DRb Fork Reproducer Script

echo_client.rb

# This script demonstrates the issue with mixing DRb calls and
# forking child processes.
#
# If the parent process calls a remote DRb method prior to forking
# a child process it appears there is still a connection in the
# DRbConn global connection pool.  This seems to cause messages to
# be sent to the wrong client and typically leads to one of three
# symptoms:
#
# 1. A message is sent to the wrong client
#    Ex: 'Failure! expected BBBB but got AAAA'
#
# 2. A message is spliced and what should be the message size is another string
#    Ex: 'Failure! too large packet 1090927110'
#
# 3. A message is spliced and what should be the Marshal version is another string
#    Ex: 'Failure! incompatible marshal file format (can't be read)
#         format version 4.8 required; 14.4 given'
#
#    For this case we added some client-side logging to dump the hex_bytes of the string:
#    [TypeError]: incompatible marshal file format (can't be read)
#      format version 4.8 required; 58.12 given
#    0x00000000  3a 0c 56 69 6d 48 61 73 68 7b 09 49 22 0b 65 6e  :.VimHash{.I".en
#    0x00000010  74 69 74 79 06 3a 06 45 54 49 43 3a 0e 56 69 6d  tity.:.ETIC:.Vim
#    0x00000020  53 74 72 69 6e 67 22 0d 76 6d 2d 31 38 37 35 37  String".vm-18757
#
#    The first bytes here which should be the marshal version look like the middle of
#    another message.

require 'drb/drb'

def echo_client(str)
  echo_obj = DRbObject.new_with_uri('druby://localhost:9999')

  1_000.times do
    begin
      ret = echo_obj.echo(str)
      if ret != str
        puts "PID #{Process.pid}: Failure! expected #{str} but got #{ret}"
        exit 1
      end
    rescue TypeError, DRb::DRbConnError => err
      puts "PID #{Process.pid}: Failure! #{err}"
      exit 1
    end
  end
  puts "PID #{Process.pid}: Succeded."
end

# Execute a DRb call in the parent process prior to forking the children
echo_obj = DRbObject.new_with_uri('druby://localhost:9999')
echo_obj.echo("CCCC")  # Comment this out and the script passes

# Create 20 child processes to echo some string to the server and verify
# the same string was returned
child_pids = []
%w(AAAA BBBB).each do |str|
  10.times do
    child_pids << Kernel.fork { echo_client(str) }
  end
end

child_pids.each { |pid| Process.wait(pid) }

echo_server.rb

require 'drb/drb'

class Echo
  def echo(str)
    str
  end
end

DRb.start_service('druby://localhost:9999', Echo.new)
DRb.thread.join

Example Run

$ ruby echo_server.rb &
[1] 28111
$ ruby echo_client.rb 
PID 28134: Failure! too large packet 67654656
PID 28128: Failure! incompatible marshal file format (can't be read)
	format version 4.8 required; 0.0 given

PID 28125: Failure! too large packet 67651874
PID 28144: Succeded.
PID 28147: Succeded.
PID 28192: Succeded.
PID 28176: Succeded.
PID 28167: Succeded.
PID 28186: Succeded.
PID 28152: Succeded.
PID 28150: Succeded.
PID 28131: Succeded.
PID 28170: Succeded.
PID 28137: Succeded.
PID 28173: Succeded.
PID 28164: Succeded.
PID 28179: Succeded.
PID 28182: Succeded.
PID 28139: Succeded.
PID 28158: Succeded.