Hi, I am setting up metrics monitoring on a new Kafka 0.10.0.1 cluster and observed the following: - kafka.network:type=RequestMetrics,name=TotalTimeMs,request=Produce 's 99th percentile is 18.71ms (remote=15.71, response send time, local and queue times are both 1ms), so far this seems to be reasonable - kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchFollower 's 99th percentile is 146.71ms (remote=146.71, others' both 1ms), this seems to be a little high for me, and if I check the 75th percentiles, it is a little bit more than 100ms - kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchConsumer 's 99th percentile is 502ms (remote=501, others' both 1ms), this looks like a bug, even the 75th percentile remote value is 500ms. The cluster has 3 kafka brokers, ping time between them is 0.2ms or 0.9ms, runs on AWS with plenty of free resources. Both clients are consumer and producer at the same time, message volume is minimal (about 50 msgs / sec / broker), message size is probably below 100byte. As I understand the documentation, the remote time should be zero for non producer metrics. Please let me know if this seems to be a bug / bad configuration / bad environment. Thanks, Sini
