Balance Scheduling

The Scheduler selects or filters the stores to be scheduled through Selector/Filter, and generates Operators according to the specific scheduling content. All Operators will be handed over to OperatorController for unified speed limit and control. If you are not familiar with these concepts, I suggest you read the article mentioned above first.

In this article, let's take a look at how PD balances the distribution of Regions through scheduling. There are many implementations of Scheduler. Among them, the scheduler responsible for region balance is called region-balance-scheduler. Its main function is to dynamically add and delete nodes as the TiKV cluster dynamically merges or splits regions, which leads to unbalanced region distribution. Algorithm to migrate Region from nodes with more data to nodes with less data, so as to realize the rebalancing of Region storage.

The article first introduces the specific structure of balance-region-scheduler in the source code, then sorts out the core process of its scheduling algorithm, and introduces in detail the key execution path of each process or the important mechanism that affects scheduling.