High-availability optimization method of memory computing system in combination with principal-subordinate backup and erasure codes

Abstract

The invention provides a high-availability optimization method of a memory computing system in combination with principal-subordinate backup and erasure codes, which comprises the steps that a distributed system is divided into multiple groups, M principal nodes and N subordinate nodes are protected by each of the groups, and values of M and N exceed 0; memory on each of the principal nodes comprises an area A and an area B, wherein the area A is used to save metadata, the area B is used to save value data, the metadata comprises indexes of key-to-value data addresses and memory distribution information, and the value data comprises value data corresponding to all the keys on the node; the memory of each of the subordinate nodes comprises an area C and an area D; the area C is used to save and backup the metadata, and the area D is used to save codes of the value data; the value data is protected by introducing an update operation of the erasure codes; and when the principal nodes are down, any of the subordinate nodes will constitute a recovery node to take the place of the principal node in order to deal with a request, and lost data is recovered by an online recovery protocol.

Description

technical field

[0001] The invention relates to the technical field of high-availability systems, in particular to a high-availability optimization method for a memory computing system combined with master-slave backup and erasure codes. Background technique

[0002] Due to its volatility, in-memory computing systems such as in-memory key-value systems usually need to use master-slave data backup to prevent data loss and service unavailability caused by downtime, such as Redis. The method of implementing a high-availability memory key-value system by using master-slave data backup mainly divides the server cluster into multiple groups, where each group includes a master node and one or more slave nodes. The master node provides key value update requests and key value query requests, and the slave nodes save the same memory information as the master node, and only receive update requests sent by the master node. When the master server fails, the slave node can become the new...

Images