DevTut

# Queue

# Multiple Workers One Sink

We want to gather data created by multiple Workers.

First we create a Queue:

sink = Queue.new

Then 16 workers all generating a random number and pushing it into sink:

(1..16).to_a.map do
  Thread.new do
    sink << rand(1..100)
  end
end.map(&:join)

And to get the data, convert a Queue to an Array:

data = [].tap { |a| a << sink.pop until sink.empty? }

# Converting a Queue into an Array

q = Queue.new
q << 1
q << 2

a = Array.new
a << q.pop until q.empty?

Or a one liner (opens new window):

[].tap { |array| array < queue.pop until queue.empty? }

# One Source Multiple Workers

We want to process data in parallel.

Let's populate source with some data:

source = Queue.new
data = (1..100)
data.each { |e| source << e }

Then create some workers to process data:

(1..16).to_a.map do
  Thread.new do
    until source.empty?
      item = source.pop
      sleep 0.5
      puts "Processed: #{item}"
    end
  end
end.map(&:join)

# One Source - Pipeline of Work - One Sink

We want to process data in parallel and push it down the line to be processed by other workers.

Since Workers both consume and produce data we have to create two queues:

first_input_source = Queue.new
first_output_sink  = Queue.new
100.times { |i| first_input_source << i }

First wave of workers read an item from first_input_source, process the item, and write results in first_output_sink:

(1..16).to_a.map do
  Thread.new do
    loop do
      item = first_input_source.pop
      first_output_source << item ** 2
      first_output_source << item ** 3
    end
  end
end

Second wave of workers uses first_output_sink as its input source and reads, process then writes to another output sink:

second_input_source = first_output_sink
second_output_sink  = Queue.new

(1..32).to_a.map do
  Thread.new do
    loop do
      item = second_input_source.pop
      second_output_sink << item * 2
      second_output_sink << item * 3
    end
  end
end

Now second_output_sink is the sink, let's convert it to an array:

sleep 5 # workaround in place of synchronization
sink = second_output_sink
[].tap { |a| a << sink.pop until sink.empty? }

# Pushing Data into a Queue - #push

q = Queue.new
q << "any object including another queue"
# or
q.push :data
  • There is no high water mark, queues can infinitely grow.
  • #push never blocks

# Pulling Data from a Queue - #pop

q = Queue.new
q << :data
q.pop #=> :data
  • #pop will block until there is some data available.
  • #pop can be used for synchronization.

# Synchronization - After a Point in Time

syncer = Queue.new

a = Thread.new do
  syncer.pop
  puts "this happens at end"
end

b = Thread.new do
  puts "this happens first"
  STDOUT.flush
  syncer << :ok
end

[a, b].map(&:join)

# Merging Two Queues

  • To avoid infinitely blocking, reading from queues shouldn't happen on the thread merge is happening on.
  • To avoid synchronization or infinitely waiting for one of queues while other has data, reading from queues shouldn't happen on same thread.

Let's start by defining and populating two queues:

q1 = Queue.new
q2 = Queue.new
(1..100).each { |e| q1 << e }
(101..200).each { |e| q2 << e }

We should create another queue and push data from other threads into it:

merged = Queue.new

[q1, q2].map do |q|
  Thread.new do
    loop do
      merged << q.pop
    end
  end
end

If you know you can completely consume both queues (consumption speed is higher than production, you won't run out of RAM) there is a simpler approach:

merged = Queue.new
merged << q1.pop until q1.empty?
merged << q2.pop until q2.empty?

# Syntax

  • q = Queue.new
  • q.push object
  • q << object # same as #push
  • q.pop #=> object

← Destructuring →