Re: beacon 0.9.58 dies after a while - broken pipe?

[Mitch Kutzko <mitch@ncsa.uiuc.edu>]

Hi,  John -- Added this change (and others) into today's release.

	http://dast.nlanr.net/projects/beacon

It's version 0.9.58-4.

Let me know it works for you -- I've received one report already saying
there are still problems with it.

Mitch

At 06:47 PM 7/31/2004 -0500, you wrote:
> On Thu, 29 Jul 2004 10:04:42 -0500
> John Kristoff <jtk@northwestern.edu> wrote:
> 
> > I also have seen this, but haven't been able to pinpoint it.  I was
> [...]
> 
> I believe I've narrowed things down quite a bit.  I ran the script
> with truss on my system and noticed that eventually I kept getting an
> EPIPE error when writing the report to the TCP pipe back to the central
> beacon.ncsa.uiuc.edu server.  This causes one of the 'print' statements
> to fail and the script to eventually stop without warning or notice.  I
> have not been able to determine why the TCP connection goes away in
> the middle of writing to the pipe, but I suspect it may be a problem
> on the server's end, since there are multiple people experiencing this
> behavior.
> 
> Mitch, if the problem is on the server end we would need you or a
> beacon server admin to debug why TCP connections are being abruptly
> lost.  Is there anything you can see wrong?
> 
> In the meantime, in order to keep my client working I am basically
> ignoring the problem.  Probably not the best thing to do, but I
> suppose for those experiencing this as well, it may be an acceptable
> option short of some kind of external watcher script or daemon tool.
> Here is a diff to the 0.9.58 RC1 -2 src/beacon.in script:
> 
>   --- beacon.in
>   +++ beacon.in.new
>   @@ -2370,7 +2370,12 @@
>           }
>           $connection_to_tcp_server = 1;
>      }
>   -
>   +
>   +### JTK: hack to avoid unexpected TCP close during writes
>   +use POSIX qw(:errno_h);
>   +$SIG{PIPE} = 'IGNORE';
>   +### JTK
>   +
>      # FIRST LINE - Authentication string: The Central Server will ignore
any
>      # TCP traffic to this port that doesn't start with the appropriate
line:
>      # "beacon.dast.nlanr.net:233.4.200.22:10006" (Or your CS Name,
group, port)
>   @@ -2555,7 +2560,10 @@
>      }
> 
>      # and terminate the connection when we're done
>   -  close($socket) || die ("Couldn't close TCP reporting socket" );
>   +### JTK: hack to avoid lost socket during unexpected TCP close
>   +#  close($socket) || die ("Couldn't close TCP reporting socket" );
>   +  close($socket) || print STDERR "Warning, couldn't close socket
$socket: $!\n";
>   +### JTK
>     
>      return;
> 
> John
> 
> 
--
Mitch Kutzko | mitch@dast.nlanr.net | mitch@ncsa.uiuc.edu | 217-333-1199
http://hobbes.ncsa.uiuc.edu/