map-reduce-ruby 1.0.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 19ac70d25d6622d30398997cae0a4d6730ebc13f8bbc31df896c5acf480923c4
+  data.tar.gz: ede2a8f736053a268175a7948324a15954475526d0589bf47416347b57a50740
+SHA512:
+  metadata.gz: 650125350b5f166ccc0fb1346c77dfa9f17471ac9ae9290b4cbc4731a5baec872a810551e0b15af064b37f73c7e103ab6281d48cd839f26d4aab6780d5cd66d1
+  data.tar.gz: 70d9f8668143cb507537e369ad0ea79d9bb9f2c33469dffccd5952116f09758a7a880831d81aa4e3b67bc548b4b36aafe321a8d553946d8734fb78c11cd1686e

data/.github/workflows/test.yml

@@ -0,0 +1,23 @@
+name: test
+on: [push, pull_request]
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        ruby: ['2.6', '2.7', '3.0']
+    steps:
+    - uses: actions/checkout@v1
+    - uses: actions/setup-ruby@v1
+      with:
+        ruby-version: ${{ matrix.ruby }}
+    - uses: actions/cache@v1
+      id: cache
+      with:
+        path: vendor/bundler
+        key: ${{ hashFiles('Gemfile.lock') }}-${{ matrix.ruby }}
+    - run: |
+        gem install bundler
+        bundle install --path=vendor/bundler
+        bundle exec rspec
+        bundle exec rubocop

data/.gitignore

@@ -0,0 +1,11 @@
+/.bundle/
+/.yardoc
+/_yardoc/
+/coverage/
+/doc/
+/pkg/
+/spec/reports/
+/tmp/
+
+# rspec failure tracking
+.rspec_status

data/.rspec

@@ -0,0 +1,3 @@
+--format documentation
+--color
+--require spec_helper

data/.rubocop.yml

@@ -0,0 +1,51 @@
+AllCops:
+  NewCops: enable
+
+Naming/FileName:
+  Exclude:
+    - lib/map-reduce-ruby.rb
+
+Style/StringConcatenation:
+  Exclude:
+    - spec/**/*.rb
+
+Metrics/BlockLength:
+  Enabled: false
+
+Gemspec/RequiredRubyVersion:
+  Enabled: false
+
+Style/MutableConstant:
+  Enabled: false
+
+Metrics/MethodLength:
+  Enabled: false
+
+Style/Documentation:
+  Enabled: false
+
+Style/NumericPredicate:
+  Enabled: false
+
+Metrics/AbcSize:
+  Enabled: false
+
+Metrics/CyclomaticComplexity:
+  Enabled: false
+
+Metrics/PerceivedComplexity:
+  Enabled: false
+
+Style/StringLiterals:
+  Enabled: true
+  EnforcedStyle: double_quotes
+
+Style/StringLiteralsInInterpolation:
+  Enabled: true
+  EnforcedStyle: double_quotes
+
+Layout/LineLength:
+  Max: 120
+
+Style/FrozenStringLiteralComment:
+  EnforcedStyle: never

data/CHANGELOG.md

@@ -0,0 +1 @@
+# CHANGELOG

data/Gemfile

@@ -0,0 +1,4 @@
+source "https://rubygems.org"
+
+# Specify your gem's dependencies in map_reduce.gemspec
+gemspec

