Cluster Fault Convergence Method and Device Based on Fault Causality Graph

Abstract

The invention provides a fault causality relationship graph-based cluster fault convergence method and a device. The method comprises the steps of acquiring the information of a cluster fault case, and establishing a fault causality relationship graph based on the information of the cluster fault case; during the occurrence of a cluster fault, judging whether the fault symptoms of the cluster fault exist in the fault causality relationship graph or not; if the fault symptoms of the cluster fault exist in the fault causality relationship graph, figuring out the primary fault of the current cluster fault based on the fault causality relationship graph and the dependency relationship of the fault symptoms in the fault causality relationship graph, and repairing the cluster fault according to the repair method of the primary fault; if the fault symptoms of the cluster fault do not exist in the fault causality relationship graph, acquiring the information of the fault case of the current cluster fault after repairing the current cluster fault, and adding the information of the fault case of the current cluster fault in the fault causality relationship graph. Based on the above method and the above device, the experiences in handling with cluster faults can be accumulated and migrated, and the requirement on the abilities of the operation and maintenance personnel is lowered. Meanwhile, the operation for eliminating cluster faults is more targeted.

Description

technical field [0001] The invention relates to computer cluster fault processing technology, in particular to a cluster fault convergence method and device based on a fault causal graph. Background technique [0002] Cluster, short for cluster communication system, is a computer system that is connected through a group of loosely integrated computer software and / or hardware to work closely together to complete computing tasks. The development trend of modern IT system clusters is multi-component (one component is a single software system or hardware device in the cluster), large-scale distributed system, and its scale and complexity are increasing day by day. This has brought great challenges to the operation and maintenance work. The main difficulties lie in the following two points: [0003] (1) A cluster is composed of multiple components, and each component undertakes specific and subdivided functions and plays different roles in the cluster; a component failure will c...

AI Technical Summary

Problems solved by technology

(2) There is a strong dependency between components, and the failure of one underlying component may spread to other components, which may appear as a failure of the upper component

[0006] However, the above method of snapshot difference also has shortcomings: (1) The correlation between the difference of the cluster snapshot and the cause of the failure is not high

During the period between two snapshots, many changes may occur in the cluster. The snapshot comparison method can find out these changes, but these changes cannot accurately reflect the root cause of the fault. At the same time, the snapshot difference information can only be used as a reference for the operation and maintenance personnel to deal with the fault. Reference, but cannot point out the fault with certainty, let alone give suggestions on detection and repair methods

(2) The historical experience of group fault handling has not been effectively accumulated

Analyzing and handling faults by means of snapshot differences, experience in dealing with new fault types cannot be accumulated in the system, and previous experience cannot be automatically applied to past fault types

Unable to solve the problem of cluster operation and maintenance requiring high experience of operation and maintenance personnel

Images