Kapacitor streaming jobs and timestamps


I am using Kapacitor to do some streaming jobs that use window nodes, and I’ve noticed some unexpected behavior. It looks like sometimes points that arrive late (compared to their timestamp) are getting bucketed into the current window. Does Kapacitor look at the timestamp to figure out whether a point belongs in a window, or is it simply based on when points arrive?


Kapacitor looks at the timestamp of the data but if points arrive significantly out of order weird things can happen.

Have a look at this https://github.com/influxdata/kapacitor/issues/1058 Any input you have for your use case is welcome.