data/Gemfile.lock

@@ -0,0 +1,58 @@
+PATH
+  remote: .
+  specs:
+    map-reduce-ruby (1.0.0)
+      json
+      lazy_priority_queue
+
+GEM
+  remote: https://rubygems.org/
+  specs:
+    ast (2.4.2)
+    diff-lcs (1.4.4)
+    json (2.5.1)
+    lazy_priority_queue (0.1.1)
+    parallel (1.20.1)
+    parser (3.0.0.0)
+      ast (~> 2.4.1)
+    rainbow (3.0.0)
+    regexp_parser (2.0.3)
+    rexml (3.2.4)
+    rspec (3.10.0)
+      rspec-core (~> 3.10.0)
+      rspec-expectations (~> 3.10.0)
+      rspec-mocks (~> 3.10.0)
+    rspec-core (3.10.1)
+      rspec-support (~> 3.10.0)
+    rspec-expectations (3.10.1)
+      diff-lcs (>= 1.2.0, < 2.0)
+      rspec-support (~> 3.10.0)
+    rspec-mocks (3.10.1)
+      diff-lcs (>= 1.2.0, < 2.0)
+      rspec-support (~> 3.10.0)
+    rspec-support (3.10.1)
+    rubocop (0.93.1)
+      parallel (~> 1.10)
+      parser (>= 2.7.1.5)
+      rainbow (>= 2.2.2, < 4.0)
+      regexp_parser (>= 1.8)
+      rexml
+      rubocop-ast (>= 0.6.0)
+      ruby-progressbar (~> 1.7)
+      unicode-display_width (>= 1.4.0, < 2.0)
+    rubocop-ast (1.4.1)
+      parser (>= 2.7.1.5)
+    ruby-progressbar (1.11.0)
+    unicode-display_width (1.7.0)
+
+PLATFORMS
+  ruby
+  x86_64-linux
+
+DEPENDENCIES
+  map-reduce-ruby!
+  rspec
+  rubocop
+
+BUNDLED WITH
+   2.2.2

data/LICENSE.txt

@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2021 Benjamin Vetter
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

