Re: slapd-ldap killed by a broken client (ITS#4117)

[Kurt@OpenLDAP.org]

At 08:11 AM 11/4/2005, raphael.ouazana@linagora.com wrote:
>On Ven 4 novembre 2005 14:35, Pierangelo Masarati wrote:
>>
>>> This problems appears to be the result of a user error.
>>> You need to tell your debugger to pass this and most
>>> other signals to slapd(8) so that slapd(8) can properly
>>> handle them.  In the case of PIPE, slapd(8) simply
>>> ignores the signal.
>>
>> If breaking the connection kills slapd __without__ gdb in the middle, as I
>> understand the ITS started from, then it's likely a bug; in this case, a
>> stack backtrace might be more appropriate to avoid issues with gdb signal
>> handling.  Please confirm, otherwise the issue will be closed.
>
>I confirm. I had this bug many times without gdb, and then I tried to
>reproduce it under gdb.
>The last trace I submited was done typing 'continue' in gdb after the
>SIGPIPE. Is it different than to tell gdb to pass the signal to slapd ?

Before running slapd in the gdb, you should tell the
gdb to pass all (excepting those gdb uses) to the
program without stopping.  That is,
        handle all pass nostop
You might also want to avoid printing certain signals,
that is,
        handle SIGHUP noprint
        handle SIGPIPE noprint