#apache-arrow #arrow #data-engineering #format #cli #apache-parquet

app evolution

๐Ÿฆ– Evolve your fixed-length data files into Apache Parquet, fully parallelized!

1 stable release

new 1.0.0 May 15, 2024
0.1.0 Nov 21, 2023

#135 in Parser implementations

Download history 4/week @ 2024-02-16 6/week @ 2024-02-23 3/week @ 2024-03-01 16/week @ 2024-03-29 6/week @ 2024-04-05 89/week @ 2024-05-10

89 downloads per month

Custom license

160KB
3K SLoC





License: MIT Crates.io (latest) codecov CI CD Tests

๐Ÿฆ– Evolve your fixed-length data files into Apache Parquet, fully parallelized!

๐Ÿ”Ž Overview

This repository hosts the evolution program which both allows you to convert existing fixed-length files into other data formats, but also allows you to create large amounts of mocked data blazingly fast. The program supports full parallelism and utilizes SIMD techniques, when possible, for highly efficient parsing of data.

To get started, follow the installation, schema setup, and example usage sections below in this README. Happy hacking! ๐Ÿ‘‹๐Ÿฅณ

๐Ÿ“‹ Table of contents

All of the code in this repo is open source and should be licensed according to LICENSE, refer to this link for more information.

๐Ÿ“ฆ Installation

The easiest way to install an evolution binary on your system is by using the Cargo package manager (which downloads it from this link).

$ cargo install evolution
    
(available features)    
 - rayon
 - nightly

Alternatively, you can build from source by cloning the repo and compiling using Cargo. See below for available optional features.

$ git clone https://github.com/firelink-data/evolution.git
$ cd evolution
$ cargo build --release

(optional: copy the binary to your users binray folder)
$ cp ./target/release/evolution /usr/bin/evolution
  • Installing with the rayon feature will utilize the rayon crate for parallel execution instead of the standard library threads. It also enables converting in chunked mode. Please see this reference for more information.
  • Installing with the nightly feature will use the nightly toolchain, which in nature is unstable. To be able to run this version you need the nightly toolchain installed on your system. You can install this by running rustup install nightly from your shell.

๐Ÿ“ Schema setup

All available commands in evolution require an existing valid schema. A schema, in this context, is a json file specifying the layout of the contents of a fixed-length file (flf). Every schema used has to adhere to this template. If you are unsure whether or not your own schema file is valid according to the template, you can use this validator tool.

An example schema can be found here, and looks like this:

{
    "name": "EvolutionExampleSchema",
    "version": 1337,
    "columns": [
        {
            "name": "id",
            "offset": 0,
            "length": 9,
            "dtype": "Int32",
            "alignment": "Right",
            "pad_symbol": "Underscore",
            "is_nullable": false
        },
        {
            "name": "name",
            "offset": 9,
            "length": 32,
            "dtype": "Utf8",
            "is_nullable": true
        },
        {
            "name": "city",
            "offset": 41,
            "length": 32,
            "dtype": "Utf8",
            "alignment": "Right",
            "pad_symbol": "Backslash",
            "is_nullable": false
        },
        {
            "name": "employed",
            "offset": 73,
            "length": 5,
            "dtype": "Boolean",
            "alignment": "Center",
            "pad_symbol": "Asterisk",
            "is_nullable": true
        }
    ]
}
  • If you are unsure about valid values for the dtype, alignment, and pad_symbol fields, please referr to the template which lists all valid values.
  • All columns have to provide the following fields name, offset, length, and is_nullable, whereas alignment and pad_symbol can be omitted (as they are in this example for the name column). If they are not provided, they will assume their default values which are "Right" and "Whitespace".
  • The default values come from the padder crate which defines the enums Alignment and Symbol, with default implementations as Alignment::Right and Symbol::Whitespace respectively.

๐Ÿš€ Example usage

If you install the program as explained above then by simply running the binary you will see the following helpful usage print:

๐Ÿฆ– Evolve your fixed-length data files into Apache Arrow tables, fully parallelized!