data/README.md

@@ -0,0 +1,211 @@
+# MapReduce
+
+**The easiest way to write distributed, larger than memory map-reduce jobs**
+
+The MapReduce gem provides the easiest way to write custom, distributed, larger
+than memory map-reduce jobs by using your local disk and some arbitrary storage
+layer like s3. You can specify how much memory you are willing to offer and
+MapReduce will use its buffers accordingly. Finally, you can use your already
+existing background job system like `sidekiq` or one of its various
+alternatives. Finally, your keys and values can be everything that can be
+serialized as json.
+
+## Installation
+
+Add this line to your application's Gemfile:
+
+```ruby
+gem 'map-reduce-ruby'
+```
+
+And then execute:
+
+    $ bundle install
+
+Or install it yourself as:
+
+    $ gem install map-reduce-ruby
+
+## Usage
+
+Any map-reduce job consists of an implementation of your `map` function, your
+`reduce` function and worker code. So let's start with an implementation for a
+word count map-reduce task which fetches txt documents from the web. Please
+note that your keys and values can be everything that can be serialized as
+json, but nothing else.
+
+```ruby
+class WordCounter
+  def map(url)
+    HTTP.get(url).to_s.split.each do |word|
+      yield(word, 1)
+    end
+  end
+
+  def reduce(word, count1, count2)
+    count1 + count2
+  end
+end
+```
+
+The `#map` method takes some key, e.g. a url, and yields an arbitrary amount of
+key-value pairs. The `#reduce` method takes the key as well as two values and
+should return a single reduced value.
+
+Next, we need some worker code to run the mapping part:
+
+```ruby
+class WordCountMapper
+  def perform(job_id, mapper_id, url)
+    mapper = MapReduce::Mapper.new(WordCounter.new, partitioner: MapReduce::HashPartitioner.new(16), memory_limit: 100.megabytes)
+    mapper.map(url)
+
+    mapper.shuffle do |partition, tempfile|
+      # store content of tempfile e.g. on s3:
+      bucket.object("map_reduce/jobs/#{job_id}/partitions/#{partition}/chunk.#{mapper_id}.json").put(body: tempfile)
+    end
+  end
+end
+```
+
+Please note that `MapReduce::HashPartitioner.new(16)` states that we want split
+the dataset into 16 partitions (i.e. 0, 1, ... 15). Finally, we need some
+worker code to run the reduce part:
+
+```ruby
+class WordCountReducer
+  def perform(job_id, partition)
+    reducer = MapReduce::Reducer.new(WordCounter.new)
+
+    # fetch all chunks of the partitions e.g. from s3:
+    bucket.list(prefix: "map_reduce/jobs/#{job_id}/partitions/#{partition}/").each do |object|
+      chunk_path = reducer.add_chunk # returns a path to a tempfile
+
+      object.download_file(temp_path)
+    end
+
+    reducer.reduce(chunk_limit: 32) do |word, count|
+      # each word with its final count
+    end
+  end
+end
+```
+
+Please note that `MapReduce::Reducer#add_chunk` returns a path to a tempfile,
+not a `Tempfile` object. This allows to limit the number of open file
+descriptors.
+
+To run your mappers, you can do:
+
+```ruby
+job_id = SecureRandom.hex
+
+list_of_urls.each_with_index do |url, index|
+  WordCountMapper.perform_async(job_id, index, url)
+end
+```
+
+And to run your reducers:
+
+```ruby
+(0..15).each do |partition|
+  WordCountReducer.perform_async(job_id, partition)
+end
+```
+
+How to automate running the mappers and reducers in sequence, depends on your
+background job system. The most simple approach is e.g. to track your mapper
+state in redis and have a job to start your reducers which waits up until your
+mappers are finished.
+
+That's it.
+
+## Internals
+
+To fully understand the performance details, the following outlines the inner
+workings of MapReduce. Of course, feel free to check the code as well.
+
+`MapReduce::Mapper#map` calls your `map` implementation and adds each yielded
+key-value pair to an internal buffer up until the memory limit is reached.
+When the memory limit is reached, the buffer is sorted by key and fed through
+your `reduce` implementation already, as this can greatly reduce the amount of
+data already. The result is written to a tempfile. This proceeds up until all
+key-value pairs are yielded. `MapReduce::Mapper#shuffle` then reads the first
+key-value pair of all already sorted chunk tempfiles and adds them to a
+priority queue using a binomial heap, such that with every `pop` operation on
+that heap, we get items sorted by key. When the item returned by `pop` e.g.
+belongs to the second chunk, then the next key-value pair of the second chunk
+is subsequently read and added to the priority queue, up until no more pairs
+are available. This guarantees that we sort all chunks without fully loading
+them into memory and is called `k-way-merge`. With every `pop` operation, your
+`reduce` implementation is continously called up until the key changes between
+two calls to `pop`. When the key changes, the key is known to be fully reduced,
+such that the key is hashed modulo the number of partitions and gets written to
+the correct partition tempfile (when `MapReduce::HashPartitioner` is used).
+
+The resulting partition tempfiles need to be stored in some global storage
+system like s3, such that your mapper workers can upload them and the reducer
+workers can download them.
+
+`MapReduce::Reducer#add_chunk` adds and registers a new tempfile path such that
+your reducer can download a mapper file for the particular partition and write
+its contents to that tempfile path. `MapReduce::Reducer#reduce` finally again
+builds up a priority queue and performs `k-way-merge`, feeds the key-value
+pairs into your reduce implementation up until a key change between `pop`
+operations occurs and yields the fully reduced key-value pair. At the end
+`#reduce` removes all the tempfiles. You can pass a `chunk_limit` to
+`MapReduce::Reducer#reduce`, which is most useful when you run on a system with
+a limited number of open file descriptors allowed. The `chunk_limit` ensures
+that only the specified amount of chunks are processed in a single run. A run
+basically means: it takes up to `chunk_limit` chunks, reduces them and pushes
+the result as a new chunk to the list of chunks to process. Thus, if your
+number of file descriptors is unlimited, you want to set it to a higher number
+to avoid the overhead of multiple runs.
+
+## Partitioners
+
+Partitioners are used to split the dataset into a specified amount of
+partitions, which allows to parallelize the work to be done by reducers.
+MapReduce comes with a `HashPartitioner`, which takes the number of partitions
+as an argument and derives the partition number from the key as follows:
+
+```ruby
+class HashPartitioner
+  def initialize(num_partitions)
+    @num_partitions = num_partitions
+  end
+
+  def call(key)
+    Digest::SHA1.hexdigest(JSON.generate(key))[0..4].to_i(16) % @num_partitions
+  end
+end
+```
+
+Thus, writing your own custom partitioner is really easy and, as it follows the
+interface of callables, could even be expressed as a simple one-liner:
+
+```ruby
+MyPartitioner = proc { |key| Digest::SHA1.hexdigest(JSON.generate(key))[0..4].to_i(16) % 8 }
+```
+
+## Development
+
+After checking out the repo, run `bin/setup` to install dependencies. Then, run
+`rake spec` to run the tests. You can also run `bin/console` for an interactive
+prompt that will allow you to experiment.
+
+To install this gem onto your local machine, run `bundle exec rake install`. To
+release a new version, update the version number in `version.rb`, and then run
+`bundle exec rake release`, which will create a git tag for the version, push
+git commits and the created tag, and push the `.gem` file to
+[rubygems.org](https://rubygems.org).
+
+## Contributing
+
+Bug reports and pull requests are welcome on GitHub at
+https://github.com/mrkamel/map-reduce-ruby
+
+## License
+
+The gem is available as open source under the terms of the [MIT
+License](https://opensource.org/licenses/MIT).

