Disable cache in bind 9.6

[Dmitry Rybin]

В Пнд, 26/01/2009 в 16:16 -0800, JINMEI Tatuya / 神明達哉 пишет:
> http://www.jinmei.org/patch/bind9-lrucache.diff
> (should be cleanly applicable to 9.6).
> and let me know if it mitigates the problem?
> 

Oh, great work. I'll try tomorrow.

> Other recommendations:
> - I previously suggested using a separate cache-only view and forward
>   all recursive queries to that view.  Have you tried that?  If you
>   have, didn't it work as I hoped?

Yes, I try it. But I can't set ttl to 0. It didn't work. Recursive query
fails, and authoritative query back to clients with ttl 0  :(

I increase memory on servers 2x QUAD CORE XEON up to 12Gb.
  PID USERNAME       THR PRI NICE   SIZE    RES STATE  C   TIME   WCPU
COMMAND
38634 bind            11   4    0  3003M  2952M RUN    2 159:28 46.44%
named

~50 views, 
max-cache-size for most views 64M;
bind uptime (after kernel: pid 667 (named), uid 53: exited on signal 11)
- 2 days and 6 hours.


 built with '--localstatedir=/var' '--disable-linux-caps'
'--with-randomdev=/dev/random' '--d
isable-openssl-version-check' '--without-openssl'
'--with-libxml2=/usr/local' '--without-idn' '--enable-largefile'
'--enable-threads' '--prefix=/usr/local' '
--mandir=/usr/local/man' '--infodir=/usr/local/info/'
'--build=x86_64-portbld-freebsd7.1'
'build_alias=x86_64-portbld-freebsd7.1' 'CC=cc' 'CFLAGS=-O2 -fno-st
rict-aliasing -pipe' 'LDFLAGS= -rpath=/usr/lib:/usr/local/lib' 'CXX=c++'
'CXXFLAGS=-O2 -fno-strict-aliasing -pipe'

rndc status:
version: 9.6.0-P1
CPUs found: 8
worker threads: 8

On another server in same configuration bind works 2 days and die
without core  kernel: pid 682 (named), uid 53: exited on signal 11

Max memory per process - 12GB. May be FreeBSD x64 can't work more then X
Gb per process?
# cat /boot/loader.conf 
kern.maxdsiz="17179869184"       # 16gb
kern.dfldsiz="17179869184"       # 16gb
kern.maxssiz="134217728"        # 128MB


> - BIND 9.7 will have a new option "attach-cache" exactly for such an
>   extraordinary operational environment as yours: it allows multiple
>   views to share a single cache to save memory.

I'll try to test 9.7 on one of the heavy load servers and post results
to you.