Hi, We have a use-case where there are close to 40 producers to a topic with 2 replicas and consumers for this topic are around 2000. We have seen that producer latency goes by 3 times when consumers grew from 500 to 2000. We changed the following properties. The machine has sufficient memory. The rate is around 100k we are targeting.
- queued.max.requests - num.io.threads - num.network.threads But there is no impact. Please, could you suggest how to go about this? Regards, Aravind