iostreams

Gem Version Build Status Downloads License Gitter chat

Input and Output streaming for Ruby.

Project Status

Production Ready, but API is subject to breaking changes until V1 is released.

Features

Supported file / stream types:

Streaming support currently under development:

  • S3
  • SFTP

Supported file formats:

  • CSV
  • Fixed width formats
  • JSON
  • PSV

Introduction

If all files were small, they could just be loaded into memory in their entirety. With the advent of very large files, often into several Gigabytes, or even Terabytes in size, loading them into memory is not feasible.

In linux it is common to use pipes to stream data between processes. For example:

# Count the number of lines in a file that has been compressed with gzip
cat abc.gz | gunzip -c | wc -l

For large files it is critical to be able to read and write these files as streams. Ruby has support for reading and writing files using streams, but has no built-in way of passing one stream through another to support for example compressing the data, encrypting it and then finally writing the result to a file. Several streaming implementations exist for languages such as C++ and Java to chain together several streams, iostreams attempts to offer similar features for Ruby.

# Read a compressed file:
IOStreams.reader('hello.gz') do |reader|
  data = reader.read(1024)
  puts "Read: #{data}"
end

The true power of streams is shown when many streams are chained together to achieve the end result, without holding the entire file in memory, or ideally without needing to create any temporary files to process the stream.

# Create a file that is compressed with GZip and then encrypted with Symmetric Encryption:
IOStreams.writer('hello.gz.enc') do |writer|
  writer.write('Hello World')
  writer.write('and some more')
end

The power of the above example applies when the data being written starts to exceed hundreds of megabytes, or even gigabytes.

By looking at the file name supplied above, iostreams is able to determine which streams to apply to the data being read or written. For example:

  • hello.zip => Compressed using Zip
  • hello.zip.enc => Compressed using Zip and then encrypted using Symmetric Encryption
  • hello.gz.enc => Compressed using GZip and then encrypted using Symmetric Encryption

The objective is that all of these streaming processes are performed used streaming so that only the current portion of the file is loaded into memory as it moves through the entire file. Where possible each stream never goes to disk, which for example could expose un-encrypted data.

Examples

While decompressing the file, display 128 characters at a time from the file.

require 'iostreams'
IOStreams.reader('abc.csv') do |io|
  p data while (data = io.read(128))
end

While decompressing the file, display one line at a time from the file.

IOStreams.each_line('abc.csv') do |line|
  puts line
end

While decompressing the file, display each row from the csv file as an array.

IOStreams.each_row('abc.csv') do |array|
  p array
end

While decompressing the file, display each record from the csv file as a hash. The first line is assumed to be the header row.

IOStreams.each_record('abc.csv') do |hash|
  p hash
end

Display each line from the array as a hash. The first line is assumed to be the header row.

array = [
  'name, address, zip_code',
  'Jack, Down Under, 12345'
]
IOStreams.each_record(array) do |hash|
  p hash
end

Write data while compressing the file.

IOStreams.writer('abc.csv') do |io|
  io.write('This')
  io.write(' is ')
  io.write(" one line\n")
end

Write a line at a time while compressing the file.

IOStreams.line_writer('abc.csv') do |file|
  file << 'these'
  file << 'are'
  file << 'all'
  file << 'separate'
  file << 'lines'
end

Write an array (row) at a time while compressing the file. Each array is converted to csv before being compressed with zip.

IOStreams.row_writer('abc.csv') do |io|
  io << %w[name address zip_code]
  io << %w[Jack There 1234]
  io << ['Joe', 'Over There somewhere', 1234]
end

Write a hash (record) at a time while compressing the file. Each hash is converted to csv before being compressed with zip. The header row is extracted from the first hash supplied.

IOStreams.record_writer('abc.csv') do |stream|
  stream << {name: 'Jack', address: 'There', zip_code: 1234}
  stream << {name: 'Joe', address: 'Over There somewhere', zip_code: 1234}
end

Write to a string IO for testing, supplying the filename so that the streams can be determined.

io = StringIO.new
IOStreams::Tabular::Writer(io, file_name: 'abc.csv') do |stream|
  stream << {name: 'Jack', address: 'There', zip_code: 1234}
  stream << {name: 'Joe', address: 'Over There somewhere', zip_code: 1234}
end
puts io.string

Read a CSV file and write the output to an encrypted file in JSON format.

IOStreams.record_writer('sample.json.enc') do |output|
  IOStreams.each_record('sample.csv') do |record|
    output << record
  end
end

Copying between files

Stream based file copying. Changes the file type without changing the file format. For example, compress or encrypt.

Encrypt the contents of the file sample.json and write to sample.json.enc

IOStreams.copy('sample.json', 'sample.json.enc')

Encrypt and compress the contents of the file sample.json with Symmetric Encryption and write to sample.json.enc

IOStreams.copy('sample.json', 'sample.json.enc', target_options: {streams: {enc: {compress: true}}})

Encrypt and compress the contents of the file sample.json with pgp and write to sample.json.enc

IOStreams.copy('sample.json', 'sample.json.pgp', target_options: {streams: {pgp: {recipient: '[email protected]'}}})

Decrypt the file abc.csv.enc and write it to xyz.csv.

