[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: (ITS#7025) "the backglue code doesn't install a handler for the Abandon operation"

hans.moser@ofd-z.niedersachsen.de wrote:
> Howard,
> Howard Chu schrieb (17.11.2011 23:54 Uhr):
>> Marc Patermann wrote:
>>> Howard Chu schrieb (01.11.2011 18:39 Uhr):
>>>> Also, this script just performs sync searches, what do you use to
>>>> generate the writes?
>>> you mean the write which kills the server?
>>> It just one simple mod of an object which I made in Apache Directory
>>> Studio.
>> I'm unable to reproduce this crash. I have your slapd.conf and test
>> data, I have run 80-some instances of your script as you've described.
> I ran 80 instances of each of the three script, where each script
> replicates one database, so 240 instances of the script in total.

OK. After starting 240 instances and performing the modify, I saw a SEGV due 
to syncprov. I've patched this in git master, please test. It runs OK for me now.
>> While the initial sync search is running I perform a modify of the base
>> entry of the subordinate database. Everything continues to run. I've
>> also waited for all of the initial refresh traffic to finish first, and
>> then perform the modify, and still no crash.
>> To be more clear - do you perform your modify while the sync searches
>> are still refreshing, or after all of the refreshes have completed?
> IIRC it does not make any difference, but I definitly crashed the server
> while at least one database still replicated.
>> Do you run the clients on the same host as slapd, or on a remote machine?
> I had slapd one one machine and the script on another one.
> Marc

   -- Howard Chu
   CTO, Symas Corp.           http://www.symas.com
   Director, Highland Sun     http://highlandsun.com/hyc/
   Chief Architect, OpenLDAP  http://www.openldap.org/project/