Hey, I'm no expert at all, but for me this sounds like a use case for Complex Event Processing (CEP). I don't know if you're aware of Flinks CEP Library [1, 2]? Maybe that solves your problem of multiple firings. But best to wait for the experts to answer your questions on handling state and firing windows :)
Best, Claudia [1]: https://flink.apache.org/news/2016/04/06/cep-monitoring.html [2]: https://ci.apache.org/projects/flink/flink-docs-master/dev/libs/cep.html -----Ursprüngliche Nachricht----- Von: Maciek Próchniak [mailto:m...@touk.pl] Gesendet: Freitag, 23. September 2016 10:36 An: user@flink.apache.org Betreff: window-like use case Hi, in our project we're dealing with a stream of billing events. Each has customerId and charge amount We want to have a process that will trigger event (alarm) when sum of charges for customer during last 4 hours exceeds certain threshold, say - 10. The triggered event should contain data from last billing event (the one that triggered alarm) One one hand we can implement it as custom state - we'd save charges (or some precomputed aggregates) from last 4 hours and trigger event when new one arrives. OTOH we've been thinking if we can make it easier by using flink windows. We tried to model our situation as sliding events (of length 4hours, slide is 1h), have some precomputed aggregate and have a custom trigger that fires on element when threshold is exceeded. It kinda works, except for the fact that state is unnecesarily large, custom trigger is a bit hacky and (worst of all) when event with charge amount e.g. 20 arrives trigger fires in all slides and we have duplicated events. That's why we currently think about implementing it by custom state... Do you have any other ideas/recommendations how can we handle such requirement? thanks, maciek
