TCP Socket: Connection reset, but Select says it's valid

[Scott Cole]

I'm running a server that sometimes encounters a
connection-reset-by-peer error when reading data from a client. I'm
calling IO.Select to verify that there are characters to be read, but
occasionally the "rescue" below is activated when I do my gets. I
presume this means the socket was closed after Select noticed there were
characters to be read but before I actually try to read them.

The client and server are in the same machine, so there's no cable to
pull out.

The messages of interest are sent every 30 seconds, and there might be 1
or 2 failures in a day. Running on Mac OS X 10.5.6.

The problem seems to occur when the client (below) connects to send a
message and another client connects before the first client's
transaction is complete. This transaction takes about 1 second. The
specified timeout is 120 seconds.

Is there anything obvious wrong with my code? I omitted quite a few
lines, I hope not too much.

Thanks-
Scott

######################### Server (client follows)
#########################

require 'gserver'

class ChatServer < GServer # Server class derived from GServer super
class
def initialize(*args)
super(*args)
# Keep a record of the client IDs allocated
# and the lines of chat
@@client_id = 0
@@chat = []
end

def serve(io) # Serve method handles connections
# Increment the client ID so each client gets a unique ID
@@client_id += 1
my_client_id = @@client_id
my_position = @@chat.size


loop do
#Every n seconds check for data
n = 0.5

selection = IO.select([io], nil, [io], n)
# If some event occurred, retrieve the data and process it...
if selection && selection[0] then
# There was a read event
begin
line = io.gets
rescue Exception => e
stat = ''
puts "\nerror reading 'line' from #{io}"
puts "#{ e } (#{ e.class })!"
# Close socket
io.close
print("\nSelection is #{selection[0]}\n")
end
if selection[2].size > 0 then
puts "%%%%%%%% Select error array is #{selection[2]}" +
Time.now.to_s
end

 

[Brian Candler]

I'm running a server that sometimes encounters a
connection-reset-by-peer error when reading data from a client. I'm
calling IO.Select to verify that there are characters to be read, but
occasionally the "rescue" below is activated when I do my gets.

Your code prints the exception class and message, what do you see?
I

presume this means the socket was closed after Select noticed there were
characters to be read but before I actually try to read them.

select returns true for an EOF-condition too. From `man 2 select`:

Three independent sets of file descriptors are watched. Those
listed
in readfds will be watched to see if characters become
available for
reading (more precisely, to see if a read will not block; in
particuâ€
lar, a file descriptor is also ready on end-of-file)

 

[Scott Cole]

Thanks for responding. I would expect to get some kind of EOF error on
an EOF condition. Instead I see something like

error reading 'line' from #<TCPSocket:0x40e3f4>
Connection reset by peer (Errno::ECONNRESET)!

Selection is #<TCPSocket:0x40e3f4>

 

[Brian Candler]

Thanks for responding. I would expect to get some kind of EOF error on
an EOF condition. Instead I see something like

error reading 'line' from #<TCPSocket:0x40e3f4>
Connection reset by peer (Errno::ECONNRESET)!

Yep, it's not a simple EOF - the far end has sent a TCP RST instead of a
FIN - but AFAIK the socket is marked 'readable' for select.

More info at:
http://unix.derkeiler.com/Newsgroups/comp.unix.programmer/2007-08/msg00176.html

Note also that this is does not mean that the socket is marked
selectable for 'error' - selection[2]. In fact, you're almost certainly
never going to find the socket marked that way. See
http://unix.derkeiler.com/Newsgroups/comp.unix.programmer/2007-11/msg00187.html