[op5-users] Réf. : Re: Réf. : Re: Réf. : Re: Nagios start time delay with Merlin

[Andreas Ericsson]

nicolas.raspail at bnpparibas.com wrote:
> op5-users-bounces at lists.op5.com wrote on 26/08/2009 14:07:07:
> 
>> nicolas.raspail at bnpparibas.com wrote:
>>> I will try the new beta. Is a tar file available or this version must 
> be 
>>> retreive from the git repository ?
>>>
>> You'll need to get it from git, I think. We still haven't gotten around
>> to providing tar-balls automagically every time we tag. That should
>> happen in the next few days though, as we're in release-frenzy at the
>> moment and usually face a slightly calmer period the first week or so
>> after a major release.
>>
> 
> I have got it from 
> http://git.op5.org/git/?p=nagios/merlin.git;a=commitdiff;h=4ae5c2ba5bb494d60a865df8a34ec66f9f32676e
> 
> 
>>>>>         * nagios has been restarted at timestamp 1251281934 and 
> since 
>>>>> that, no check have been made until timestamp 1251282809 ! 15 mins 
> of 
>>>>> update, that is a lot of time.
>>>> And far, far more than we're experiencing here. Merlin is designed in
>>>> such a way that it rather drops messages than interferes with the
>>>> running Nagios daemon, so what you're seeing is almost certainly not
>>>> a result of Merlin doing something weird.
>>>>
>>>> This should be alleviated by upgrading to the latest Merlin version
>>>> though, since it ignores all events Nagios throws at it until Nagios
>>>> has entered the main event execution loop. Under such circumstances,
>>>> the startup time can't be affected at all by Merlin.
>>>>
>>> Ok, but without Merlin, my Nagios starts immediately some checks. With 
> 
>>> NDO,
>>> there is also a delay (7/8 mins), and the mysql server is very busy 
> during
>>> this period. Maybe there is a problem else, but I can't find where.
>>>
>> With the latest Merlin, checks should start immediately with merlin too.
>>
> 
> I will compile it and git it a try
> 
>>>>> Also, During these tests, I have see my check and hosts latencies 
> grow 
>>> up 
>>>>> and now, with Merlin enabled, I have a large number of orphaned 
>>> checks.
>>>>> With Merlin I have the following latencies (and it is increasing as 
> I 
>>>>> write my email) :
>>>>>
>>>>> Service Check Latency:  0.00 / 1299.86 / 160.156 sec
>>>>> Host Check Execution Time:      2.54 / 3.19 / 2.564 sec
>>>>> Host Check Latency:     0.00 / 723.49 / 304.415 sec
>>>>>
>>>>> Before Merlin, i don't have the exact values, but I remember that 
> the 
>>>>> service latency was under 50s and the host latency under 1s
>>>>>
>>>> Was the latency slowly increasing before, or was it totally stable?
>>>> Merlin does add a small overhead to the processing of each check
>>>> result, status update and a plethora of other things. If your
>>>> latency was previously increasing slowly, Merlin will make it
>>>> increase faster. If it was stable before, it's possible that the
>>>> (very small) overhead that Merlin adds is pushing it over the
>>>> limit so that the latency starts converging on infinity.
>>> Before Merlin, the latency was totally stable. I understand why Merlin 
> and 
>>> NDO add
>>> a small overhead, but what I'm facing is a huge overhead.
>>>
>>> I have disabled Merlin and enabled NDO to compare. Here is my actual 
>>> latency with NDO
>>> after 15/20 minutes :
>>>
>>> Service Check Execution Time:   0.04 / 30.03 / 0.389 sec
>>> Service Check Latency:  0.00 / 1952.96 / 300.960 sec
>>> Host Check Execution Time:      2.54 / 2.69 / 2.566 sec
>>> Host Check Latency:     0.00 / 17.98 / 5.105 sec
>>>
>>> And the latency is still decreasing as I write my email.
>>>
>> Ah, I think I know what's happening now. Merlin schedules a reaper
>> event once every 5 seconds, where it basically just sits and waits
>> for input from the daemon. This reaper event can be dropped
>> completely if no pollers or peers are configured, so I'll make
>> sure to add something to that effect. This should make Merlin
>> perform better than NDOUtils, since we're transferring fewer
>> events and we're transferring them with a more efficient protocol.
> 
> Should I test the beta5 of Merlin or wait for the release where you
> will make this change ?
> 

Please test the beta5 of Merlin. The release-tag will be set where
the beta5 tag is now, unless our testing uncover some truly amazing
deficiency that we have so far been completely unable to spot (not
very likely). In all likelihood, 0.6.2 will therefore be the same
code as is in 0.6.2-beta5.

> Right now, with NDO, my latencies seems to be stable with theses values
> 
> Service Check Execution Time:   0.04 / 30.02 / 0.372 sec
> Service Check Latency:  0.00 / 1299.86 / 46.321 sec
> Host Check Execution Time:      2.54 / 2.62 / 2.563 sec
> Host Check Latency:     0.00 / 3.45 / 1.179 sec
> 

That's a fairly high service check latency though. Hopefully Merlin
will cope better than that.

-- 
Andreas Ericsson                   andreas.ericsson at op5.se
OP5 AB                             www.op5.se
Tel: +46 8-230225                  Fax: +46 8-230231

Considering the successes of the wars on alcohol, poverty, drugs and
terror, I think we should give some serious thought to declaring war
on peace.