I have an analysis in PI AF that uses EventCount on 2 attributes and then divide them to find a ratio. Eventually, it triggers a notification if the ratio is less than 75%.
When I reproduce the same calculation in DataLink or when I run a Preview or a Backfilling, the results are good. However, when the analysis run by itself (triggered every hour), the results are not always good : if the number of events is high (about >600), the count is correct. If the number of events is low (<300) the count return by the analysis is roughly twice the value that I expect. Between 300 and 600 events is a gray zone : sometimes the count is correct, sometimes it is not.
Therefore, the count may be correct on 1 attribute and bad on the other attribute. In this case, the ratio is out of range and a false notification is sent.
Here are the expressions used in my analysis :
NbRang : Eventcount('Rang','*-4h','*')
NbID : Eventcount('Pl-m ID','*-4h','*')/2
PctLu : if NbRang > 600 then (100 * NbID / NbRang) else NoOutput()
Note that this analysis was working fine until the upgrade from PI Server 2016 to PI Server 2018.
Any idea on what is going on ?