Replica Consistency Management Stratege Of Metadata Cluster For Key-value Stores

In order to improve the efficiency of data access, most people adopt theorganization management to separate file and metadata in modern storage system.However, to design a high performance and high reliability metadata cluster indistributed file system, needs to solve an important problem is replica consistency.Traditional consistency maintenance methods commonly used to ensure that strongconsistency, but strong consistency is easy to impact system performance. Introducingthe key-value storage system, because of RDBS lack of load balance, data availabilityand query performance. The key-value system can effectively meet the metadatainformation of high concurrency, high performance access requirements.To this end, this paper proposes a replica consistency management strategy ofmetadata cluster for key-value stores, and elaborate the maintain method in theaspects of replicas placement, read and write, restore and update, eventually achieve aflexible and user configurable replicas consistency constraint. Firstly, the paperintroduces the method by using consistent hash to locate the metadata quickly in complexcluster network. Because of the key-value system’s characteristics, it puts forward the flexibleand reliable read-write strategy. Then it introduces how to recover the replications when somenodes are down in the cluster to improve the robustness of the cluster. Moreover, it describesthe update propagation and the synchronous detection to ensure eventual consistency of thedata in detail. Finally through the read-write tests to verify the cluster efficient performance,and to ensure the system can still provide the reliable service under the high load condition bythe stress test of high concurrent access.