[Xorp-hackers] Assert in recent cvs

[Ben Greear]

Pavlin Radoslavov wrote:
> Ben Greear <greearb at candelatech.com> wrote:
> 
>> We're still trying to figure out if this is reproducible, but
>> here's the stack trace.  This is with all of my patches applied,
>> so it's possible something we are doing wrong.
> 
> Please send instructions how to reproduce the problem: rtrmgr
> command line options, configuration file and any actions you did.
> I need the command line options so I know where the log output is
> redirected to.
> 
> Also, please reverse-back the rev. 1.68 change to the
> rtrmgr/module_manager.cc file:
> 1.68      +5 -3;  commitid: 11fdf48f7e12041a7; xorp/rtrmgr/module_manager.cc
> 
> In particular, change back the two XLOG_*() messages to fprintf(),
> and see whether you still get the error.
> This will guide us choosing the next step in debugging the problem.

Ok, here is how you reproduce the problem:

I have a program called 'logchopper' that reads from stdin
and writes to log files, rotating them when they are large.
This way I can 'log' stdout and stderr by piping it into
logchopper.

When I shutdown my xorp instances, I do a 'killall -r xorp', which
sometimes kills the logchoppers before the parent xorp.  This causes
a sigpipe, I guess, and then xorp_rtrmgr dumps core.

I think you could use 'tee' in place of my logchopper and
have the same result.  If you want, I can send you the logchopper
source.

I'm going to try just commenting out the assert in the sig-pipe
handler.  In my case, xorp will be killed shortly anyway, and even if
not, I'd rather xorp run w/out stdout/stderr (or maybe exit cleanly)
than crash.

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com