[Dnsmasq-discuss] Occasional "communications error", how to diagnose?

[Simon Kelley]

On 13/12/2023 15:25, Chris Green wrote:
> I run dnsmasq version 2.89 on my laptop which is running [x]ubuntu
> 23.04.
> 
> I have systemd.resolvd disabled.
> 
> I'm occasionally seeing the following error when getting a host's IP:-
> 
>      chris$ host homepi
>      ;; communications error to 127.0.0.1#53: timed out
>      homepi has address 192.168.1.113
>      chris$ ps -ef | grep dnsmasq
>      dnsmasq      933       1  0 Dec06 ?        00:00:22 /usr/sbin/dnsmasq -x /run/dnsmasq/dnsmasq.pid -u dnsmasq -7 /etc/dnsmasq.d,.dpkg-dist,.dpkg-old,.dpkg-new --local-service --trust-anchor=.,20326,8,2,e06d44b80b8f1d39a95c0b0d7c65d08458e880409bbc683457104237c7f8ec8d
>      chris      86541    3774  0 15:05 pts/1    00:00:00 grep --color=auto dnsmasq
>      chris$
> 
> As can be seen dnsmasq is running and subsequent queries work without any
> error (or delay).  The above timeout is a few seconds, maybe five or a bit
> less.
> 
> There's no dnsmasq related error message in syslog (nothing for today at
> all).  The system homepi is a Raspberry Pi on the same LAN as the laptop
> running dnsmasq, The error isn't only for one particular host, I've seen
> it for other systems on my LAN.
> 
> Can anyone suggest what might be causing the error and/or how to diagnose
> what's wrong?
> 

It looks like the first query (or its reply) was dropped, host retried, 
and it worked second time around.

Since DNS transport is normally across UDP, which is defined as 
unreliable, this is completely normal. Except that the UDP packets are 
not actually traversing a network, they're going via the lo interface 
within one machine. I'm sure there are circumstances where UDP packets 
can get dropped in the kernel when going via the lo interface, but it 
shouldn't happen very often. Is the machine under heavy load or memory 
pressure? Maybe a network reconfiguration event could drop packets?

Simon.