MOO-cows Mailing List Archive

[Prev][Next][Index][Thread]

BUG: Problems with 1.8.0p5 and fast multiple connect requests.



Hi guys,

here's an interesting one.  Jeni has been playing with serving web pages from 
her own personal MOO at our site.  She has found that, when she's using frames 
to get 3 pages in on one screen, she gets wierd results.  Sometimes she gets 
them all fine.  Sometimes she gets none of them, mostly 1 or 2 don't ever get 
sent.  The log file for this period of connects goes as follows.  Accepts 
obviously mean that it has allowed the connect.  A Disconnect for it then 
follows in course.  The *** error means it failed to connect and GET the text.  
We've included more of the working ones to show that the order and speed of 
requests doesn't seem to be the only factor.  jenimac is her Mac, darkhorse is 
my Sun.

Note we are having no problems with single web pages or player connects.  It 
*may* also be related to requests in quick succession from the same client, 
therefore.

Jul 23 13:15:33: ACCEPT: #-6679 on port 5556 from jenimac, port 1507
Jul 23 13:15:34: ***  An event requires
attention
Jul 23 13:15:35: *** Accepting new network connection: An event requires
attention
Jul 23 13:15:35: ACCEPT: #-6680 on port 5556 from jenimac, port 1510
Jul 23 13:15:36: DISCONNECTED: #-6679 on port 5556 from jenimac, port 1507
Jul 23 13:15:38: DISCONNECTED: #-6680 on port 5556 from jenimac, port 1510
Jul 23 13:15:40: ACCEPT: #-6681 on port 5556 from darkhorse.CV.COM, port 36180
Jul 23 13:15:42: DISCONNECTED: #-6681 on port 5556 from darkhorse.CV.COM, port
36180
Jul 23 13:15:50: ACCEPT: #-6682 on port 5556 from darkhorse.CV.COM, port 36181
Jul 23 13:15:51: DISCONNECTED: #-6682 on port 5556 from darkhorse.CV.COM, port
36181
Jul 23 13:15:53: ACCEPT: #-6683 on port 5556 from darkhorse.CV.COM, port 36182
Jul 23 13:15:54: DISCONNECTED: #-6683 on port 5556 from darkhorse.CV.COM, port
36182
Jul 23 13:15:58: ACCEPT: #-6684 on port 5556 from darkhorse.CV.COM, port 36183
Jul 23 13:15:59: DISCONNECTED: #-6684 on port 5556 from darkhorse.CV.COM, port
36183
Jul 23 13:16:24: ACCEPT: #-6685 on port 5556 from darkhorse.CV.COM, port 36184
Jul 23 13:16:27: DISCONNECTED: #-6685 on port 5556 from darkhorse.CV.COM, port
36184
Jul 23 13:16:28: ACCEPT: #-6686 on port 5556 from darkhorse.CV.COM, port 36185
Jul 23 13:16:28: ACCEPT: #-6687 on port 5556 from darkhorse.CV.COM, port 36186
Jul 23 13:16:28: ACCEPT: #-6688 on port 5556 from darkhorse.CV.COM, port 36187
Jul 23 13:16:29: DISCONNECTED: #-6687 on port 5556 from darkhorse.CV.COM, port
36186
Jul 23 13:16:30: DISCONNECTED: #-6686 on port 5556 from darkhorse.CV.COM, port
36185
Jul 23 13:16:30: DISCONNECTED: #-6688 on port 5556 from darkhorse.CV.COM, port
36187

Now, "Accepting new network connection:" is from net_sysv_tcp.c, 
proto_accept_connection().  It means that either t_listen() or t_accept(), both 
system calls, have returned an error.  The rest of the error we are assuming is 
from the following t_errno value:

     TLOOK               An asynchronous event  has  occurred  on
                         this  transport  endpoint  and  requires
                         immediate attention.
                         
though since the MOO ignores the value set here, we have no way of finding out, 
other than starting to hack the code to find it.  Since this is a timing issue, 
we believe, that may be counter-productive.

What we *think* might be happening is shown in the paste below.  If so, I think 
the best solution would be to try the t_listen/t_accept again?

>From the t_accept man page:

     For both types of endpoints, t_accept() will  fail  and  set
     t_errno  to  TLOOK  if there are indications (for example, a
     connect or disconnect) waiting to be received on  that  end-
     point.

Either way, this is now somewhat over my head, and so we figured you'd want to 
see it.  Any ideas?  If we *have* traced this to the correct point, is there any 
chance of patching this so that if TLOOK is set, it retries the t_accept()?  
This is proving quite an irritation for us.

Many thanks for reading this far down a long and boring post,

Moredhel/TC.  MOOtiny. (spsyc.nott.ac.uk/128.243.31.4 8888)



Home | Subject Index | Thread Index