On 2005-01-10, Chris Stromsoe wrote:

> I'm running bind 9.2.4 and am regularly seeing it use > 1.5Gb ram.
> I'm loading ~ 1050 zones and an RBL zone. The zones on disk are
> 132680Mb (measured with du). rndc dumpdb produces a 40Mb dump file.


> Immediately after startup completes, bind is using slightly less than 500Mb.
> max-cache-size is set to 250M. The machine has 1Gb of physical memory + 2Gb
> swap.
>
> I would expect memory usage to stabilize somewhere around 750Mb, but it
> doesn't. After a few weeks of running, the amount of swapping becomes
> noticeable and I have to kill and restart named. Are there issues with
> max-cache-size in 9.2.4? Or am I missing something obvious?


I've seen this problem too (also with max-cache-size set). See some
comments in my recent thread, but just an aside -- I was really
skeptical of rbldnsd, especially because it's loosely based on djb-ware,
but it seems to do a really good job of serving dnsbl zones without
eating up tons of memory (while still maintaining good performance).
It's super easy to setup.

Increasing the memory and swap on the machine is one way to at least
sort of help this problem; that's what I've done... I'd be interested in
hearing better solutions, though.

Side note (w/r/t killing / restarting named) - I know it's not the right
thing to do, but I almost always end up having to kill -9 named - even
rndc halt doesn't seem to do the trick - rndc stop or rndc halt stops
named from running but the process doesn't seem to die even if I wait
quite some time.