BIND - out of memory

[Jan Arild Lindstrøm]

Hi,

would be great if someone could explain why this happens and how we can correct
it.

Regards
Jan Arild Lindstrøm



At 09:13 26/03/2009, Jan Arild Lindstrøm wrote:
>Hi,
>
>I allready tried 9.4.3, and it happened there.
>
>Trying 9.6.0-P1 gives the same result:
>
>        recursive clients: 1083/49900/50000
>        
>        --cut--
>        26-Mar-2009 08:04:39.736 database: adb: fetch of 'mars.csd.unb.ca' A failed: out of memory
>        26-Mar-2009 08:04:39.737 database: adb: fetch of 'dns.guangzhou.gd.cn' A failed: out of memory
>        --cut--
>
>        nsXX(root) / 502# plimit 9474
>        9474:   /local/named-new/local/sbin/named -f -c /etc/named/named.conf.new -u n
>           resource              current         maximum
>          time(seconds)         unlimited       unlimited
>          file(blocks)          unlimited       unlimited
>          data(kbytes)          unlimited       unlimited
>          stack(kbytes)         unlimited       unlimited
>          coredump(blocks)      unlimited       unlimited
>          nofiles(descriptors)  unlimited       unlimited
>          vmemory(kbytes)       unlimited       unlimited
>
>        tcp-clients 5000;
>        clients-per-query 2500;
>        max-clients-per-query 5000;
>        recursive-clients 50000;
>
>        (Big numbers to be sure they are not the reason for the 1000 limit.)
>
>9.6.1b1 info:
>        file local/named-new/local/sbin/named:       
>        ELF 64-bit MSB executable SPARCV9 Version 1, UltraSPARC3 Extensions Required, dynamically linked, not stripped
>
>        Sun Studio Express 3/09 (-xtarget=ultraT2plus -m64):
>        nsXX(root) named-new 543# /local/named-new/local/sbin/named -V
>        BIND 9.6.1b1 built with '--prefix=/local' '--localstatedir=/var' '--with-openssl=/local/openssl' '--with-randomdev=/dev/urandom'        
>        '--enable-threads' '--with-libtool' '--enable-static=yes' '--disable-shared' '--sysconfdir=/etc/named' 'CC=/opt/StudioExpress/SSX0903/bin/cc' 
>        'CFLAGS= -xtarget=ultraT2plus -m64' 'LDFLAGS= -xtarget=ultraT2plus -m64' 'CPPFLAGS= -xtarget=ultraT2plus -m64' 
>        'CXX=/opt/StudioExpress/SSX0903/bin/CC' 'CXXFLAGS= -xtarget=ultraT2plus -m64'
>
>        SunOS nsXX.xxx.xx 5.10 Generic_138888-01 sun4v sparc SUNW,T5140 Solaris
>
>BIND 9.4.3
>        Sun Studio 12: -fast -xtarget=ultraT1 -m64
>BIND 9.6.0-P1:
>        Sun Studio Express 11/08: -fast -xtarget=native64
>
>I tried it on another server also, and same thing happens: As soon as recursive clients pass
>1000/default, "out of memory" messages start to flood the log.
>
>Anyway:
>        During peak of the day our prod servers are around 700 recursive clients, so we are not 
>        affected by the 1000/default "limit" yet. But if we flush the cache we will have a problem 
>        (as always), since then suddenly there is +8000 recursive clients there within one second, 
>        and the number just grows until 50000, and the server is jammed. Hence we try very hard 
>        to not flush the whole cache or restart BIND. I have asked before about that problem, but
>        no solution have been found. But perhaps now, the 1000/default "limit" and out-of-memory
>        might be the/one of the reasons for this.
>
>
>Regards
>Jan Arild Lindstrom
>
>
>At 22:41 25/03/2009, Doug Barton wrote:
>>Jan Arild Lindstrøm wrote:
>>> Hi,
>>> 
>>> more findings ...
>>> 
>>> BIND 9.6.1b1
>>> 
>>> No matter what I set in named.conf, it starts to give "out of memory" when recursive
>>> clients pass 1000. I see that 1000 is the default value for recursive-clients.
>>
>>Did you try backing up to 9.6.0-P1 to see if the same behavior exists
>>there?
>>
>>
>>Doug
>
>_______________________________________________
>bind-users mailing list
>bind-users at lists.isc.org
>https://lists.isc.org/mailman/listinfo/bind-users