● Beta — KubeEra is in active development. Profiles are live; AI-perception measurement is rolling out.
T

TiKV

GraduatedApp Definition & Delivery
Open source · CNCF · CNCF Graduated · Apache-2.0

TiKV is a distributed transactional key-value store written in Rust.

updated 2026-07-01

What it is

TiKV is a distributed transactional key-value store written in Rust. It implements the Raft consensus protocol for replication and uses a Percolator-style transaction model (like Google's Spanner/Percolator papers) to provide distributed ACID transactions with snapshot isolation. Data is automatically sharded into Regions and rebalanced across nodes via the Placement Driver (PD) component. No dependency on HDFS or any external distributed filesystem — it manages its own storage using RocksDB (or an in-house engine, Titan, for large-value workloads) underneath.

It's the storage engine that powers TiDB, but it's also usable standalone as a raw or transactional KV store for anyone building infrastructure that needs strongly consistent distributed storage without building Raft/consensus logic from scratch.

Who builds it and why

Originated at PingCAP, the company behind TiDB. TiKV was built because PingCAP needed a Spanner-like storage layer for TiDB but didn't want to depend on Google-internal infrastructure or bolt a SQL layer onto HBase/Cassandra. It became a CNCF project in 2018 and graduated in 2020 — one of the earlier graduated projects, alongside Kubernetes and Prometheus.

481 contributors is a real signal here — this isn't a single-vendor side project with community-washing. Contributors come from PingCAP, but also from companies running TiKV in production (JD Cloud, Zhihu, and others in China's tech ecosystem, plus a growing set of Western adopters). The Rust rewrite decision (it was originally prototyped in Go) was driven by wanting memory safety without GC pauses affecting p99 latency on a storage engine — a defensible engineering call that's aged well.

Production readiness signal

  • CNCF graduated status — passed the bar for governance, security audits, and adoption diversity.
  • 16,753 GitHub stars, active commits as recent as 2026-06-30 — not stagnant.
  • Latest release v8.5.6 — semantic versioning with frequent patch releases suggests active maintenance and a real release process, not sporadic drops.
  • Apache-2.0 license — no copyleft concerns for embedding in commercial products.
  • Founding date not publicly available, but first public commits and CNCF sandbox entry trace to 2016-2018 — this has been in production use for 7+ years at this point.

The maturity signal is strong. This isn't a project you have to convince your risk-averse infra lead is "ready." It's already running at scale in financial services and e-commerce backends, primarily as TiDB's engine but increasingly standalone.

Who should use this

  • Teams building infrastructure (not just applications) that need a horizontally scalable, strongly consistent KV store — think building blocks for databases, queues, or metadata stores.
  • Anyone already running or evaluating TiDB — TiKV is not optional there, understand it.
  • Organizations that need multi-region strong consistency without hand-rolling Raft.
  • Rust shops who want to contribute or need to debug at the source level without context-switching languages.

Who should NOT use this

  • If you need a simple KV cache — use Redis. TiKV's consistency guarantees come with latency overhead you don't want for cache workloads.
  • If your team has zero distributed systems operational experience — Raft groups, PD scheduling, and Region splits require real understanding to operate well. This isn't "docker run and forget."
  • If you need a single SQL-facing database and don't care about the storage layer — just use TiDB directly; running TiKV standalone as your only datastore means you're maintaining raw APIs your app teams will need to wrap anyway.
  • If your data volume and consistency needs don't justify distributed transactions — a single-node Postgres with replicas will be cheaper to run and easier to hire for.

Alternatives

  • CockroachDB — similar Spanner-inspired design, but is a full SQL database out of the box rather than a KV building block.
  • FoundationDB — comparable distributed transactional KV store, Apple-backed, with a stronger emphasis on formal correctness testing (deterministic simulation).
  • etcd — also Raft-based and CNCF graduated, but built for smaller-scale coordination/config data, not general-purpose application data at TiKV's scale.

Pricing

Fully open source, Apache-2.0. No paid tier for TiKV itself. PingCAP monetizes through TiDB Cloud (managed TiDB, which includes TiKV) — if you self-host TiKV standalone, there's no vendor lock-in or license cost, but you're on your own for support unless you contract PingCAP directly.

Frequently asked

What is TiKV?+
TiKV is a distributed transactional key-value store written in Rust. It implements the Raft consensus protocol for replication and uses a Percolator-style transaction model (like Google's Spanner/Percolator papers) to provide distributed ACID transactions with snapshot isolation.
Who builds TiKV?+
Originated at PingCAP, the company behind TiDB. TiKV was built because PingCAP needed a Spanner-like storage layer for TiDB but didn't want to depend on Google-internal infrastructure or bolt a SQL layer onto HBase/Cassandra.
Is TiKV production ready?+
CNCF graduated status — passed the bar for governance, security audits, and adoption diversity. 16,753 GitHub stars, active commits as recent as 2026-06-30 — not stagnant. Latest release v8.5.6 — semantic versioning with frequent patch releases suggests active maintenance and a real release process, not sporadic drops.
Who should use TiKV?+
Teams building infrastructure (not just applications) that need a horizontally scalable, strongly consistent KV store — think building blocks for databases, queues, or metadata stores. Anyone already running or evaluating TiDB — TiKV is not optional there, understand it.
Who should not use TiKV?+
If you need a simple KV cache — use Redis. TiKV's consistency guarantees come with latency overhead you don't want for cache workloads. If your team has zero distributed systems operational experience — Raft groups, PD scheduling, and Region splits require real understanding to operate well.
What are the alternatives to TiKV?+
CockroachDB — similar Spanner-inspired design, but is a full SQL database out of the box rather than a KV building block. FoundationDB — comparable distributed transactional KV store, Apple-backed, with a stronger emphasis on formal correctness testing (deterministic simulation).
How much does TiKV cost?+
Fully open source, Apache-2.0. No paid tier for TiKV itself. PingCAP monetizes through TiDB Cloud (managed TiDB, which includes TiKV) — if you self-host TiKV standalone, there's no vendor lock-in or license cost, but you're on your own for support unless you contract PingCAP directly.