data/Rakefile

@@ -0,0 +1,10 @@
+require "bundler/gem_tasks"
+require "rspec/core/rake_task"
+
+RSpec::Core::RakeTask.new(:spec)
+
+require "rubocop/rake_task"
+
+RuboCop::RakeTask.new
+
+task default: %i[spec rubocop]

data/bin/console

@@ -0,0 +1,14 @@
+#!/usr/bin/env ruby
+
+require "bundler/setup"
+require "map_reduce"
+
+# You can add fixtures and/or initialization code here to make experimenting
+# with your gem easier. You can also use a different console, if you like.
+
+# (If you use this, don't forget to add pry to your Gemfile!)
+# require "pry"
+# Pry.start
+
+require "irb"
+IRB.start(__FILE__)

data/bin/setup

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+IFS=$'\n\t'
+set -vx
+
+bundle install
+
+# Do any other automated setup that you need to do here

data/lib/map-reduce-ruby.rb

@@ -0,0 +1 @@
+require "map_reduce"

data/lib/map_reduce.rb

@@ -0,0 +1,16 @@
+require "tempfile"
+require "json"
+require "digest"
+require "fileutils"
+require "tmpdir"
+require "lazy_priority_queue"
+require "map_reduce/version"
+require "map_reduce/priority_queue"
+require "map_reduce/temp_path"
+require "map_reduce/mergeable"
+require "map_reduce/reduceable"
+require "map_reduce/hash_partitioner"
+require "map_reduce/mapper"
+require "map_reduce/reducer"
+
+module MapReduce; end

data/lib/map_reduce/hash_partitioner.rb

@@ -0,0 +1,32 @@
+module MapReduce
+  # The MapReduce::HashPartitioner calculates a partition for the passed keys
+  # using SHA1 modulo the desired number of partitions.
+
+  class HashPartitioner
+    # Initializes a HashPartitioner.
+    #
+    # @param num_partitions [Fixnum] The desired number of partitions.
+    #   Typically 8, 16, 32, 64, etc. but can be everything according to your
+    #   needs.
+    #
+    # @example
+    #   MapReduce::HashPartitioner.new(16)
+
+    def initialize(num_partitions)
+      @num_partitions = num_partitions
+    end
+
+    # Calculates the partition for the specified key.
+    #
+    # @param key The key to calculate the partition for. Can be everything
+    #   that can be serialized as json.
+    # @returns [Integer] The partition number.
+    #
+    # @example
+    #   partitioner.call("some key")
+
+    def call(key)
+      Digest::SHA1.hexdigest(JSON.generate(key))[0..4].to_i(16) % @num_partitions
+    end
+  end
+end

