I’m looking for ways to improve throughput of my Flink job. I used the code here: https://stackoverflow.com/questions/33904793/kafka-flink-performance-issues by @rmetzger that was helpful in figuring my throughput, and I can get 50k msg/core/s and ~70MB/s from a single topic/1 partition. However I am confused: the job ‘busy’ status is barely 3%, the CPU is ~200mi on the TM, and I know my Kafka cluster can get me at least 100MB/s per partition, so somehow I’m only at 70% of what I could expect, with the job not being very busy. If I read from 2 topics, 1 partition each, I get to 140MB/s, so the job seems perfectly able to handle that, yet with 1 topic/1partition is is not able to reach max throughput. I tuned

fetch.max.bytes

,

max.partition.fetch.bytes

,

max.records

as well as

receive.buffer.bytes

to much larger numbers than default, making little to no difference, so I feel like it is not a consumer config issue. I’m just not sure what else it may be at this point. Any pointed to how to troubleshoot this would be appreciated. Thanks