[Dnsmasq-discuss] excessive cpu consumption in 2.39rc1

[Simon Kelley]

Carlos Carvalho wrote:
> Simon Kelley (simon at thekelleys.org.uk) wrote on 12 April 2007 18:19:
>  >Carlos Carvalho wrote:
>  >> I've been using rc1 without problems but today I noticed a big cpu
>  >> consumption. I also get many of these messages in the log:
>  >> 
>  >> dnsmasq[6137]: overflow: 27 log entries lost
>  >> 
>  >> I disabled log-queries and log-dhcp and still get them(?!). Any ideas
>  >> about what's going on?
>  >
>  >Is the CPU being used by dnsmasq, or is the machine busy because of 
>  >other load?
> 
> At that moment it was both.
> 
>  >That message is given by the latest code when is cannot log 
>  >stuff fast enough, which could be because syslogd is not running fast 
>  >enough, or the disk is too busy to write the logs. If you do --log-max=0 
>  >then the old, blocking behaviour is restored, so that you'll be able to 
>  >see what dnsmasq was trying to log.
> 
>  >If the extra CPU load comes from dnsmasq, there might be a bug in the 
>  >new logging code. Are you using syslog, or syslog-ng?
> 
> syslog. Didn't have time to change yet.
> 
> I found strange that it complained about overflow even without
> log-queries and log-dhcp but this only happened right after starting,
> so it could be because of the overall load.
There are quite a lot of log lines created at startup, so that makes 
some sense.
> 
> I'll turn the logging again and use --log-max=0 to see what happens.

OK. I suspect that this a symptom of a loaded system: the trade-off for 
lost logging is that dnsmasq keeps working fast, without waiting for syslog.

If you see messages like this, when dnsmasq shouldn't be logging, that's 
  a bug.

Increasing log-max (the default is 5) will reduce the number of lost 
messages.


Cheers
> 
> _______________________________________________
> Dnsmasq-discuss mailing list
> Dnsmasq-discuss at lists.thekelleys.org.uk
> http://lists.thekelleys.org.uk/mailman/listinfo/dnsmasq-discuss
>