data/lib/map_reduce/mapper.rb

@@ -0,0 +1,113 @@
+module MapReduce
+  # The MapReduce::Mapper class runs the mapping part of your map-reduce job.
+
+  class Mapper
+    include Mergeable
+    include Reduceable
+    include MonitorMixin
+
+    attr_reader :partitions
+
+    # Initializes a new mapper.
+    #
+    # @param implementation Your map-reduce implementation, i.e. an object
+    #   which responds to #map and #reduce.
+    # @param partitioner [#call] A partitioner, i.e. an object which responds
+    #   to #call and calculates a partition for the passed key.
+    # @param memory_limit [#to_i] The memory limit, i.e. the buffer size in
+    #   bytes.
+    #
+    # @example
+    #  MapReduce::Mapper.new(MyImplementation.new, partitioner: HashPartitioner.new(16), memory_limit: 100.megabytes)
+
+    def initialize(implementation, partitioner: HashPartitioner.new(32), memory_limit: 100 * 1024 * 1024)
+      super()
+
+      @implementation = implementation
+      @partitioner = partitioner
+      @memory_limit = memory_limit.to_i
+
+      @buffer_size = 0
+      @buffer = []
+      @chunks = []
+    end
+
+    # Passes the received key to your map-reduce implementation and adds
+    # yielded key-value pair to a buffer. When the memory limit is reached, the
+    # chunk is sorted and written to a tempfile.
+    #
+    # @param key The key to pass to the map-reduce implementation.
+    #
+    # @example
+    #   mapper.map("some_key")
+    #   mapper.map("other_key")
+
+    def map(*args, **kwargs)
+      @implementation.map(*args, **kwargs) do |new_key, new_value|
+        synchronize do
+          @buffer.push([new_key, new_value])
+
+          @buffer_size += JSON.generate([new_key, new_value]).bytesize
+
+          write_chunk if @buffer_size >= @memory_limit
+        end
+      end
+    end
+
+    # Performs a k-way-merge of the sorted chunks written to tempfiles while
+    # already reducing the result using your map-reduce implementation and
+    # splitting the dataset into partitions. Finally yields each partition with
+    # the tempfile containing the data of the partition.
+    #
+    # @example
+    #   mapper.shuffle do |partition, tempfile|
+    #     # store data e.g. on s3
+    #   end
+
+    def shuffle(&block)
+      return enum_for(:shuffle) unless block_given?
+
+      write_chunk if @buffer_size > 0
+
+      partitions = {}
+
+      reduce_chunk(k_way_merge(@chunks), @implementation).each do |pair|
+        partition = @partitioner.call(pair[0])
+
+        (partitions[partition] ||= Tempfile.new).puts(JSON.generate(pair))
+      end
+
+      @chunks.each { |tempfile| tempfile.close(true) }
+      @chunks = []
+
+      partitions.each_value(&:rewind)
+
+      partitions.each do |partition, tempfile|
+        block.call(partition, tempfile)
+      end
+
+      partitions.each_value { |tempfile| tempfile.close(true) }
+
+      nil
+    end
+
+    private
+
+    def write_chunk
+      tempfile = Tempfile.new
+
+      @buffer.sort_by! { |item| JSON.generate(item.first) }
+
+      reduce_chunk(@buffer, @implementation).each do |pair|
+        tempfile.puts JSON.generate(pair)
+      end
+
+      tempfile.rewind
+
+      @chunks.push(tempfile)
+
+      @buffer_size = 0
+      @buffer = []
+    end
+  end
+end

data/lib/map_reduce/mergeable.rb

@@ -0,0 +1,56 @@
+module MapReduce
+  # The MapReduce::Mergeable mixin provides the k-way-merge operation used by
+  # mappers as well as reducers.
+
+  module Mergeable
+    private
+
+    # Performs the k-way-merge of the passed files using a priority queue using
+    # a binomial heap. The content of the passed files needs to be sorted. It
+    # starts by reading one item of each file and adding it to the priority
+    # queue. Afterwards, it continously pops an item from the queue, yields it
+    # and reads a new item from the file the popped item belongs to, adding the
+    # read item to the queue. This continues up until all items from the files
+    # have been read. This guarantees that the yielded key-value pairs are
+    # sorted without having all items in-memory.
+    #
+    # @param files [IO, Tempfile] The files to run the k-way-merge for. The
+    #   content of the files must be sorted.
+
+    def k_way_merge(files)
+      return enum_for(:k_way_merge, files) unless block_given?
+
+      queue = PriorityQueue.new
+
+      files.each_with_index do |file, index|
+        line = file.eof? ? nil : file.readline
+
+        next unless line
+
+        key, value = JSON.parse(line)
+
+        queue.push([key, value, index], JSON.generate(key))
+      end
+
+      loop do
+        key, value, index = queue.pop
+
+        return unless index
+
+        yield([key, value])
+
+        line = files[index].yield_self { |file| file.eof? ? nil : file.readline }
+
+        next unless line
+
+        key, value = JSON.parse(line)
+
+        queue.push([key, value, index], JSON.generate(key))
+      end
+
+      files.each(&:rewind)
+
+      nil
+    end
+  end
+end

