Case Study: Zhihu boosts big data performance with PingCAP TiDB

Horizontally Scaling the Hive Metastore Database by Migrating from MySQL to TiDB

Zhihu, China’s largest knowledge sharing platform, needed to improve the performance and scalability of its big data infrastructure as query volumes grew. The company was struggling with slow Hive ad-hoc queries, heavy pressure on HDFS, and NameNode RPC bottlenecks, and turned to PingCAP’s TiDB to help manage the massive metadata workload.

PingCAP’s TiDB was used to replace MySQL for Apache Hive Metastore and to store and index HDFS metadata, while also powering a real-time alerting system for large Hive queries and a cache for HDFS ContentSummary in the Router layer. The results were dramatic: large query execution time dropped from 30–40 seconds to 6–7 seconds, and ordinary `getContentSummary` latency fell from several minutes to less than 10 ms, significantly easing pressure on NameNode and improving overall cluster stability.

Zhihu

Mengyu Hu

Platform Engineer