Wrong current values with CDEF

[X-dark]

Hi,

I'm using thold 0.4.3 with Cacti 0.8.7e and PIA 2.5.

I want to make alert when the ratio between two values of a data template is below a given value. I have created two thold templates that should have the same behavior but based on one value of the data template or the other.

The values that thold use to decide if the threshold should be triggered is wrong. It seems to be calculated with values belonging to different polling cycles.

Here is a detailed explanation of what I have setup:

My data template with two counters "received" and "forwarded":


The first thold template created:

The CDEF is

Code: Select all

cdef=CURRENT_DATA_SOURCE,ALL_DATA_SOURCES_NODUPS,CURRENT_DATA_SOURCE,-,/,100,*

The second thold template created:

The CDEF is

Code: Select all

cdef=ALL_DATA_SOURCES_NODUPS,CURRENT_DATA_SOURCE,-,CURRENT_DATA_SOURCE,/,100,*

I've instantiated the two templates above for a given graph instance. If I go to the instance of the first template I get the good and expected value:


And if I go to the instance of the second template I get, as expected, the same value as with the first template:


Now, comes the "not working as expected" part. Here is what I get when I display the thold states. Not that the value are not the same as the ones seen before (the screenshots have been taken within the same polling cycle) and even differs between them. One of the value is superior to 100% which never occurs with the graph.


Does anyone have a suggestion ?

NB: I know the versions I use are not the latest but as my Cacti installation involve 5 Cacti instances (with 2,000 hosts and 600,000 data sources) upgrade must be thoroughly tested. If this is a known bug with this thold version and I can upgrade only thold without changing Cacti, PIA or Boost version to solve it, it should be doable.

[TheWitness]

If you want accurate data, you should migrate to the RPN expression instead of the CDEF method. The issue with the CDEF is that we have to go to the RRDfile for some of the data, which if say for example you are using boost, defeats the purpose of boost, and if both items have not been polled, results in inaccurate data.

The RPN Expression route should provide clearer answers while at the same time, not affecting scalability.

[X-dark]

Thanks.

If you want accurate data, you should migrate to the RPN expression instead of the CDEF method.

OK. Then I need to update to thold >= 0.4.6. Is there any constraint regarding Thold/Cacti/PIA/Boost versions? There is nothing about this in Thold README.

The issue with the CDEF is that we have to go to the RRDfile for some of the data, which if say for example you are using boost, defeats the purpose of boost, and if both items have not been polled, results in inaccurate data.

It happens boost on demand updating is disabled on this specific Cacti instance.

The RPN Expression route should provide clearer answers while at the same time, not affecting scalability.

Good.

[TheWitness]

For your first question, I would be at PIA 2.9+, or in other words the 'G' version of Cacti. I would not goto the 'H' version due to some bugs to be cleaned in an upcoming release.

[X-dark]

For your first question, I would be at PIA 2.9+, or in other words the 'G' version of Cacti. I would not goto the 'H' version due to some bugs to be cleaned in an upcoming release.

Ok. This is what I was afraid of. I've begun to make some test yesterday with only a newer thold version and this does not seems to work very well.