Questions about join semantics about goka HOT 2 CLOSED

When consuming a stream, you can enrich the stream by joining it with tables using ctx.Join or ctx.Lookup, for example, an event "user-logged-in" arrives; you lookup a device-table using the device id in the user-logged-in event and combine the device information with the event. The result can be emitted downstream or stored in a group table. So this is a Stream-Table join. The callbacks will be triggered by the stream and you'll see all events from the stream. The joined tables are passive, they will be kept up-to-date, but you typically don't get triggered for their updates (unless you use UpdateCallback).

In your case, you want to count how many bookings a user has performed. I would do something similar to your approaches, but I would not pack user and booking in the same processor. I'd have one processor for the booking-table and one processor for the user-table. The booking processor would emit into a topic "user-booked-stream" using the user as key. The user processor would consume (with multiple callbacks) from the user streams plus this "user-booked-stream".

The at least once issue would have to be solved by keeping a set of bookings of the user instead of simply counting them. Does that help you further?

from goka.

@db7

To do exactly once counting, that's exactly what I was thinking about (keeping set of ids instead of a counter, at least during some limited timeframe/window).

Thanks a lot for your recommendations, it helped a lot!

from goka.