The basic structure of watercress

This article according to the Infoq Chinese station to the Watercress Hong Qiangning (@hongqn) the communication Exchange reorganization. Hong Qiangning introduced the structure and components of watercress, and shared some of the team experiences of the Watercress Foundation Platform Department. The screenshot is from Hong Qiangning's share of the CTO club in 2013.

Guest Introduction

Hong Qiangning, chief architect of Watercress. The first full-time employee in watercress. After graduating from Tsinghua University, Hong Qiangning has been doing embedded systems. began contacting the Python language in 2002, turning from a hardware engineer into a software engineer, with a deep understanding of how a language works at the bottom of the computer.

Architecture

Watercress the entire infrastructure can be roughly divided into online and offline two chunks. The online section is similar to most Web sites: Before using the LVs to do ha, using Nginx to do the reverse agent, forming a load-balanced layer; Application layer is mainly doing operations, the results returned to the previous users, Dae platform is the two years of construction, now most of the application of watercress are basically running in the Dae above the Application of the basic services are similar to other sites, MySQL, memcached, Redis, BEANSTALKD, not the same is the NoSQL choice--beansdb, this is our open source in a few years ago, the KV database, but also the domestic relatively early open source of the KV database.

The BEANSDB project can be said to be a simplified version of the AWS DynamoDB, which was launched in 2008 and 2009 Open source, with the Tokyo cabinet as the storage engine, and 2010, with the Bitcask storage format rewriting the storage engine for better performance. Beansdb a hash of the key to find the node to achieve distribution and redundancy, a write operation will write several nodes, and now the configuration is to write three read a copy. The main feature of BEANSDB is the support of mass KV databases--compared to Redis, which supports dozens of g to hundreds of g of memory kv databases, BEANSDB can support up to hundreds of T data. In addition BEANSDB the biggest benefit is the operation dimension is very simple, the performance, the usability, the expansion is very good, also achieved the final consistency.