Liu Qi

The Scaling Problem Nobody Wanted to Solve

Somewhere inside Wandoujia's data center, around 2014, Liu Qi was managing one of the world's largest Redis clusters - and watching it buckle. Not from bad engineering. From a limit baked into the architecture of traditional databases: they weren't built for the kind of horizontal scale that China's internet companies were hitting simultaneously. You could shard, you could hack, you could hold your breath. But the problem didn't go away.

Most engineers route around this kind of obstacle. Liu Qi, along with colleagues Ed Huang and Dylan Cui, decided to route through it. In April 2015, the three of them founded PingCAP in Beijing with a single conviction: the right response to an impossible trade-off is to stop accepting the trade-off.

The result was TiDB - an open-source, distributed SQL database with MySQL compatibility, HTAP architecture, and the kind of horizontal scalability that makes the sharding nightmare disappear. Ten years later, it runs the most demanding data workloads at Pinterest, Plaid, Bolt, Bank of China, JD.com, and 1,500+ organizations worldwide.

Assembly Language at 20, Distributed SQL at 30

Liu Qi started writing x86 assembly code in his college junior year - getting into software encryption and decryption when most students were still learning loops. That early instinct for low-level systems thinking would define the rest of his career. At JD.com, China's largest B2C retailer by transaction volume, he worked deep in infrastructure. At Wandou Labs, he operated at the scale where most databases begin to lie to you about their limits.

His GitHub username is "ngaut" - display name: goroutine. A Go concurrency primitive. Not a product name, not a personal brand. Just a clue about what the man finds interesting. He's been contributing to open source since 2011, long before "open-source CEO" became a career arc anyone could plan.

The co-authorship of Codis, a proxy-based Redis cluster solution written in Go, came out of that same Wandoujia period. Codis solved the Redis scaling problem in production. TiDB, conceived around the same time, attacked the deeper question: what does a database look like if you design it for horizontal scale from the start, not as an afterthought?

Open Source as a Business Model - and a Philosophy

The conventional wisdom in enterprise software is that you open-source the edges and keep the core proprietary. Liu Qi inverted this. TiDB is fully open-source. TiKV, TiDB's distributed key-value storage layer written in Rust, was donated to the Cloud Native Computing Foundation and graduated as a CNCF project - the infrastructure equivalent of a degree from a credible institution, earned in public.

The reason isn't idealism. It's trust engineering. "Open source cracked the trust code for international expansion," is how his team describes it. When a Japanese payment company discovered TiDB on GitHub during an AWS Aurora crisis - found it, evaluated the source directly, deployed it, and made it a market standard in Japan - that wasn't marketing. That was the architecture of openness doing its job.

The Architecture of TiDB: Why HTAP Matters

The standard story in data infrastructure is that you keep your transactional database and your analytical database separate. Two systems, two teams, two maintenance burdens, and a replication lag between them that everyone pretends doesn't matter until it does.

TiDB's HTAP (Hybrid Transactional/Analytical Processing) architecture is a direct argument against this arrangement. TiFlash, TiDB's columnar storage engine, handles analytical queries against live transactional data - without moving data to a separate system, without batch ETL, without the wait. The design reflects Liu Qi's engineering instinct: if the problem is a forced choice between two things you need, stop accepting the premise.

The storage layer, TiKV, was written in Rust rather than C++ - a deliberate choice to avoid the memory safety issues that haunt long-running infrastructure software. The distributed SQL layer handles MySQL compatibility so that migrating doesn't require rewriting applications. The Kubernetes operator handles deployment, scaling, and failover so that teams don't need a dedicated DBA to manage production clusters.

Three Open-Source Projects, One Career Arc

Liu Qi's technical identity is anchored in three open-source projects. Codis, the Redis cluster proxy, solved a real problem fast and became widely adopted by Chinese internet companies dealing with Redis at scale. TiKV, the distributed key-value store, became infrastructure that other projects - not just TiDB - could build on, and earned CNCF graduation as a result. TiDB, the distributed SQL layer, assembled these ideas into a complete system.

Each project followed the same pattern: identify a real operational pain, build in the open, iterate against production workloads, let the community stress-test it. This isn't the research lab model or the stealth startup model. It's something closer to building a product and running a long-term public experiment simultaneously.

In 2025, PingCAP recorded $13.1 million in annual revenue with a 430-person team - numbers that reflect a company still investing heavily in product and community while extracting commercial value from its cloud and enterprise offerings. By 2026, employee count had grown to 600+.

Building Across Borders

Liu Qi operates between Beijing and Sunnyvale - a geography that reflects PingCAP's ambition to be genuinely global, not just internationally available. The engineering hubs span US, Europe, Japan, and Southeast Asia. The open-source community that builds TiDB is distributed by design - 1,000+ contributors across time zones, none of whom needed a visa to participate.

The Series D in November 2020 - $270 million led by GGV Capital, Access Technology Ventures, and a consortium of investors - gave PingCAP the runway to build TiDB Cloud as a fully managed service, enabling companies to run distributed SQL without managing the infrastructure underneath it. The pitch to enterprises: the database complexity is real, but you don't have to own it.

In 2024, PingCAP hosted the third annual HTAP Summit at the Computer History Museum in Mountain View - a choice of venue that says something about how Liu Qi frames the work. Not a sales conference. A gathering about the history and future of data infrastructure, held in a building that exists to document where computing came from.

Principles, Not Slogans

Liu Qi has articulated the principles that guide TiDB's development in writing. They're not abstract. They describe specific decisions that cost something to make - pausing OLAP development when OLTP users reported problems, for instance, or identifying root causes rather than shipping feature requests that only addressed symptoms. The principle of "make it work, make it right, make it fast" sounds familiar because it's true, but PingCAP applies it with the patience that most software companies run out of somewhere between series A and series B.

Three Projects. One Engineer. Real Production.

Under the Hood