TiKV — Distributed Transactional Key-Value Store on Raft

What it is

TiKV is a distributed transactional key-value store written in Rust. It is a CNCF-graduated project that powers TiDB (a distributed SQL database). TiKV provides horizontal scaling, strong consistency via the Raft consensus protocol, geo-replication, and ACID transactions with Percolator-style MVCC.

It targets infrastructure teams building distributed systems that need a reliable storage layer with transactional guarantees. TiKV runs as a standalone key-value store or as the storage engine behind TiDB.

How it saves time or tokens

Building a distributed, strongly consistent storage layer from scratch is a multi-year effort. TiKV provides this as a ready-to-deploy component. The Raft consensus protocol handles leader election and data replication automatically, and the Percolator transaction model provides distributed ACID without manual coordination.

TiKV's RawKV API is simple enough for key-value workloads, while the TxnKV API supports multi-key transactions across shards.

How to use

1. Install TiUP (PingCAP's cluster manager):

curl --proto '=https' --tlsv1.2 -sSf https://tiup-mirrors.pingcap.com/install.sh | sh

1. Start a local TiKV cluster for testing:

tiup playground --mode tikv-slim

1. Interact with TiKV using the Rust, Go, Java, or Python client:

from tikv_client import RawClient

client = RawClient.connect(['127.0.0.1:2379'])
client.put(b'key1', b'value1')
value = client.get(b'key1')
print(value)  # b'value1'

Example

// Go client example with transactions
package main

import (
    "context"
    "github.com/tikv/client-go/v2/txnkv"
)

func main() {
    client, _ := txnkv.NewClient([]string{"127.0.0.1:2379"})
    txn, _ := client.Begin()

    txn.Set([]byte("account:alice"), []byte("1000"))
    txn.Set([]byte("account:bob"), []byte("500"))

    err := txn.Commit(context.Background())
    // Both writes are atomic -- either both succeed or neither does
}

Related on TokRepo

• AI Tools for Database -- Distributed database tools and storage engines
• AI Tools for DevOps -- Infrastructure tools for distributed systems

Common pitfalls

• TiKV requires PD (Placement Driver) for cluster coordination. Running TiKV without PD is not supported in production.
• Raft consensus requires an odd number of replicas (3, 5, 7) for proper leader election. A 2-node cluster cannot tolerate any node failure.
• The Percolator transaction model adds latency compared to single-node databases. Expect 2-5ms per transaction in a well-configured cluster.