data/lib/map_reduce/priority_queue.rb

@@ -0,0 +1,49 @@
+module MapReduce
+  # The MapReduce::PriorityQueue implements a min priority queue using a
+  # binomial heap.
+
+  class PriorityQueue
+    # Initializes the priority queue.
+    #
+    # @example
+    #   MapReduce::PriorityQueue.new
+
+    def initialize
+      @queue = MinPriorityQueue.new
+      @sequence_number = 0
+    end
+
+    # Adds a new item to the priority queue while the key is used for sorting.
+    # The object and key can basically be everything, but the key must be some
+    # comparable object.
+    #
+    # @param object The object to add to the priority queue.
+    # @param key The key to use for sorting.
+    #
+    # @example
+    #   priority_queue = MapReduce::PriorityQueue.new
+    #   priority_queue.push("some object", "some key")
+
+    def push(object, key)
+      @queue.push([@sequence_number, object], key)
+
+      @sequence_number += 1
+    end
+
+    # Pops the min item from the queue.
+    #
+    # @returns The popped object.
+    #
+    # @example
+    #   priority_queue = MapReduce::PriorityQueue.new
+    #   priority_queue.push("object1", "key1")
+    #   priority_queue.push("object2", "key2")
+    #   priority_queue.pop
+
+    def pop
+      _, object = @queue.pop
+
+      object
+    end
+  end
+end

data/lib/map_reduce/reduceable.rb

@@ -0,0 +1,38 @@
+module MapReduce
+  # The MapReduce::Reduceable mixin allows to reduce an arbitrary chunk using
+  # the specified map-reduce implementation.
+
+  module Reduceable
+    private
+
+    # Reduces the specified chunk, e.g. some enumerable, using the specified
+    # map-reduce implementation using a lookahead of one to detect key changes.
+    # The reduce implementation is called up until a key change is detected,
+    # because the key change signals that the reduce operation is finished for
+    # the particular key, such that it will then be yielded.
+    #
+    # @param chunk The chunk to be reduced. Can e.g. be some enumerable.
+    # @param implementation The map-reduce implementation.
+
+    def reduce_chunk(chunk, implementation)
+      return enum_for(:reduce_chunk, chunk, implementation) unless block_given?
+
+      last_item = chunk.inject do |prev_item, cur_item|
+        prev_key = prev_item[0]
+
+        # Here we can compare without serializing the keys to json first,
+        # because we reduce a chunk which includes a deserialization step.
+
+        if prev_key == cur_item[0]
+          [prev_key, implementation.reduce(prev_key, prev_item[1], cur_item[1])]
+        else
+          yield(prev_item)
+
+          cur_item
+        end
+      end
+
+      yield(last_item) if last_item
+    end
+  end
+end

