Explain TiKV and PD terminologies.
This document explains some important terminologies of TiKV and PD.
A TiKV Node is a physical node in the cluster, which might be a virtual machine, a container, etc. Within each Node, there can be one or more stores.
The Node, Store, and Region regularly report their status to the Placement Driver.
Placement Driver (PD)
The Placement Driver (PD) is the cluster manager of TiKV. It periodically records the cluster information, makes decisions to move/split/merge TiKV Regions across nodes according to the application workload and storage capacities. This process is called scheduling.
A replica of a Region is called a peer. Multiple peers of the same Region replicate data via the Raft consensus algorithm.
Data is distributed across TiKV instances via the Raft consensus algorithm, which is based on the Raft paper (“In Search of an Understandable Consensus Algorithm”) from Diego Ongaro and John Ousterhout.
TiKV shards continuous ranges of keys into Regions, and replicates Regions via the Raft protocol. When data size increases until a threshold is reached, a Region will be split into multiple ones. Conversely, if the size of the Region shrinks due to data deletion, two adjacent Regions can be merged into one.
A Store is an instance of a TiKV server in which multiple peers are stored.
TiKV provides transactions that allow you to read and write across any keys with Snapshot Isolation regardless of the physical placement of the Region. TiKV also provides the pessimistic transactions that are semantically similar to
SELECT ... FOR UPDATE in SQL.
The underlying transaction model is similar to Google’s Percolator, a system built for processing updates to large data sets.