Alleviating Nagios i/o contention problem

[Matthias Flacke]

On 9/25/10 2:30 PM, Frost, Mark {PBC} wrote:
> Greetings, listers,
> 
>  
> 
> We've got an on-going issue with i/o contention.  There's the obvious
> problem that we've got a whole lot of things all writing to the same
> partition.  In this case, there's just one big chunk of RAID 5 disk on a
> single controller so I don't believe that making more partitions is
> going to help.
> 
>  
> 
> On this same partition we have:
> 
>  
> 
> 1) Nagios 3.2.1 running as the central/reporting server for a couple of
> other Nagios nodes that are sending check results via NSCA. 
> Approximately 6-7K checks.
> 
>  
> 
> 2) pnp4nagios 0.6.2 (with rrd 1.4.2) writing graph data.
> 
>  
> 
> There's a 2nd server configured identically to the first that's acting
> as a "hot spare" so it also receives check data from the 2 distributed
> nodes and writes its own copy of the graph data locally as well.
> 
>  
> 
> At the moment I'm concerned about the graphdata, but because I can only
> see i/o utilization as an aggregate, I can't tell what is the worst
> component on that filesystem -- status.dat updates?  graph data?  writes
> to the var/spool directory?  We also look at continued growth so this is
> only going to get worse.
> 
>  
> 
> These systems are quite lightly loaded from a CPU (2 dual-core CPUs) and
> memory (4GB) perspective, but the i/o to the nagios filesystem is
> queuing now.
> 
>  
> 
> We're about to order new hardware for these servers and I want to make a
> reasonable choice.  I'd like to make some reasonable changes without
> requiring too exotic of a setup.  I believe these servers are currently
> Dell 2950s and they're all running Suse Linux 10.3 SP2.
> 
>  
> 
> My first thought was to potentially move the graphs to a NAS share which
> would shift that i/o to the network.  I don't know how that would work
> though and it would ultimately be an experiment.
> 
>  
> 
> What experiences do people out there have handling this kind of i/o and
> what have you done to ease it?

You didn't say how many of your checks create perfdata - but I assume
that most of your disk I/O is related to RRD updates.
RRD cached (see http://docs.pnp4nagios.org/pnp-0.6/rrdcached for PNP
integration) is a good means to collect multiple RRD updates and burst
write the RRD files.

status.dat and the checkresults directory are always good candidates to
be stored on a ramdisk, especially since they're volatile data. As a
side note: status.dat on ramdisk is a pure boost for the CGIs :).
I know people which also store nagios.log on a ramdisk and regularily
save them via rsync onto a hard disk.

My own systems with ~4000 checks and ~20.000 performance relevant data
sets went down from 30% to less than 2% wait I/O with rrdcached and
ramdisk use.

Cheers,
-Matthias

------------------------------------------------------------------------------
Start uncovering the many advantages of virtual appliances
and start using them to simplify application deployment and
accelerate your shift to cloud computing.
http://p.sf.net/sfu/novell-sfdev2dev
_______________________________________________
Nagios-users mailing list
Nagios-users at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nagios-users
::: Please include Nagios version, plugin version (-v) and OS when reporting any issue. 
::: Messages without supporting info will risk being sent to /dev/null