drop in fetch-follower requests

[Wesley Chow <we...@chartbeat.com>]

We deployed a Samza process that effectively doubled our fetch rates in our
cluster and noticed a significant drop fetch-follower requests. What is the
difference between fetch-consumer and fetch-follower?

[image: Inline image 1]

We noticed that some topics moved along undisturbed, sending in the same
rate of data they had been before. Other topics saw a drop in traffic. It
seems like the pattern is that the topics with reduced message rate were
derived from our source topics, ie consumers of a source topic that
published to another topic.

It doesn't seem like we hit any sort of machine resource limits. We're well
under network, CPU, and disk IO utilization. This is a 6 node cluster with
all topics replicated 3 times (ie, one leader, two followers).

Any thoughts on what might have happened?

Thanks,
Wes