TiDB Configuration Myth

A running TiDB cluster is composed of many running processes of different role: PD/TiKV/TiDB, and many features’ behavior depend on the specific configuration.

Currently there are a whole lot of things related to configuration: config file, global/instance/session variables, which really makes people crazy. Let alone multiplied with the number of components.

Could someone please provide:

  • a full list of methods to configure a TiDB cluster
  • design purpose of each method
  • interaction between them

And if it turns out that the current design is misleading and hard to use, maybe it’s time for a top-down re-design.

Thanks a lot!