NoSQL databases: Consistency of data

NoSQL database: Consistent read consistency for data

• Strong consistency
  Access to any node in the cluster at any time, resulting in consistent data results;

• User consistency
  The same user, access to the cluster during the data consistent;
  Resolve user consistency: Use sticky sessions to bind sessions to specific nodes for processing;
  This reduces the performance of the load balancer;

• Final consistency
  The data synchronization between the nodes in the cluster is inconsistent with the temporary data, but after the data synchronization is complete, the final consistency is achieved.

Update consistency

• Pessimistic Way

  使用写锁大幅降低系统响应能力可能导致死锁

• Optimistic Way

  先让冲突发生，再检测顺序自动合并的处理方式极具“领域特定”问题

Easing "consistency constraints"

• Cap theorem
  Consistency (consistency), availability (availability), and partition tolerance (Partition tolerance), 3 properties can only meet 2 at a time;

Explanation of partition tolerance: clusters are still available if the cluster is divided into multiple communications failures

• CA system

    单服务器  集群中出现”分区“，就不可用

• Pa/pc

    集群出现”分区“时，需要在”一致性“ 和“可用性”间权衡  一般会牺牲部分一致性（eg：使用最终一致性），保证可用性

Relaxing "persistence" constraints

More stringent durability means more performance loss;

• Sacrificing "persistence" for better performance
• Replication "Persistence" failure

    主节点故障，未同步到从节点的数据丢失  主节点恢复，故障期间更新的数据冲突  解决方案：针对单个请求指定其所需的持久性

Attached Mind Map

Reference

The essence of NoSQL

Posted by: Big CC | 02jul,2014
Blog: blog.me115.com [Subscribe]
Weibo: Sina Weibo