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

Re: back-bdb core dump (ITS#3358)

rhafer@suse.de wrote:

>On Friday 08 October 2004 13:08, rhafer@suse.de wrote:
>>Well I was now able to get it crashing with "cachesize 1" as well. I just
>>used a bit different test case. I have now two clients, one is just adding
>>a 10000 entry LDIF and the other is adding an removing 10 entry in a loop.
>>After a few interation of the loop the server crashed. (BTW it seems that
>>on SMP machines the crash is triggered faster)
>What sees to happen is that another thread calls bdb_cache_lru_add, which 
>cleans out the EntryInfo structure, before bdb_cache_delete_cleanup is 
>called. I was however not yet able to come up with a reasonable fix for that.
Hm... I guess that makes sense, since the entry lock was freed when the 
transaction got committed. I've committed a proposed fix to CVS HEAD, 
but since I never reproduced the original problem I cannot verify the 
fix. Please test...

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support