We have a 3 node data cluster that contains index with 3 primary shards and 2 replica
Initially, each data node had 1 primary shard but when we had to replace one of the data nodes recently, all the primary shards are now on a single data node. The other 2 data nodes contain replicas only.
Is there a performance (read and write both) issue with the latter when compared to former?
I am reading conflicting information and not sure what is correct.
If I want to redistribute primary shards, how do I do it?