Usage: evolution [OPTIONS] <COMMAND>

Commands:
  convert  Convert a fixed-length file (.flf) to parquet
  mock     Generate mocked fixed-length files (.flf) for testing purposes
  help     Print this message or the help of the given subcommand(s)

Options:
      --n-threads <NUM-THREADS>  Set the number of threads (logical cores) to use when multi-threading [default: 1]
  -h, --help                     Print help
  -V, --version                  Print version

As you can see from above, the functionality of the program comprises of the two main commands: convert and mock. If you installed the program with the rayon feature you will also have access to a third command called c-convert. This stands for chunked-convert and is an alternative implementation. Documentation for this command is work-in-progress.

  • If you want to see debug prints during execution, set the RUST_LOG environment variable to DEBUG before executing the program.

๐Ÿ—๏ธ๐Ÿ‘ทโ€โ™‚๏ธ Converting

Convert a fixed-length file (.flf) to parquet

Usage: evolution convert [OPTIONS] --in-file <IN-FILE> --out-file <OUT-FILE> --schema <SCHEMA>

Options:
  -i, --in-file <IN-FILE>
          The fixed-length file to convert
  -o, --out-file <OUT-FILE>
          Specify output (target) file name
  -s, --schema <SCHEMA>
          Specify the .json schema file to use when converting
      --buffer-size <BUFFER-SIZE>
          Set the size of the buffer (in bytes)
      --thread-channel-capacity <THREAD-CHANNEL-CAPACITY>
          Set the capacity of the thread channel (number of messages)
  -h, --help
          Print help

To convert a fixed-length file called old-data.flf, with associated schema located at ./my/path/to/schema.json, to a parquet file with name converted.parquet, you could run the following command:

$ evolution convert --in-file old-data.flf --out-file converted.parquet --schema ./my/path/to/schema.json

๐Ÿ‘จโ€๐ŸŽจ Mocking

Generate mocked fixed-length files (.flf) for testing purposes

Usage: evolution mock [OPTIONS] --schema <SCHEMA>

Options:
  -s, --schema <SCHEMA>
          Specify the .json schema file to mock data for
  -o, --out-file <OUT-FILE>
          Specify output (target) file name
  -n, --n-rows <NUM-ROWS>
          Set the number of rows to generate [default: 100]
      --force-new
          Set the writer option to fail if the file already exists
      --truncate-existing
          Set the writer option to truncate a previous file if the out file already exists
      --buffer-size <MOCKER-BUFFER-SIZE>
          Set the size of the buffer (number of rows)
      --thread-channel-capacity <MOCKER-THREAD-CHANNEL-CAPACITY>
          Set the capacity of the thread channel (number of messages)
  -h, --help
          Print help

For example, if you wanted to mock 1 billion rows of a fixed-length file from a schema located at ./my/path/to/schema.json with the output name mocked-data.flf and enforce that the file should not already exist, you could run the following command:

$ evolution mock --schema ./my/path/to/schema.json --out-file mocked-data.flf --n-rows 1000000000 --force-new

๐Ÿงต Threading

There exists a global setting for the program called --n-threads which dictates whether or not the invoked command will be executed in single- or multithreaded mode. This argument should be a number representing the number of threads (logical cores) that you want to use. If you try and set a larger number of threads than you system has logical cores, then the program will use all available logical cores. If this argument is omitted, then the program will run in single-threaded mode.

Note that running in multithreaded mode only really has any clear increase in performance for substantially large workloads.

If you are unsure how many logical cores your CPU has, the easiest way to find out is by simply running the program with the --n-threads option set to a large number. The program will check how many logical cores you have and see whether this option exceeds the possible value. If the value you passed is greater than the number of logical cores on your system, then the number of logical cores available will be logged to you on stdout.

You could also potentially use one of the commands below depending on your host system.

Windows

$ Get-WmiObject Win32_Processor | Select-Object Name, NumberOfCores, NumberOfLogicalProcessors

Use the value found under NumberOfLogicalProcessors.

Unix

$ lscpu | grep -E '^Thread|^Core|^Socket|^CPU\('

The number of logical cores is calculed as: threads per core X cores per socket X sockets.

๐Ÿ“œ License

All code is to be held under a general MIT license, please see LICENSE for specific information.

Dependencies

~30โ€“44MB
~824K SLoC