how many zookeeper servers need to serve ~100 Kafka machines

[#1]

for now we have 15 Kafka machines in the cluster ,

all machines are are installed with RHEL 7.9 and the HW machine is DELL physical machine

Kafka version is 2.7 , and we have 3 zookeeper servers that serve the Kafka cluster

we decided to extend the Kafka cluster to ~100 machines , because Total Throughput In Megabytes increased dramatic

- note  according to Kafka confluent calculator we need around 100 Kafka machines

in that case I am wonder if our 3 zookeeper servers are enough to serve this huge cluster machines?

• addition I want to say that our 3 zookeeper servers are already serve other application as HDFS , YARN , HIVE , spark etc.

[#2]

Hello Mike

Likely your existing 3 Zookeeper nodes can serve your expansion requirements

You can monitor the CPU and network of the Zookeeper nodes when your Kafka cluster is growing, when reaching the throughput limit, you can expand your zookeeper to 5 nodes

Remember the zookeeper nodes need to keep in sync all the time, so the more zookeeper nodes the more traffic will be added to keep them in sync, while those nodes handling the Kafka requests; so it doesn't mean the more the better

I would suggest to stay with 3 zookeeper nodes while expanding your kafka cluster with close monitoring, and consider to grow to 5 when the CPU/network throughput reaching the limit

You can also consider to tune the zookeeper nodes e.g. dedicated disks, better network throughput, isolate zookeeper process, disable swaps