Features

• Converts CSV files into Parquet.
• Creates Partitions

Usage

Project Status: Dev Testing - Seems to be working, but needs more testing. Do not use in production.

Designed to run on a Hadoop edge node.

Examples

java -jar toparquet.jar -if myfile1.csv myfile2.csv -o myparquet

java -jar toparquet.jar -ip folder/mydir/ -o myparquet -s myschema.par

More Information

java -jar toparquet.jar -h

Advanced/Json Configuration Guide

Development Readme

Goals

The motivation of this project is to allow an in-memory data processing framework, such as Spark, to be able to process a large file that might not fit into a node's memory if it were a CSV. In some unfortunate circumstances, spark might not have enough memory to do the conversion itself.

• Ability to quickly and easily convert large files into parquet with minimal memory usage.
• Preprocess files to be more easily used by spark processing.

Non-Goals

• This project does not intend to become a data transformation / processing tool in its own right.

Feature Roadmap

• Stable CSV to Parquet Conversion
• Streaming Conversion via Command Line Pipe
• Failure Tolerance
• Add Calculated Columns to Output
• Basic Filtering
• Explode Records (To deal with high-skew joins)
• Dedupe Records
• Isolate Large Columns into their own Files
• Multiply Readers for Input Bottlenecks
• Multiply Writers for Output Bottlenecks
• Array Support
• Per Partition Ordering
• Option to use Cloud Object Storage instead of Hadoop
• Multi-Node Processing?
• Other Input Formats?