[Pdns-users] recursor stops answering under stess

[Chris Foote]

On Thu, 14 Dec 2006, Juergen Georgi wrote:

> On Dec 14, 15:12, bert hubert wrote:
>> Subject: Re: [Pdns-users] recursor stops answering under stess
>> On Thu, Dec 14, 2006 at 02:39:03PM +0100, Juergen Georgi wrote:
>>> while doing perfomance testing on an Sun-Fire-V240 (Sparc, Solaris-10),
>>> pdns_recursor 3.1.4 just stopped answering on its bge0 interface.
>>> However, it was still responding to queries on 127.0.0.1.
>>
>> Did you bind it to a specific IP address or to 0.0.0.0?
>
> Yes, a specific IP: local-address=127.0.0.1,129.143.2.25
>
>>> The silence on bge0 occurred at about 500.000 queries
>>> and continued even after shutdown of queryperf.
>>>
>>> Could this be the effect of throttling - a client punishment
>>> somehow?
>>
>> No. Does your machine have multiple IP addresses?
>
> Yes, 129.143.2.25 is bge0:1, bge0:2 is 129.143.2.72 and
> is used as query-local-address.
>
> In addition, the dns service runs in a Solaris zone.

Hi Juergen.

If the bge0:1 ethernet alias 129.143.2.25 is not within the IP subnet
that the main bge0 interface uses, then you might have run into the
same issue as I did on a Linux box where the ARP table's entry for the
default gateway had been aged out; ARP requests for the default gateway
had a IP source of the alias, not the main interface, and the gateway
router correctly doesn't reply.

If so, `tcpdump -i bge0 arp` would show this.  The trick under Linux
was to use the 'arptables' tool to mangle outgoing ARP requests - i.e.:

/sbin/arptables -A OUTPUT \
          -s 172.16.3.25 -d 192.168.3.1 -o eth0 \
          --opcode=Request \
          -j mangle --mangle-ip-s 192.168.3.66

(where:
  - 172.16.3.25 was the eth0:0 alias
  - 192.168.3.1 was the default gateway IP
  - 192.168.3.66 was the eth0 address
)

I don't know if a similar facility exists in Solaris though.

Cheers,
Chris