, String].

import com.databricks.hls.DBGenomics

DBGenomics.transform("pipe", df, Map(
  "cmd" -> "[\"grep\", \"-v\", \"#INFO\"]"
  "inputFormatter" -> "vcf",
  "outputFormatter" -> "vcf",
  "in_vcfHeader" -> "infer"))


Option keys and values are always strings. From Python, you provide options through the arg_map argument or as keyword args. From Scala, you provide options as a Map[String, String]. Option names are case insensitive; for example inputFormatter, inputformatter, and INPUTFORMATTER are all equivalent.

Option Description
cmd The command, specified as an array of strings, to invoke the piped program. The program’s stdin receives the formatted contents of the input DataFrame, and the output DataFrame is constructed from its stdout. The stderr stream will appear in the executor logs.
inputFormatter Converts the input DataFrame to a format that the piped program understands. Built-in input formatters are text, csv, and vcf.
outputFormatter Converts the output of the piped program back into a DataFrame. Built-in output formatters are text, csv, and vcf.
env_* Options beginning with env_ are interpreted as environment variables. For example, providing the option env_ANIMAL=MONKEY results in an environment variable with key ANIMAL and value MONKEY being provided to the piped program.

Some of the input and output formatters take additional options.

VCF input formatter:

Option Description

How to determine a VCF header from the input DataFrame. Possible values:

  • infer: Derive a VCF header from the DataFrame schema.
  • The complete contents of a VCF header starting with ##
  • A Hadoop filesystem path (for example, dbfs://...) to a VCF file. The header from this file is used as the VCF header for each partition.

The CSV input and output formatters accept most of the same options as the CSV data source Supported options. You must prefix options to the input formatter with in_, and options to the output formatter with out_. For example, in_quote sets the quote character when writing the input DataFrame to the piped program.

The following options are not supported:

  • path options are ignored
  • The parserLib option is ignored. univocity is always used as the CSV parsing library.

Example usage

The Pipe Transformer can be used with any command line tool. Try any of the following examples.

Example notebook