Re: (ITS#4216) slapd terminates unexpectedly

[hyc@symas.com]

James MacNider wrote:
> We used 10 clients and 10 threads per client spread across 2 test driver
> machines (so a total of 50 threads on each machine).  The time it takes
> for the problem to occur varies...it has happened as quickly as 10
> seconds or as long as 20 minutes. 
>   
Just as an aside, in our experience using more than two threads per 
client doesn't actually produce any more useful server load.

By the way, is syslog active on slapd while running this test? Is slapd 
configured with reverse-lookups enabled? The point is to identify where 
any non-socket descriptors are coming from, if any. If you can run slapd 
under a debugger with a breakpoint at the point where the error is 
detected (daemon.c line 1862 would be a good choice), and then run lsof 
on the slapd process when the breakpoint is triggered, that may give 
some useful information. Also print slap_daemon, readfds, and writefds 
at that point.

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