I have a 3 node Kafka cluster with a replication factor of 3, min in sync replicas of 2 and required acks set to all.
Everything works fine until I turn off one broker node, then the Java Producer stops producing for ~15 seconds until it works again. I have figured out that this delay is the zookeeper session timeout ms value. It looks like the Producer is waiting until new partition leaders are elected, and this apparently only occurs after the zookeeper session timed out.
Does anyone know if this behavior is normal and if it's possible to reduce this delay to like a few Ms? I would like that the Producer switches instantly after one node fails.