I have an issue with the aggregate operator time window.
It seems that it has some kind of divergence in the time when he triggers the window.
I'm using a tumbling, partitioned window with time parameter, but if I give him lets say 120 secs sometimes ge fire it much sooner.
I thought that the aggregate is start countiing for each group individualy from the first tuple that arrive to this group.
I've been told that time window starts to count from the moment the job is been uploaded, but how can it be true in parition mode?
help please. thank you