Bramble

Map-reduce with ActiveJob + database

Rationale

We have some staff-only views that expose stats about how people use our app. Eventually, our tables grew so large that MySQL wouldn't aggregate them all at once. So we can use this to generate those stats over time.

Usage

• Setup ActiveJob with a queue named :bramble

• Setup Redis and give Bramble a connection object:

  my_redis_connection = Redis.new # Your connection settings here!
  Bramble.config do |conf|
    conf.redis_conn = my_redis_connection
    # By default, keys expire after 1 day.
    # Assign a value here to override that:
    # conf.expire_after = 60 * 60 * 24 * 7 # 7 days
    # conf.expire_after = nil # never expire
  end

• Define a module with map, reduce and items(options = {}) functions:

  module LetterCount
    # Generate a list of items based on some input
    def self.items(filepath)
      File.read(filepath).split(" ")
    end

    # .map is called with each item in the input
    def self.map(word)
      letters = word.upcase.each_char

      # call `yield` to emit a key-value pair for processing
      letters.each { |letter| yield(letter, 1) }
    end

    # .reduce is called with
    # - `yield` key (first argument)
    # - array of `yield` values (second argument)
    def self.reduce(letter, observations)
      # letter => "A"
      # observations => [1, 1, 1, 1, 1]
      observations.length
    end

    # If a .map or .reduce hits an error,
    # it will be rescued and passed here.
    # To cause the job to fail, raise it again.
    # Otherwise, let it pass
    def self.on_error(err)
      Bugsnag.notify(err)
      # Or, to trigger a faiure:
      # raise(err)
    end
  end

Inputs and outputs are serialized with JSON, so some Ruby types will be lost (eg, Symbols).

• Start a job with a handle, module, and an (optional) argument for finding data:

  # used for fetching the result later:
  handle = "shakespeare-letter-count"

  # will be sent to `.items(filepath)`
  hamlet_path = "./shakespeare/hamlet.txt"

  # Begin the process:
  Bramble.map_reduce(handle, LetterCount, hamlet_path)

• Later, fetch the result using the handle:

  result = Bramble.get("shakespeare-letter-count")
  result.running?         # => false
  result.finished?        # => true
  result.data             # => { "A" => 100, "B" => 100, ... }
  result.percent_finished # 1.0
  result.percent_mapped   # 1.0
  result.percent_reduced  # 1.0
  result.finished_at      # 2016-05-16 12:31:00 UTC

• Delete the saved result:

  Bramble.delete("shakespeare-letter-count")

Or delete everything:

  Bramble.delete_all

Todo

• Write some more adapters: Memcached, ActiveRecord
• Warn if no storage is configured
• Do we have atomicity issues? Occasional test failures
• Consolidate storage in Redis to a single key? (Could some keys be evicted while others remain?)

Development

• rake test