Reading from fifo with no writers blocks the whole process

Hi,

in my program I'm creating a thread, in which I open a FIFO and read
data from it.

pipe_listener_thread = Thread.new(ARGV[0]) do |fifoname|
  while true # loop to reopen fifo, when it ends because of no writers
    fi = File.open(fifoname, "r")
    while line = fi.gets
      puts "msg: #{line}"
    end
    fi.close
  end
end

Everything works fine, except that when the thread is waiting to open
the fifo, whole process - all threads seem to be blocked. Such thing
doesn't happen when waiting for data (inner loop) - where the thread
alone gets blocked.

Please tell, if there's something wrong with this behavior, or it should
work that way.

Regs,
Tomasz Wrobel

···

--
Posted via http://www.ruby-forum.com/.

in my program I'm creating a thread, in which I open a FIFO and read
data from it.

pipe_listener_thread = Thread.new(ARGV[0]) do |fifoname|
  while true # loop to reopen fifo, when it ends because of no writers
    fi = File.open(fifoname, "r")
    while line = fi.gets
      puts "msg: #{line}"
    end
    fi.close
  end
end

Everything works fine, except that when the thread is waiting to open
the fifo, whole process - all threads seem to be blocked. Such thing
doesn't happen when waiting for data (inner loop) - where the thread
alone gets blocked.

First of all you should change your code to use the block form of File.open. This avoids keeping the file descriptor open in case of errors.

You can do even more, see http://blog.rubybestpractices.com/posts/rklemme/001-Using_blocks_for_Robustness.html

Please tell, if there's something wrong with this behavior, or it should
work that way.

I believe this is perfectly ok. Since IO indicates eof if there is no more data in the fifo, it makes sense to block on open as long as there is nothing to read. Otherwise your program would mindlessly be looping and burning CPU.

I do wonder though why reading from a named fifo ever reaches eof. You would expect the reading to continue. OTOH that way you will be aware if a program terminates which writes to the fifo and can react on it appropriately.

Kind regards

  robert

···

On 30.04.2009 11:42, Tomasz Wrobel wrote: