[mrtg] Intermittent Graphs

[Nico Kadel-Garcia]

Alex van den Bogaerdt wrote:
> On Sat, Feb 02, 2008 at 08:55:46AM -0500, Darren.Terry at cox.com wrote:
>
>   
>>      This past week I migrated a very, very large MRTG installation to
>> new servers and updated software. Everything went well with the
>> relocation with the exception of one group of hardware. The group of
>> CMTS systems that I am graphing aren't graphing correctly. They have
>> graphs which are very intermittent. They look like saw blades. I believe
>> the reason that they are so intermittent is that the poll isn't
>> finishing before the next one needs to start. So, I am receiving the
>> lock file error in email. However, this doesn't make sense. I upgraded
>> to servers that are LEAST five times stronger. In addition to this,
>> according to the FAQ on the MRTG site, the graphs should be populated
>> with the previous data rather than zero.
>>
>> 	If anyone has any ideas about how to speed up the poll or
>> anything else that might help, I am all ears (or eyes?)!! 
>>     
>
> Lock file errors, meaning it takes too much time to finish one round,
> and the next one is started before the prevous one has finished.
>
> Perhaps two polling periods is enough for those counters to wrap
> around, similarly to the infamous 114Kbps at 5-minute intervals.
>
> Things to try:
> * increase IO speed
> * use HC counters
> * run multiple MRTG instances parallel
>
> A temporary workaround may be to increase the polling interval from
> five to six minutes (or maybe seven), but not to ten as that would
> most likely will result in more, not less, saw blade-looking.
>
> HTH
> Alex
>   
This is compounded by unpingable routers, that introduce massive delays 
into the polling. Breaking up mrtg into checks of small numbers of 
machines is handy: running a pre-check with "ping" or the Nagios 
"check_alive" tool before running MRTG on a target is also handy.