There is a good research paper from IBMon this topic - it’s a little outdated, but I’m sure that the main statements are still preserved and only improved. The bottom line is that the overhead incurred by Docker is pretty small when it comes to CPU and memory, but for heavy IO applications you need to be a little more careful. Depending on the workload, I would put Kafka in the heavy IO group, so this is probably not a problem. Kafka benefits a lot from fast disk access, so if you run your containers in some kind of distributed platform with storage attached to a SAN or NFS, or something like that, I would suggest that you notice the difference. But if you only chose containers to facilitate deployment and run them on the same physical machine, I would suggest that the difference would be negligible.
, , , , .