High latencies problem.

[D. Emmanuel Feinsmith]

Dear Alessandro,

You are more than likely eating up the cpu and memory with the  
memcpy's executed by each fork of your check_nrpe and check_icmp  
services. You can prove this out to yourself by using top to observe  
the behaviour of the nagios processes. I would also suggest that there  
is nothing else eating up CPU and memory on your nagios server box and  
keep the box dedicated. Running top will show if there is resource  
contention on your monitoring server. Keep in mind that check_nrpe is  
amongst the slowest possible commands nagios can execute because it  
has to wait for whatever timeout period you entered in your client  
nrpe.cfg for the nrpe daemon to respond. This can take seconds in some  
cases. A much more scalable solution is to enable passive checks  
(using nsca/send_nsca) on some or all of your clients)

I would suggest the following things (from the nagios performance  
tuning guide):

# Check service latencies to determine best value for maximum  
concurrent checks. Nagios can restrict the number of maximum  
concurrently executing service checks to the value you specify with  
the max_concurrent_checks option. This is good because it gives you  
some control over how much load Nagios will impose on your monitoring  
host, but it can also slow things down. If you are seeing high latency  
values (> 10 or 15 seconds) for the majority of your service checks  
(via the extinfo CGI), you are probably starving Nagios of the checks  
it needs. That's not Nagios's fault - its yours. Under ideal  
conditions, all service checks would have a latency of 0, meaning they  
were executed at the exact time that they were scheduled to be  
executed. However, it is normal for some checks to have small latency  
values. I would recommend taking the minimum number of maximum  
concurrent checks reported when running Nagios with the -s command  
line argument and doubling it. Keep increasing it until the average  
check latency for your services is fairly low.

# Optimize host check commands. If you're checking host states using  
the check_ping plugin you'll find that host checks will be performed  
much faster if you break up the checks. Instead of specifying a  
max_attempts value of 1 in the host definition and having the  
check_ping plugin send 10 ICMP packets to the host, it would be much  
faster to set the max_attempts value to 10 and only send out 1 ICMP  
packet each time. This is due to the fact that Nagios can often  
determine the status of a host after executing the plugin once, so you  
want to make the first check as fast as possible. This method does  
have its pitfalls in some situations (i.e. hosts that are slow to  
respond may be assumed to be down), but you'll see faster host checks  
if you use it. Another option would be to use a faster plugin (i.e.  
check_fping) as the host_check_command instead of check_ping.

# Schedule regular host checks. Scheduling regular checks of hosts can  
actually help performance in Nagios. This is due to the way the cached  
check logic works (see below). Prior to Nagios 3, regularly scheduled  
host checks used to result in a big performance hit. This is no longer  
the case, as host checks are run in parallel - just like service  
checks. To schedule regular checks of a host, set the check_interval  
directive in the host definition to something greater than 0.

# Enable cached host checks. Beginning in Nagios 3, on-demand host  
checks can benefit from caching. On-demand host checks are performed  
whenever Nagios detects a service state change. These on-demand checks  
are executed because Nagios wants to know if the host associated with  
the service changed state. By enabling cached host checks, you can  
optimize performance. In some cases, Nagios may be able to used the  
old/cached state of the host, rather than actually executing a host  
check command. This can speed things up and reduce load on monitoring  
server. In order for cached checks to be effective, you need to  
schedule regular checks of your hosts (see above). More information on  
cached checks can be found here.

For more, see:

http://nagios.sourceforge.net/docs/3_0/tuning.html

If none of this works, you may have to use passive checks or multiple  
nagios instances to drop your latency.

Bon Chance!
Daniel.

On Feb 17, 2009, at 8:41 AM, Alessandro Ren wrote:

