In our kafka-mediated ELK environment, we find that filebeat templates direct that the kafka.partition data item be type long, and metricbeat templates direct that it be type Object. This generates conflicts downstream in e.g. kibana.
I believe this prevents me from plotting data from the two sources on the same visualization. But that may be my ignorance.
I'd love to hear how other people are dealing with this facet of the beats world, especially if someone can point out a thing I'm missing because I'm just ignorant of some aspects of ELK.