data/lib/map_reduce/reducer.rb

@@ -0,0 +1,107 @@
+module MapReduce
+  # The MapReduce::Reducer class runs the reducer part of your map-reduce job.
+
+  class Reducer
+    include Mergeable
+    include Reduceable
+    include MonitorMixin
+
+    class InvalidChunkLimit < StandardError; end
+
+    # Initializes a new reducer.
+    #
+    # @param implementation Your map-reduce implementation, i.e. an object
+    #   which responds to #map and #reduce.
+    #
+    # @example
+    #   MapReduce::Reducer.new(MyImplementation.new)
+
+    def initialize(implementation)
+      super()
+
+      @implementation = implementation
+
+      @temp_paths ||= []
+    end
+
+    # Adds a chunk from the mapper-phase to the reducer by registering a
+    # tempfile and returning the path to that tempfile, such that you can
+    # download a chunk e.g. from s3 and write the content to this tempfile.
+    #
+    # @returns [String] The path to a tempfile.
+    #
+    # @example
+    #   chunk_path = reducer.add_chunk
+    #   File.write(chunk_path, "downloaded blob")
+
+    def add_chunk
+      temp_path = TempPath.new
+
+      synchronize do
+        @temp_paths.push(temp_path)
+      end
+
+      temp_path.path
+    end
+
+    # Performs a k-way-merge of the added chunks and yields the reduced
+    # key-value pairs. It performs multiple runs when more than `chunk_limit`
+    # chunks exist. A run means: it takes up to `chunk_limit` chunks,
+    # reduces them and pushes the result as a new chunk. At the end it
+    # removes all tempfiles, even if errors occur.
+    #
+    # @param chunk_limit [Integer] The maximum number of files to process
+    #   during a single run. Most useful when you run on a system where the
+    #   number of open file descriptors is limited. If your number of file
+    #   descriptors is unlimited, you want to set it to a higher number to
+    #   avoid the overhead of multiple runs.
+    #
+    # @example
+    #   reducer = MapReduce::Reducer.new(MyImplementation.new)
+    #
+    #   chunk1_path = reducer.add_chunk
+    #   # write data to the file
+    #
+    #   chunk2_path = reducer.add_chunk
+    #   # write data to the file
+    #
+    #   reducer.reduce(chunk_limit: 32) do |key, value|
+    #     # ...
+    #   end
+
+    def reduce(chunk_limit:, &block)
+      return enum_for(:reduce, chunk_limit: chunk_limit) unless block_given?
+
+      raise(InvalidChunkLimit, "Chunk limit must be >= 2") unless chunk_limit >= 2
+
+      begin
+        loop do
+          slice = @temp_paths.shift(chunk_limit)
+          files = slice.select { |temp_path| File.exist?(temp_path.path) }
+                       .map { |temp_path| File.open(temp_path.path, "r") }
+
+          begin
+            if @temp_paths.empty?
+              reduce_chunk(k_way_merge(files), @implementation).each do |pair|
+                block.call(pair)
+              end
+
+              return
+            end
+
+            File.open(add_chunk, "w") do |file|
+              reduce_chunk(k_way_merge(files), @implementation).each do |pair|
+                file.puts JSON.generate(pair)
+              end
+            end
+          ensure
+            files.each(&:close)
+            slice.each(&:delete)
+          end
+        end
+      ensure
+        @temp_paths.each(&:delete)
+      end
+    end
+  end
+end

data/lib/map_reduce/temp_path.rb

@@ -0,0 +1,42 @@
+module MapReduce
+  # The MapReduce::TempPath generates a tempfile path and automatically deletes
+  # the file when the object is garbage collected or manually deleted. Using
+  # this class instead of Tempfile allows to have less open file descriptors.
+
+  class TempPath
+    attr_reader :path
+
+    # Initializes a new tempfile path.
+    #
+    # @example
+    #   temp_path = MapReduce::TempPath.new
+    #   File.write(temp_path.path, "blob")
+
+    def initialize
+      @path = Dir::Tmpname.create("") do
+        # nothing
+      end
+
+      FileUtils.touch(@path)
+
+      ObjectSpace.define_finalizer(self, self.class.finalize(@path))
+    end
+
+    # @api private
+
+    def self.finalize(path)
+      proc { FileUtils.rm_f(path) }
+    end
+
+    # Allows to manually delete the tempfile.
+    #
+    # @example
+    #   temp_path = MapReduce::TempPath.new
+    #   File.write(temp_path.path, "blob")
+    #   temp_path.delete
+
+    def delete
+      FileUtils.rm_f(path)
+    end
+  end
+end

