csv2parquet: Create Parquet files from CSV

This simple tool creates Parquet files from CSV input, using a minimal installation of Apache Drill. As a data format, Parquet offers strong advantages over comma-separated values for big data and cloud computing needs; csv2parquet is designed to let you experience those benefits more easily.

Much credit for this goes to Tugdual "Tug" Grall; csv2parquet essentially automates the process he documents in Convert a CSV File to Apache Parquet With Drill.

csv2parquet is now in public beta. Feedback, comments, bug reports, and feature requests are all appreciated. See "About and Contact" below to reach the author.

Usage

csv2parquet CSV_INPUT PARQUET_OUTPUT [--column-map ...] [--types ...]

csv_input is a CSV file, whose first line defines the column names. parquet_output is the Parquet output (i.e., directory in which one or more Parquet files are written.) Note that csv2parquet is currently specifically designed to work with CSV files whose first line defines header/column names.

Customizing Column Names

By default, Parquet column names have the same name as the CSV header. You can specify a different name for each output column with the --column-map option. When used, it must be followed by an even number of strings, i.e. a sequence of pairs. In each pair, the first string is the CSV file column name, and the second is the Parquet column name to use instead:

csv2parquet data.csv data.parquet --column-map "First Column" "Primary Column" "Another Column" "Special Name"

In this example, two of the CSV columns are named "First Column" and "Another Column". The created Parquet file will store data from these columns under "Primary Column" and "Special Name", respectively.

(A perfectly good CSV column name may not be valid as a Parquet column name - for example, a header name with a period, like "Min. Investment". In this situation, you must use --column-map to provide a column name that Parquet can accept, or edit the source CSV file.)

Column Types

By default, csv2parquet assumes all columns are of type string, but you can declare specific columns to be any Drill data type. You do this using the --types option, whose syntax is similar to --column-map. On the command line, you write --types, followed by an even number of strings that encode a sequence of pairs. In each pair, the first string matches the name of the CSV column. (Not the Parquet column name, if that is different.) The second string is one of the Drill data types, such as "INT", "FLOAT", "DATE", and so on. For example:

csv2parquet data.csv data.parquet --types "First Column" "INT" "Another Column" "FLOAT"

Note you can pass both --types and --column-map to csv2parquet at once:

# On one long line:
csv2parquet data.csv data.parquet --column-map "First Column" "Primary Column" "Another Column" "Special Name" --types "First Column" "INT" "Another Column" "FLOAT"

# Split across lines, for readability:
csv2parquet data.csv data.parquet \
    --column-map "First Column" "Primary Column" "Another Column" "Special Name" \
    --types "First Column" "INT" "Another Column" "FLOAT"

Troubleshooting

If you encounter a bug, run again with the --debug option. and note the directory name which is printed out at startup. Many files, logs, and other info useful for troubleshooting are stored in a temporary folder. --debug prevents this from being deleted after the program completes. See in particular script, script_stderr and script_stdout from that folder. To report bugs, see "About and Contact" below.

Installation

Your system must have:

• Python 3 (version 3.5 or later).
• A quick-and-easy installation of Apache Drill, version 1.4 or 1.5 - see below.

There are no other dependencies. You can simply copy the csv2parquet script wherever you'd like, and run it.

If you do not currently have Drill installed, simply download the tarball, uncompress it, and add its bin directory in your $PATH. No additional setup is needed. (cvs2parquet just uses the drill-embedded executable.)

Currently, csv2parquet runs on OS X and Linux. It has not been tested on Windows, though Windows support is intended, and I appreciate comments, pull requests, etc. to support Windows users.

Regarding Python versions: Note that Python 3 safely installs alongside Python 2 with no conflict: even the executables are named differently ("python" for 2.7, and "python3" for 3.x). So you can simply install it to run csv2parquet today on any system you control.

Future Work

In terms of priority:

• Adding certain important features, including:
  • delimiters other than commas
  • CSV files without header lines
• Running csv2parquet on Windows

About and Contact

Written by Aaron Maxwell. Contact him at [email protected].

Licensed under GPLv3.

For bug reports, please run with the --debug option (see "Troubleshooting" above), and email the script, script_stderr and script_stdout files to the author, along with a description of what happened, and a CSV file that will reproduce the error.