Getting top 10 results from many storage nodes.

[amano-kenji]

Let's assume that there are 3,000 storage nodes. I want to retrieve the 10 largest values from a TiDB server which has to query 3,000 storage nodes.

In this case, maintaining 3,000 connections would be challenging.

• Can it be done efficiently?
• Perhaps, I should assign 300 storage nodes to each TiDB server and aggregate local top-10 results from TiDB servers at the application level? In this case, I would have to design a database schema that is entirely contained within one TiDB server.