dnsperf and BIND memory consumption

Vinny Abello vinny at tellurian.com
Thu Jan 1 05:47:10 UTC 2009


> -----Original Message-----
> From: Danny Mayer [mailto:mayer at gis.net]
> Sent: Thursday, January 01, 2009 12:05 AM
> To: Jinmei_Tatuya at isc.org
> Cc: Vinny Abello; dougb at FreeBSD.org; bind-users at isc.org
> Subject: Re: dnsperf and BIND memory consumption
> 
> JINMEI Tatuya / $B?@L at C#:H wrote:
> > At Tue, 9 Dec 2008 15:26:25 -0500,
> > Vinny Abello <vinny at tellurian.com> wrote:
> >
> >> Has anybody else tried this patch for you? I haven't had time to
> >> look into this at all. If nobody has tried this yet, I'll get around
> >> to it when I can and let you know the result.
> >
> > No one else other than by myself.  It worked perfectly for me, i.e.,
> I
> > could reproduce the problem and I could completely eliminate the leak
> > with the patch.  One thing I was not certain about in an off-list
> > discussion that led to this patch was that the patch reportedly
> solved
> > the leak only partially.  I've been hoping to confirm that, but
> > unfortunately I've not got any followup since then.
> >
> > So, basically, I believe the problem was solved, it would also help
> if
> > you could confirm it.
> 
> Personally, I'm not convinced that it will make a difference outside of
> Windows. The fix is to make sure a lock gets destroyed when done and
> the
> function exits. On Windows the lock gets created and memory is
> allocated
> for it outside of the function using it and needs to be explicitly
> destroyed, but my understanding of pthreads was that this worked
> differently on Unix and the lock structure would be automatically
> destroyed when the function is exited since the lock was local to the
> function.
> 
> I'd be interested in seeing the results of this.
>  w
> Danny

I just loaded up the BIND 9.5.1 port on FreeBSD 7.0 AMD64 with threads. I don't see the prominent memory leak present on my system any longer. I lost track of this thread and think two different changes might have been made, however. One via the port and one in BIND itself, so I don't know which fix is actually in effect here, but the results seem good so far. Maybe Doug and Jinmei and comment on their respective changes just for the record.

-Vinny



More information about the bind-users mailing list