data/lib/map_reduce/version.rb

@@ -0,0 +1,3 @@
+module MapReduce
+  VERSION = "1.0.0"
+end

data/map-reduce-ruby.gemspec

@@ -0,0 +1,38 @@
+require_relative "lib/map_reduce/version"
+
+Gem::Specification.new do |spec|
+  spec.name          = "map-reduce-ruby"
+  spec.version       = MapReduce::VERSION
+  spec.authors       = ["Benjamin Vetter"]
+  spec.email         = ["vetter@flakks.com"]
+
+  spec.summary       = "The easiest way to write distributed, larger than memory map-reduce jobs"
+  spec.description   = "The MapReduce gem is the easiest way to write custom, distributed, larger "\
+                       "than memory map-reduce jobs"
+  spec.homepage      = "https://github.com/mrkamel/map-reduce-ruby"
+  spec.license       = "MIT"
+
+  spec.required_ruby_version = Gem::Requirement.new(">= 2.5.0")
+
+  spec.metadata["homepage_uri"] = spec.homepage
+  spec.metadata["source_code_uri"] = "https://github.com/mrkamel/map-reduce-ruby"
+  spec.metadata["changelog_uri"] = "https://github.com/mrkamel/map-reduce/blob/master/CHANGELOG.md"
+
+  # Specify which files should be added to the gem when it is released.
+  # The `git ls-files -z` loads the files in the RubyGem that have been added into git.
+  spec.files = Dir.chdir(File.expand_path(__dir__)) do
+    `git ls-files -z`.split("\x0").reject { |f| f.match(%r{\A(?:test|spec|features)/}) }
+  end
+  spec.bindir        = "exe"
+  spec.executables   = spec.files.grep(%r{\Aexe/}) { |f| File.basename(f) }
+  spec.require_paths = ["lib"]
+
+  spec.add_development_dependency "rspec"
+  spec.add_development_dependency "rubocop"
+
+  spec.add_dependency "json"
+  spec.add_dependency "lazy_priority_queue"
+
+  # For more information and examples about making a new gem, checkout our
+  # guide at: https://bundler.io/guides/creating_gem.html
+end

metadata

@@ -0,0 +1,126 @@
+--- !ruby/object:Gem::Specification
+name: map-reduce-ruby
+version: !ruby/object:Gem::Version
+  version: 1.0.0
+platform: ruby
+authors:
+- Benjamin Vetter
+autorequire: 
+bindir: exe
+cert_chain: []
+date: 2021-07-05 00:00:00.000000000 Z
+dependencies:
+- !ruby/object:Gem::Dependency
+  name: rspec
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+- !ruby/object:Gem::Dependency
+  name: rubocop
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+- !ruby/object:Gem::Dependency
+  name: json
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+- !ruby/object:Gem::Dependency
+  name: lazy_priority_queue
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+description: The MapReduce gem is the easiest way to write custom, distributed, larger
+  than memory map-reduce jobs
+email:
+- vetter@flakks.com
+executables: []
+extensions: []
+extra_rdoc_files: []
+files:
+- ".github/workflows/test.yml"
+- ".gitignore"
+- ".rspec"
+- ".rubocop.yml"
+- CHANGELOG.md
+- Gemfile
+- Gemfile.lock
+- LICENSE.txt
+- README.md
+- Rakefile
+- bin/console
+- bin/setup
+- lib/map-reduce-ruby.rb
+- lib/map_reduce.rb
+- lib/map_reduce/hash_partitioner.rb
+- lib/map_reduce/mapper.rb
+- lib/map_reduce/mergeable.rb
+- lib/map_reduce/priority_queue.rb
+- lib/map_reduce/reduceable.rb
+- lib/map_reduce/reducer.rb
+- lib/map_reduce/temp_path.rb
+- lib/map_reduce/version.rb
+- map-reduce-ruby.gemspec
+homepage: https://github.com/mrkamel/map-reduce-ruby
+licenses:
+- MIT
+metadata:
+  homepage_uri: https://github.com/mrkamel/map-reduce-ruby
+  source_code_uri: https://github.com/mrkamel/map-reduce-ruby
+  changelog_uri: https://github.com/mrkamel/map-reduce/blob/master/CHANGELOG.md
+post_install_message: 
+rdoc_options: []
+require_paths:
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: 2.5.0
+required_rubygems_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+requirements: []
+rubygems_version: 3.0.3
+signing_key: 
+specification_version: 4
+summary: The easiest way to write distributed, larger than memory map-reduce jobs
+test_files: []