ruby-concurrency/concurrent-ruby

View on GitHub
docs-source/channel.in.md

Summary

Maintainability
Test Coverage
## Examples

Let's start by creating a channel with a capacity of 2 messages.

```ruby
ch = Concurrent::Promises::Channel.new 2
```

We push 3 messages, 
then it can be observed that the last thread pushing is sleeping 
since the channel is full. 

```ruby
threads = Array.new(3) { |i| Thread.new { ch.push message: i } } #
sleep 0.01 # let the threads run
threads
```

When message is popped the last thread continues and finishes as well.

```ruby
ch.pop
threads.map(&:join)
```

Same principle applies to popping as well.
There are now 2 messages int he channel.
Lets create 3 threads trying to pop a message, 
one will be blocked until new messages is pushed.

```ruby
threads = Array.new(3) { |i| Thread.new { ch.pop } } #
sleep 0.01 # let the threads run
threads 
ch.push message: 3
threads.map(&:value)
```

### Promises integration

However this channel is implemented to **integrate with promises**
therefore all operations can be represented as futures.

```ruby
ch = Concurrent::Promises::Channel.new 2
push_operations = Array.new(3) { |i| ch.push_op message: i }
```

> We do not have to sleep here letting the futures execute as Threads.
Since there is capacity for 2 messages the Promises are immediately resolved 
without ever allocating a Thread to execute. 
Push and pop operations are often more efficient.
The remaining pending push operation will also never require another thread,
instead it will resolve when a message is popped from the channel
making a space for a new message.    

```ruby
ch.pop_op.value!
push_operations.map(&:value!)

pop_operations = Array.new(3) { |i| ch.pop_op }
ch.push message: 3 # (push|pop) can be freely mixed with (push_o|pop_op)
pop_operations.map(&:value) 
```

### Selecting over channels

A selection over channels can be created with the `.select_channel` factory method. It
will be fulfilled with a first message available in any of the channels. It
returns a pair to be able to find out which channel had the message available.

```ruby
ch1    = Concurrent::Promises::Channel.new 2
ch2    = Concurrent::Promises::Channel.new 2
ch1.push 1 
ch2.push 2 

Concurrent::Promises::Channel.select([ch1, ch2])
ch1.select(ch2)

Concurrent::Promises.future { 3 + 4 }.then_channel_push(ch1)
Concurrent::Promises::Channel. 
    # or `ch1.select_op(ch2)` would be equivalent
    select_op([ch1, ch2]).
    then('got number %03d from ch%d') { |(channel, value), format| 
      format format, value, [ch1, ch2].index(channel).succ
    }.value! 
```

### `try_` variants

All blocking operations ({#pop}, {#push}, {#select}) have non-blocking variant
with `try_` prefix. 
They always return immediately and indicate either success or failure.

```ruby
ch
ch.try_push 1
ch.try_push 2
ch.try_push 3
ch.try_pop
ch.try_pop
ch.try_pop
```

### Timeouts

All blocking operations ({#pop}, {#push}, {#select}) have a timeout option.
Similar to `try_` variants it will indicate success or timing out, 
when the timeout option is used.

```ruby
ch
ch.push 1, 0.01
ch.push 2, 0.01
ch.push 3, 0.01
ch.pop 0.01
ch.pop 0.01
ch.pop 0.01
```

### Backpressure

Most importantly the channel can be used to create systems with backpressure.
A self adjusting system where the producers will slow down 
if the consumers are not keeping up.

```ruby
channel = Concurrent::Promises::Channel.new 2
log     = Concurrent::Array.new

producers = Array.new 2 do |i|
  Thread.new(i) do |i|
    4.times do |j|
      log.push format "producer %d pushing %d", i, j      
      channel.push [i, j]      
    end
  end
end

consumers = Array.new 4 do |i|
  Thread.new(i) do |consumer|
    2.times do |j|
      from, message = channel.pop
      log.push format "consumer %d got %d. payload %d from producer %d", 
                      consumer, j, message, from       
      do_stuff      
    end
  end
end

# wait for all to finish
producers.map(&:join)
consumers.map(&:join)
# investigate log
log
```

The producers are much faster than consumers 
(since they `do_stuff` which takes some time)  
but as it can be seen from the log they fill the channel 
and then they slow down 
until there is space available in the channel.

If permanent allocation of threads to the producers and consumers has to be avoided,
the threads can be replaced with promises
that run a thread pool.

```ruby
channel = Concurrent::Promises::Channel.new 2
log     = Concurrent::Array.new

def produce(channel, log, producer, i)
  log.push format "producer %d pushing %d", producer, i      
  channel.push_op([producer, i]).then do
    i + 1 < 4 ? produce(channel, log, producer, i + 1) : :done    
  end      
end

def consume(channel, log, consumer, i)
  channel.pop_op.then(consumer, i) do |(from, message), consumer, i|
    log.push format "consumer %d got %d. payload %d from producer %d", 
                    consumer, i, message, from       
    do_stuff
    i + 1 < 2 ? consume(channel, log, consumer, i + 1) : :done       
  end
end

producers = Array.new 2 do |i|
  Concurrent::Promises.future(channel, log, i) { |*args| produce *args, 0 }.run
end

consumers = Array.new 4 do |i|
  Concurrent::Promises.future(channel, log, i) { |*args| consume *args, 0 }.run
end

# wait for all to finish
producers.map(&:value!)
consumers.map(&:value!)
# investigate log
log
```

### Synchronization of workers by passing a value

If the capacity of the channel is zero 
then any push operation will succeed only 
when there is a matching pop operation
which can take the message.
The operations have to be paired to succeed. 

```ruby
channel = Concurrent::Promises::Channel.new 0
thread = Thread.new { channel.pop }; sleep 0.01 #
# allow the thread to go to sleep
thread
# succeeds because there is matching pop operation waiting in the thread 
channel.try_push(:v1)
# remains pending, since there is no matching operation 
push = channel.push_op(:v2)
thread.value
# the push operation resolves as a pairing pop is called
channel.pop
push
```