IOStreams.copy('abc.csv.enc', 'xyz.csv')

Read ABC, PGP encrypt the file and write to xyz.csv.pgp, applying 

IOStreams.copy('ABC', 'xyz.csv.pgp',
               source_options: [:enc],
               target_options: [pgp: {email_recipient: '[email protected]'})

Philosopy

IOStreams can be used to work against a single stream. it's real capability becomes apparent when chainging together multiple streams to process data, without loading entire files into memory.

Linux Pipes

Linux has built-in support for streaming using the | (pipe operator) to send the output from one process to another.

Example: count the number of lines in a compressed file: 

gunzip -c hello.csv.gz | wc -l

The file hello.csv.gz is uncompressed and returned to standard output, which in turn is piped into the standard input for wc -l, which counts the number of lines in the uncompressed data.

As each block of data is returned from gunzip it is immediately passed into wc so that it can start counting lines of uncompressed data, without waiting until the entire file is decompressed. The uncompressed contents of the file are not written to disk before passing to wc -l and the file is not loaded into memory before passing to wc -l.

In this way extremely large files can be processed with very little memory being used.

Push Model

In the Linux pipes example above this would be considered a "push model" where each task in the list pushes its output to the input of the next task.

A major challenge or disadvantage with the push model is that buffering would need to occur between tasks since each task could complete at very different speeds. To prevent large memory usage the standard output from a previous task would have to be blocked to try and make it slow down.

Pull Model

Another approach with multiple tasks that need to process a single stream, is to move to a "pull model" where the task at the end of the list pulls a block from a previous task when it is ready to process it.

IOStreams

IOStreams uses the pull model when reading data, where each stream performs a read against the previous stream when it is ready for more data.

When writing to an output stream, IOStreams uses the push model, where each block of data that is ready to be written is pushed to the task/stream in the list. The write push only returns once it has traversed all the way down to the final task / stream in the list, this avoids complex buffering issues between each task / stream in the list.

Example: Implementing in Ruby: gunzip -c hello.csv.gz | wc -l

  line_count = 0
  IOStreams::Gzip::Reader.open("hello.csv.gz") do |input|
    IOStreams::Line::Reader.open(input) do |lines|
      lines.each { line_count += 1}
    end
  end
  puts "hello.csv.gz contains #{line_count} lines"

Since IOStreams can autodetect file types based on the file extension, IOStreams.reader can figure which stream to start with:

  line_count = 0
  IOStreams.reader("hello.csv.gz") do |input|
    IOStreams::Line::Reader.open(input) do |lines|
      lines.each { line_count += 1}
    end
  end
  puts "hello.csv.gz contains #{line_count} lines"

Since we know we want a line reader, it can be simplified using IOStreams.line_reader:

  line_count = 0
  IOStreams.line_reader("hello.csv.gz") do |lines|
    lines.each { line_count += 1}
  end
  puts "hello.csv.gz contains #{line_count} lines"

It can be simplified even further using IOStreams.each_line:

  line_count = 0
  IOStreams.each_line("hello.csv.gz") { line_count += 1}
  puts "hello.csv.gz contains #{line_count} lines"

The benefit in all of the above cases is that the file can be any arbitrary size and only one block of the file is held in memory at any time.

Chaining

In the above example only 2 streams were used. Streams can be nested as deep as necessary to process data.

Example, search for all occurrences of the word apple, cleansing the input data stream of non printable characters and converting to valid US ASCII.

  apple_count = 0
  IOStreams::Gzip::Reader.open("hello.csv.gz") do |input|
    IOStreams::Encode::Reader.open(input, 
                                   encoding:       'US-ASCII', 
                                   encode_replace: '', 
                                   encode_cleaner: :printable) do |cleansed|
      IOStreams::Line::Reader.open(cleansed) do |lines|
        lines.each { |line| apple_count += line.scan('apple').count}
      end
  end
  puts "Found the word 'apple' #{apple_count} times in hello.csv.gz"

Let IOStreams perform the above stream chaining automatically under the covers:

  apple_count = 0
  IOStreams.each_line("hello.csv.gz", 
                      encoding:       'US-ASCII', 
                      encode_replace: '', 
                      encode_cleaner: :printable) do |line|
    apple_count += line.scan('apple').count
  end

  puts "Found the word 'apple' #{apple_count} times in hello.csv.gz"

Notes

  • Due to the nature of Zip, both its Reader and Writer methods will create a temp file when reading from or writing to a stream. Recommended to use Gzip over Zip since it can be streamed.
  • Zip becomes exponentially slower with very large files, especially files that exceed 4GB when uncompressed. Highly recommend using GZip for large files.

To completely implement io streaming for Ruby will take a lot more input and thoughts from the Ruby community. This gem represents a starting point to get the discussion going.

By keeping this gem a 0.x version and not going V1, we can change the interface as needed to implement community feedback.

Versioning

This project adheres to Semantic Versioning.

Author

Reid Morrison

License

Copyright 2018 Reid Morrison

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.