History

Inspired by Google Spanner, PingCAP started developing TiKV in 2015, and released the first version of TiKV along with TiDB in 2016. Up to April 27, 2018, PingCAP has released TiDB/TiKV 2.0, which has a lot new features and great performance gains compared to 1.0.

Concurrency Control[03]

TiKV has a Timestamp Oracle(TSO) to provide globally unique timestamp. The core transaction model of TiKV is called 2-Phase Commit powered by MVCC. There are two stages within each transaction: - PreWrite: - Create a startTS timestamp. Select one row as the primary row and others as secondary rows. - Check whether there are locks on this row or whether there are commits after the startTS. If conflicts exists, the transaction will be rollback. If not, lock the row. - Repeat the second step on other rows. - Commit: - Write to the CF_WRITE with current timestamp commitTS. - Release all the locks.

Data Model

Query Execution

Query Interface[04]

TiKV support queries such as simple Key-Value, transactional Key-Value and push-down. But no matter it’s transactional Key-Value or push-down, it will be transformed to simple Key-Value operations in TiKV.

Storage Model[05]

TiKV uses RocksDB as it underlying storage layer. Data is stored in RocksDB within each TiKV Store.

Stored Procedures[06]

System Architecture[07][08]

TiKV is built on top of RocksDB, where all data in a TiKV node shares two RocksDB instances. One is for data, and the other is for Raft log. There are some major components in TiKV:

• Placement Driver (PD): Manages the metadata about Nodes, Stores, Regions mapping, and makes decisions for data placement and load balancing.
• Node: A physical node in the cluster. Each node contains one or more Stores.
• Store: Stores data in local disks using RocksDB. Each store contains one or more regions.
• Region: The basic unit of Key-Value data movement and corresponds to a data range in a Store. Each Region is replicated to multiple Nodes and form a Raft group. A replica of a Region is called a Peer.

Views[06]