One more comment - ad process size, I did measure the process sizes via
'top', and the excessive memory was really and without a doubt allocated
by named. While the machine has only 2GB of RAM, top reported named has
allocated much more than that, swap was in use and free swap was
steadily diminishing.
The recommend I killed the named process, all the swap was suddenly
free, and top reported large amount of Free memory that was not there
the moment ago.
Look at this graph:
http://www.mujweb.cz/nakamura/dns/leak.png
The cliffs after the peaks correspond with manual restart of bind.
This is how the same machine memory allocation looks after downgrading
to 9.9.6 during the Sunday evening:
http://www.mujweb.cz/nakamura/dns/noleak.png
Note the orange parts of the graph representing the Free memory in
FreeBSD memory management system - before, the Free memory was quickly
consumed after each named restart, but the situation is radically
different now. The server still has a steady pool of Free memory, that
means memory that was never used since the restart of the named, since
FreeBSD marks most of the deallocated memory as Inactive and cleans it
just moments before it's needed again.
That clearly means that the overall memory requirements are dramatically
lower with 9.9.6, and also that these requirements do not grow with
time, but remain stable on a resolver with the same volume and structure
of queries, which is what one would expect.
I am sorry I cannot provide the xml statistics you asked for, but the
data I provided in the leakinfo.tgz archive should be sufficient to
prove there is indeed a memory leak.
--
Best Regards,
Daniel Ryšlink
System Administrator
Dial Telecom a. s.
Křižíkova 36a/237
186 00 Praha 3, Česká Republika
Tel.:+420.226204627
daniel.rysl...@dialtelecom.cz
-----------------------------------------------
www.dialtelecom.cz
Dial Telecom, a.s.
Jednoduše se připojte
-----------------------------------------------
On 01/27/2015 06:46 AM, Mukund Sivaraman wrote:
Hi Daniel
On Mon, Jan 26, 2015 at 02:56:44PM +0100, Daniel Ryšlink wrote:
Downgraded to BIND 9.9.6, the leak is gone, using the same named.conf,
same HW, same environment.
It is highly likely there is really a memory leak problem in Bind
9.10.
Because many of these reports are on FreeBSD 10 in a resolver
configuration, we are checking to see if we are able to reproduce it
there.
Meanwhile, please can you enable statistics-channels in named.conf and
send us a dump of the XML statistics along with process sizes reported
by ps when named grows very large?
Mukund
_______________________________________________
Please visit https://lists.isc.org/mailman/listinfo/bind-users to unsubscribe
from this list
bind-users mailing list
bind-users@lists.isc.org
https://lists.isc.org/mailman/listinfo/bind-users
