Inside Pika: How 360’s Large‑Scale Redis‑Compatible Store Powers Billions of Requests

Background Introduction

On November 26, 2017, at 13:30, the first Pika community salon was held in the 360 A‑building conference hall near the 798 Art Factory in Beijing.

What Is Pika

Pika is a large‑capacity Redis‑compatible storage jointly developed by 360 DBA and the infrastructure team. It is not meant to replace Redis but to complement it, offering persistent storage to address Redis limitations in massive‑scale scenarios such as slow recovery, costly master‑slave sync, single‑thread bottlenecks, limited data capacity, and high memory costs.

Within 360, Pika serves almost all business lines, handling 100 billion daily requests and storing up to 30 TB (compressed), roughly equivalent to 100 TB of Redis memory data, saving substantial server costs. Around 30 companies have deployed it at scale in the community.

Technical Resources

GitHub: https://github.com/Qihoo360/pika

Technical discussion group: 294254078

Related articles (original Chinese titles): "All About Large‑Capacity Redis‑Like Storage – Pika", "Pika Master‑Slave Replication Workflow", "Pika Binlog Details", "Pika 2.2 Official Release", "Hundred‑Billion‑Level High‑Performance KV Storage Ecosystem".

Event Review

Despite cold weather, nearly 150 participants attended, far exceeding the expected small‑scale gathering.

The salon aimed to give attendees a comprehensive, objective understanding of Pika, addressing misconceptions and showcasing its real‑world capabilities.

Topic 1: Pika Technology Review and Outlook

Speaker: Song Zhao, core developer from 360WEB Platform Infrastructure.

He covered Pika’s inception, the 1.0 release, open‑source journey, iterative improvements up to version 2.2.6, engine breakthroughs, operational commands, current challenges, and future directions.

He announced two eagerly awaited features: the distributed sync tool pika_hub supporting dual‑master and multi‑point synchronization.

Topic 2: Scaling Pika for Huanxin’s Massive Instant‑Messaging Cloud Platform

Speaker: Zhang Yunqian, senior performance engineer at Huanxin.

He explained Huanxin’s KV storage needs, why Pika was chosen, and how it reduced storage costs while improving stability.

As data grew, Huanxin augmented Pika with Codis for further scaling, employing sync/async migrations and rapid slot migration to enhance expansion capabilities, and plans to explore a Codis + Pika + Ceph stack.

Topic 3: Pika in Weibo’s Architecture

Speaker: Lan Jiangzhou, DBA from Sina Weibo.

He demonstrated Weibo’s migration of nearly 40 TB from Redis to Pika, saving the equivalent of ~300 servers.

The talk highlighted challenges encountered, solutions, and constructive feedback for the Pika team.

Topic 4: Billion‑Scale Access of Pika Inside 360

Speaker: Zhang Heng, DBA and product manager for Pika at 360WEB.

He presented internal usage statistics: by 2017, 360 operated 4,000 Redis/Pika instances but only modestly increased server count, saving roughly ¥8 million in hardware costs.

He emphasized Pika’s disk‑based, highly compressed storage model that leverages idle disk space without impacting existing Redis workloads, and illustrated suitable and unsuitable scenarios.

Prizes and Closing

Attendees received books, custom mouse pads, and other gifts.

The 360 Technology Training Camp’s seventh session concluded successfully, with plans for future events.

Additional Resources

Speaker slide download: https://eyun.360.cn/surl_ywspCbYbYiX (extraction code: b319)

Event video: http://www.itdks.com/eventlist/detail/1693