I'm looking for some job design insight. I have two process functions in my system. The first operates event data directly from devices (keyed by a device ID). That process function generates side outputs that are keyed by a rule identifier for further processing and feed into the other process function. I would like for the second process function to maintain an understanding of the latest event time seen for each device, but the current layout doesn't achieve that. Some ideas I had • Make the first process function generate a side output of something simple like

{device: <id>, time: <event time>}

, broadcast that side output stream and connect it to the stream that feeds the second process function. • Generate multiple events out of the first process function like

{rule: <id>, device: <id>, time: <event time>}

where all that changes is rule ID so that this side output can be connected to the other side output that is keyed by the same namespace. • Something else? The former seems easier but perhaps not the proper intent of broadcast state. The latter seems like a lot of extra traffic/noise in the system.