Re: (ITS#6405) test043-delta-syncrepl failed - producer and consumer databases differ

[hyc@symas.com]

Michael Ströder wrote:
> Howard Chu wrote:
>> Howard Chu wrote:
>>> Michael Ströder wrote:
>>>> Howard Chu wrote:
>>>>> michael@stroeder.com wrote:
>>>>>> Download content of testrun/ here:
>>>>>>
>>>>>> http://www.stroeder.com/temp/openldap/its6405-test043-testrun.tar.gz
>>>>>
>>>>> Shows that a change was replicated to the consumer and ignored on the
>>>>> consumer. But since there's no SYNC logging enabled, we don't see the
>>>>> reason why the consumer did this...
>>>>
>>>> It does not happen very often. This time I had to run it almost 50
>>>> times.
>>>>
>>>> Hope I got the sync log level right herein:
>>>>
>>>> http://www.stroeder.com/temp/openldap/its6405-test043-testrun-sync-log.tar.bz2
>>>>
>>>
>>> I think you have an OS / clock problem. The skipped op in this case got
>>> assigned an entryCSN older than the immediately preceding op. (The
>>> missing op
>>> has CSN 20091130201002.575384Z. The preceding op was
>>> 20091130201002.577244Z)
>>> This test doesn't run any concurrent operations, each new op is only
>>> submitted
>>> after the previous one completes, so this is not a thread concurrency
>>> issue,
>>> your system clock is going backwards.
>>>
>> A patch for this is now in libldap/util-int.c in HEAD.
>
> Hmm, this is a virtual machine running in vmware-player 64 bit. But I didn't
> suspect to have a clock issue therein since the host is a rather fast
> dual-core notebook. I will investigate this.

The one thing you can rely on as far as clocks in VMware is that they will be 
wrong, regardless of your CPU speed.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1236

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