memget errors in bindlog

[Mark.Andrews at nominum.com]

	Named will stabalise its memory useage after ~7 days, max_cache_ttl
	dominates this and is a compile time parameter.  This can be brought
	down to a few hours without much problems.  max_ncache_ttl should be
	<= max_cache_ttl, the cleaning-interval should also be lowered to ~30
	mins if you do this.

	There are some options which can also effect memory usage.
	"cleaning-interval 0;" will disable garbage collection.
	"host-statistics yes;" will allocate memory per nameserver
		talked to over the life of the process.

	Mark

> This is a snapshot of the resources on the server in question:
> 
> bash-2.03$ ulimit -a
> core file size (blocks)     unlimited
> data seg size (kbytes)      unlimited
> file size (blocks)          unlimited
> open files                  64
> pipe size (512 bytes)       10
> stack size (kbytes)         8192
> cpu time (seconds)          unlimited
> max user processes          15941
> virtual memory (kbytes)     unlimited
> 
> 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> bash-2.03$ df -k
> Filesystem            kbytes    used   avail capacity  Mounted on
> /proc                      0       0       0     0%    /proc
> /dev/dsk/c0t0d0s0     192423   33479  139702    20%    /
> /dev/dsk/c0t0d0s3    2405630 1149486 1208032    49%    /usr
> fd                         0       0       0     0%    /dev/fd
> /dev/dsk/c0t0d0s7    1984230   83500 1841204     5%    /var
> /dev/dsk/c0t0d0s6     480815  316973  115761    74%    /etc/namedb
> /dev/dsk/c0t0d0s4    1156662   19228 1079601     2%    /export/home
> /dev/dsk/c0t0d0s5    1389006    5928 1327518     1%    /opt
> swap                 1728736       8 1728728     1%    /tmp
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> bash-2.03$   ps -eo pid,rss,vsz,comm |sort +1nr |head
> 20155 148264 149824 /usr/local/sbin/named
>   216 2040 2496 /usr/sbin/vold
>   157 1952 2416 /usr/lib/autofs/automountd
>   185 1928 2480 /usr/sbin/nscd
>   235 1888 2864 /usr/lib/dmi/dmispd
>  9717 1824 2384 -bash
>   168 1808 3272 /usr/sbin/syslogd
> 23862 1616 2264 /usr/local/sbin/named-xfer
> 23863 1616 2264 /usr/local/sbin/named-xfer
>  9714 1536 1968 /usr/local/sbin/sshd
> 
> Anything look wrong with it?
> 
> 
> Mark.Andrews at nominum.com wrote:
> 
> > > We have a machine running bind 8.2.2p5 on solaris 7.  Although this
> > > machine is specifically used only for named, it is a primary for the
> > > company and gets slammed with queries hard.  It is an Ultra10 440 with a
> > >
> > > gig of RAM.  The problem is this:
> > > If I do not set the datasize for the cache, named will run to a certain
> > > point of memory (not exactly sure of how much, still trying to capture
> > > that data) size and just die.  It does not flush its own cache, (first
> > > in, first out) it just dies.
> >
> >         See limit/ulimit for what the per process memory limit is set to.
> >         "options { datasize #; };" just sets that limit.
> >
> >         Since you have 1G of memory why don't you just set the per process
> >         limit to 900M.  Also make sure you have enough swap to cope when na
> med
> >         forks to invoke named-xfer for incoming zone transfers.
> >
> >         Mark
> >
> > > This is the error in the logs that we get:
> > >
> > > 22-Feb-2000 23:50:47.995 default: critical: savedata: memget
> > > 22-Feb-2000 23:50:47.995 panic: critical: savedata: memget
> > >
> > > 23-Feb-2000 16:44:29.201 default: critical: savedata: memget
> > > 23-Feb-2000 16:44:29.201 panic: critical: savedata: memget
> > >
> > > 27-Feb-2000 01:39:57.757 default: critical: savedata: memget
> > > 27-Feb-2000 01:39:57.757 panic: critical: savedata: memget
> > >
> > > If I set the datasize in the named.conf file, it will reach that size
> > > and die, but I have a cron job in place that checks if named is running
> > > every 5 minutes and restarts it if found to not be running.
> > >
> > > While that works, I would like to be able to actually heal this wound
> > > and not keep a bandaid on it.  Or at the very least, understand why it
> > > does this.
> > >
> > > Any help would be appreciated.
> > >
> > > D-.
> > --
> > Mark Andrews, Nominum Inc. / Internet Software Consortium
> > 1 Seymour St., Dundas Valley, NSW 2117, Australia
> > PHONE: +61 2 9871 4742                 INTERNET: Mark.Andrews at nominum.com
> 
> 
> 
--
Mark Andrews, Nominum Inc. / Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: Mark.Andrews at nominum.com