Bind uses 100% CPU when internet is down

I'm running BIND 8.4.6 on a Ubuntu "Dapper" system, for caching internet DNS
requests as well as providing DNS for the local network.

Internet access is via dialup. What is occuring is that when the internet
is down, the named daemon after a while (maybe an hour or two) will start
chewing up nearly 100% of the CPU. This continues until it is killed off
or the internet is dialed up. Once the internet is connected, the named
daemon drops to normal CPU utilization levels within seconds. There are
no named diagnostics to be found in /var/log/messages.

I've done some searching with Google and found a lot of references
to CPU utilization problems in named, but have not found anything yet
relating it to whether or not the internet is connected. Is this a bug? A
configuration problem? Will updating to bind 9 be of any help?

--
Roger Blake
(Subtract 10 for email.)

Roger Blake <(E-Mail Removed)> wrote:
> I'm running BIND 8.4.6 on a Ubuntu "Dapper" system, for caching
> ...
> internet is down, the named daemon after a while (maybe an hour or
> two) will start chewing up nearly 100% of the CPU. This continues
> ...
> I've done some searching with Google and found a lot of references
> to CPU utilization problems in named, but have not found anything
> yet relating it to whether or not the internet is connected. Is this
> a bug? A configuration problem? Will updating to bind 9 be of any
> help?

I thought the ISC folks wanted everyone migrating to BIND 9 ages ago
so it might be good to do just on first principles...

WRT the other bit - have you tried to profile the 8.4.6 named when it
gets into this situation? Say starting with an strace to see which
system calls (if any) it is making, and then with something to show
which routines are being excercised within the named binary?

rick jones
--
denial, anger, bargaining, depression, acceptance, rebirth...
where do you want to be today?
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

In article <s0Mwh.44$(E-Mail Removed)>, Rick Jones wrote:
> I thought the ISC folks wanted everyone migrating to BIND 9 ages ago
> so it might be good to do just on first principles...

That's probably something that should be done. Didn't think it would
make much difference on a nameserver for a small LAN, but you never
know.

> WRT the other bit - have you tried to profile the 8.4.6 named when it
> gets into this situation? Say starting with an strace to see which

That's a good idea. Though I may just try the BIND 9 update first
and do the profiling if the problem persists.

--
Roger Blake
(Subtract 10 for email.)

On 2007-02-02, Roger Blake <(E-Mail Removed)> wrote:

> I'm running BIND 8.4.6 on a Ubuntu "Dapper" system, for caching internet DNS
> requests as well as providing DNS for the local network.
>
> Internet access is via dialup. What is occuring is that when the internet
> is down, the named daemon after a while (maybe an hour or two) will start
> chewing up nearly 100% of the CPU. This continues until it is killed off
> or the internet is dialed up. Once the internet is connected, the named
> daemon drops to normal CPU utilization levels within seconds. There are
> no named diagnostics to be found in /var/log/messages.
>
> I've done some searching with Google and found a lot of references
> to CPU utilization problems in named, but have not found anything yet
> relating it to whether or not the internet is connected. Is this a bug? A
> configuration problem? Will updating to bind 9 be of any help?

Two things: bind-8.4.6 is quite old; bind-8.x support is currently in
maintenance mode only. You should consider updating to bind-9.3.4.
Second, for a dial-up connection you probably should consider setting
up bind in caching-only mode. That way it won't keep trying to pull
down zone updates from the internet when you're not connected.

--

John ((E-Mail Removed))

>> I thought the ISC folks wanted everyone migrating to BIND 9 ages ago
>> so it might be good to do just on first principles...

> That's probably something that should be done. Didn't think it would make
> much difference on a nameserver for a small LAN, but you never know.

I'm not sure who "small" is "small", but for tiny LANs you may want to take
a look at dnsmasq. It's just perfect for home LANs behind a NAT router,
for example.


Stefan

In article <(E-Mail Removed)>, John Thompson wrote:
> Two things: bind-8.4.6 is quite old; bind-8.x support is currently in
> maintenance mode only. You should consider updating to bind-9.3.4.

I am building BIND 9 from source as this is being written, will give it
a try. I don't need a lot of the new features (I have absolutely no
interest in IPV6, for example), but if the newer version does not
exhibit this problem I'll use it.

> Second, for a dial-up connection you probably should consider setting
> up bind in caching-only mode. That way it won't keep trying to pull
> down zone updates from the internet when you're not connected.

That's supposed to be what I have, using a set of pre-packaged config
files. The only modificatons were to add the "forwarders" option to
point to my ISP's name servers, and to add a zone for the local network,
which consists of 3 computers that all share the dialup connection.

--
Roger Blake
(Subtract 10 for email.)