Reducers are able to store whatever they want in state without limits. This puts difficult to pinpoint pressure on the platform and produces bursty performance followed by bursty slow downs.
Perhap could lookup at the returning state and decide some default strategy for breaking state up. However the effectiveness of a such a default strategy being applicable to a wide range of specific data model and workloads is questionable.
An alternate strategy would be to decide which workloads perhap is to carry and let it emit events outside into Kafka or another system to allow other systems to process additional workloads.
Reducers are able to store whatever they want in state without limits. This puts difficult to pinpoint pressure on the platform and produces bursty performance followed by bursty slow downs.
Perhap could lookup at the returning state and decide some default strategy for breaking state up. However the effectiveness of a such a default strategy being applicable to a wide range of specific data model and workloads is questionable.
An alternate strategy would be to decide which workloads perhap is to carry and let it emit events outside into Kafka or another system to allow other systems to process additional workloads.