Research And Implementation Of Failover Technology On Database Cluster

With the increasing requirement of critical and continuous task, high availability of database management system is paid more attention by industry. Data redundancy and cluster is the first choice to achieve high availability, but failover efficiency of cluster system affect the system's high availability. Therefore, to achieve one cluster system of high failover efficiency has high practical value. the failover model Pronto _PF focused on transaction integrity based on MySQL is implemented.Thesis detailed analyze shared nothing and shared storage architecture, the former has high failover efficiency and scalable performance, but the fault handling capacity is weak. However, shared storage architecture has strong capacity of fault handling and less efficient failover. And then introduce the failure predication mechanism to promote the efficiency of the failover, and prove the correctness and rationality of the failover algorithm based on predication. And introduced the transaction ID attribute for failover, designed and implemented the extended event structure in the binary log file: HA_log_event. Finally based on failure predication and the extended event structure we building a failover model Pronto_PF, which mainly consists of three modules, the node switch module convene the election agreement during the node failure predication and select the appropriate coordinator according to transaction ID attribute; the transaction read/write module introduce the concept of the transaction information table and record the implementation information of transaction in the system, and determine whether the loss of the node affairs according to the analysis and comparison of the information in the transaction information table and binary log file in the node; the transaction analysis module realize the transaction integrity according to the transaction read/write module by undoing and redoing the transaction.In this paper, we construct the failover model Pronto _PF based on the extend log, then analyze and implement the three modules in accordance with its architecture, which applied the ROWA-A strategy, heartbeat detection strategy and redo logs strategy. Finally, Pronto_PF's performance evaluation and comparison with existing high availability cluster show that the failover efficiency and transaction integrality capabilities of the model meet the high availability requirements.