For our use case, the Milvus Standalone edition looks to be about right (< 1M vectors). However, we would want to make use of the alleged H/A feature, an old master / slave setup, where data is replicated from the master to the other instances. Should the master fail, then a slave is elected as the new master. Or at least this is my understanding. It would be offering only redundancy and not additional read capacity (although it would be excellent if we could send pure read requests to the redundant instances).