> On 2/17/2009 1:32 PM, D. Emmanuel Feinsmith wrote:
>
>     Answers bellow,
>> Alessandro,
>>
>> 1.  what is the breakdown between passive and active checks? For
>> passive checks, there are many ways to increase the # of services
>> through bypassing the command pipe (which nsca writes to). With
>> passive checks done in this way I've gone to 50,000 services with
>> under 10 second latency.
>>
>     All active checks, no passive.
>
>> 2.  how many of those services are check_icmp or check_ping? If there
>> is a good number of those, you can use fping to reduce the # of fork/
>> exec's that nagios has to perform, which is a major area of resource
>> utilization within the nagios server.
>>
>     Less than 5% are ping checks and we use check_icmp for all those.
> Most checks are check_nrpe,.
>
>> 3. Are you using a performance data handler or OCSP? If so, you might
>> either find a way to get rid of these entirely, or be sure you are
>> using file based performance handling at the very minimum.
>>
>     I am using perfparse to write to mysql. Disabling it has no effect
> in the latency.
>
>> The key to nagios scalability and latency reduction is to educe the #
>> of fork/exec's to the smallest amount possible and keep away from the
>> command pipe as much as you can if you are passive-check heavy. If  
>> you
>> are using all active checks, then you can balance the load between
>> active and passive checks and thereby gain some speed.
>>
>
>     In my other nagios with just 2600 services, we see around 200
> nagios processes running in average, in the 11600 services system, the
> average is 30 processes, it seems that the event loop in lagging, is  
> is
> not starting enough processes, thus raising the latency.
>
>     Thank you Daniel.
>> Daniel.
>>
>> On Feb 17, 2009, at 8:17 AM, Alessandro Ren wrote:
>>
>>
>>>    Hello,
>>>
>>> I have a nagios system running with 427 hosts and 11160 services and
>>> since I reached 8000 services, I am having problems with the latency
>>> beeing around 100s and 200s.
>>>    use_large_installation_tweaks is enabled, max_concurrent_checks
>>> have
>>> been tested with 0 and higher values and I have tested this setup in
>>> two
>>> different HWs, a dual core with 4GB RAM 32 bits a a Dual Xeon Dual
>>> core
>>> 64bits with 8GB of RAM. We are using REdHat enterprise 5.
>>>    Also reaper is already at 2s, host checks with cache horizon are
>>> enabled with a max retry of 3, all services check every 5min.
>>>    I have no service dependency set up.
>>>    I've noticed that nagios is not spawning too many processes as
>>> another nagios I have running which has far less servicexs and it
>>> seems
>>> that the event loop if lagging behing, in my debugs.
>>>    Any ideas what could I do to fix that? Have I reached a limit in
>>> nagios pooler code?
>>>
>>>    Tks.
>>>
>>> -- 
>>> Alessandro Ren
>>> http://www.opservices.com.br
>>> alessandro.ren at opservices.com.br
>>>
>>> ------------------------------------------------------------------------------
>>> Open Source Business Conference (OSBC), March 24-25, 2009, San
>>> Francisco, CA
>>> -OSBC tackles the biggest issue in open source: Open Sourcing the
>>> Enterprise
>>> -Strategies to boost innovation and cut costs with open source
>>> participation
>>> -Receive a $600 discount off the registration fee with the source
>>> code: SFAD
>>> http://p.sf.net/sfu/XcvMzF8H
>>> _______________________________________________
>>> Nagios-devel mailing list
>>> Nagios-devel at lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/nagios-devel
>>>
>>
>>
>> ------------------------------------------------------------------------------
>> Open Source Business Conference (OSBC), March 24-25, 2009, San  
>> Francisco, CA
>> -OSBC tackles the biggest issue in open source: Open Sourcing the  
>> Enterprise
>> -Strategies to boost innovation and cut costs with open source  
>> participation
>> -Receive a $600 discount off the registration fee with the source  
>> code: SFAD
>> http://p.sf.net/sfu/XcvMzF8H
>> _______________________________________________
>> Nagios-devel mailing list
>> Nagios-devel at lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/nagios-devel
>>
>
> ------------------------------------------------------------------------------
> Open Source Business Conference (OSBC), March 24-25, 2009, San  
> Francisco, CA
> -OSBC tackles the biggest issue in open source: Open Sourcing the  
> Enterprise
> -Strategies to boost innovation and cut costs with open source  
> participation
> -Receive a $600 discount off the registration fee with the source  
> code: SFAD
> http://p.sf.net/sfu/XcvMzF8H
> _______________________________________________
> Nagios-devel mailing list
> Nagios-devel at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/nagios-devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.monitoring-lists.org/archive/developers/attachments/20090217/9218917d/attachment.html>
-------------- next part --------------
------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
-------------- next part --------------
_______________________________________________
Nagios-devel mailing list
Nagios-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nagios-devel