For the distributed version of Milvus 2.0.2, I tested to increase the number of querynodes online. It seems that the data will not be load balanced, and the memory usage of newly started nodes is very low. It will be balanced after restarting the service. What is the reason for this?
How much data do you have? each 512M data is a segment, and it only balanced by segment level.
If you have smaller data amount and you want to scale your qps, memory replica is exactly you are looking for and we will release on 2.1