class NIO::Selector

Selectors monitor IO objects for events of interest

def self.backends

See `#backend` method definition for all possible backends

Return supported backends as symbols

def self.backends
  [:ruby]
end

def backend

* :unknown - libev w\ unknown backend
* :io_uring - libev w\ Linux io_uring (experimental)
* :linuxaio - libev w\ Linux AIO io_submit (experimental)
* :port - libev w\ I/O completion ports
* :select - libev w\ SysV select
* :kqueue - libev w\ BSD kqueue
* :poll - libev w\ POSIX poll
* :epoll - libev w\ Linux epoll
* :java - Java NIO on JRuby
* :ruby - pure Ruby (i.e IO.select)
Supported backends are:
Return a symbol representing the backend I/O multiplexing mechanism used.

def backend
  :ruby
end

def close

Close this selector and free its resources

def close
  @lock.synchronize do
    return if @closed
    begin
      @wakeup.close
    rescue IOError
    end
    begin
      @waker.close
    rescue IOError
    end
    @closed = true
  end
end

def closed?

Is this selector closed?

def closed?
  @closed
end

def deregister(io)

Deregister the given IO object from the selector

def deregister(io)
  @lock.synchronize do
    monitor = @selectables.delete IO.try_convert(io)
    monitor.close(false) if monitor && !monitor.closed?
    monitor
  end
end

def empty?

def empty?
  @selectables.empty?
end

def initialize(backend = :ruby)

Create a new NIO::Selector

def initialize(backend = :ruby)
  raise ArgumentError, "unsupported backend: #{backend}" unless [:ruby, nil].include?(backend)
  @selectables = {}
  @lock = Mutex.new
  # Other threads can wake up a selector
  @wakeup, @waker = IO.pipe
  @closed = false
end

def register(io, interest)

* :rw - is the IO either readable or writeable?
* :w - is the IO writeable?
* :r - is the IO readable?
of events. Valid event types for interest are:
Register interest in an IO object with the selector for the given types

def register(io, interest)
  unless defined?(::OpenSSL) && io.is_a?(::OpenSSL::SSL::SSLSocket)
    io = IO.try_convert(io)
  end
  @lock.synchronize do
    raise IOError, "selector is closed" if closed?
    monitor = @selectables[io]
    raise ArgumentError, "already registered as #{monitor.interests.inspect}" if monitor
    monitor = Monitor.new(io, interest, self)
    @selectables[monitor.io] = monitor
    monitor
  end
end

def registered?(io)

Is the given IO object registered with the selector?

def registered?(io)
  @lock.synchronize { @selectables.key? io }
end

def select(timeout = nil)

Select which monitors are ready

def select(timeout = nil)
  selected_monitors = Set.new
  @lock.synchronize do
    readers = [@wakeup]
    writers = []
    @selectables.each do |io, monitor|
      readers << io if monitor.interests == :r || monitor.interests == :rw
      writers << io if monitor.interests == :w || monitor.interests == :rw
      monitor.readiness = nil
    end
    ready_readers, ready_writers = Kernel.select(readers, writers, [], timeout)
    return unless ready_readers # timeout
    ready_readers.each do |io|
      if io == @wakeup
        # Clear all wakeup signals we've received by reading them
        # Wakeups should have level triggered behavior
        @wakeup.read(@wakeup.stat.size)
      else
        monitor = @selectables[io]
        monitor.readiness = :r
        selected_monitors << monitor
      end
    end
    ready_writers.each do |io|
      monitor = @selectables[io]
      monitor.readiness = monitor.readiness == :r ? :rw : :w
      selected_monitors << monitor
    end
  end
  if block_given?
    selected_monitors.each { |m| yield m }
    selected_monitors.size
  else
    selected_monitors.to_a
  end
end

def wakeup

level-triggered behavior.
has the same effect as invoking it just once. In other words, it provides
Invoking this method more than once between two successive select calls

any such thread exists.
Wake up a thread that's in the middle of selecting on this selector, if

def wakeup
  # Send the selector a signal in the form of writing data to a pipe
  begin
    @waker.write_nonblock "\0"
  rescue IO::WaitWritable
    # This indicates the wakeup pipe is full, which means the other thread
    # has already received many wakeup calls, but not processed them yet.
    # The other thread will completely drain this pipe when it wakes up,
    # so it's ok to ignore this exception if it occurs: we know the other
    # thread has already been signaled to wake up
  end
  nil
end

Modules

Classes