Design Of HA Solution For Virtual Machine Based On OpenStack

The virtual machine is the key to the flexible allocation of computing resources by the cloud platform.The management of virtual machines is also one of the core services of OpenStack.Achieving high availability of virtual machines is of great significance to ensure the stability of cloud platform services.However,due to the design concept of the OpenStack community,the OpenStack cloud platform lacks such an important feature.Therefore,many enterprises in the industry have proposed their own solutions for the HA characteristics of virtual machines,but there are more or less defects.Therefore,this topic analyzes the design demands of virtual machine HA,and proposes a sophisticated virtual machine HA architecture based on the OpenStack cloud platform.This architecture can solve the problems that the existing solutions in the industry do not solve,and add virtual machine HA features to the cloud platform,and ensure the availability of cloud platform virtual machines.In this thesis,the core of the virtual machine HA architecture is the efficient monitoring of the compute nodes in the cluster and the fencing mechanism that can effectively prevent brain splitting.This solution is based on the clustermaker+corosync for cluster management.It is automatically detected by the cluster resource monitoring + OCF script for different virtual machine exception scenarios,and automatically restored according to the preset instruction set when the abnormal state is monitored.The fencing mechanism uses a reliable power fence device as a resource to be added to the cluster resource management for supervision.When the fault resource needs to be isolated,the node state is checked by the fence script.When fencing is required,the node’s evacuation attribute is set to True.Then,the STONITHd module sends a fence command to the fence device,and finally fence the fault resource by means of power isolation;the virtual machine recovery mechanism filters the compute node according to the recovery command issued by the cluster management software or the fence mechanism,and then invokes the nova command to restore the virtual machines on the failed node;then combines the metadata service in OpenStack to avoid full recovery of the virtual machine.Finally,the deployment test is performed according to the overall architecture diagram of the virtual machine,and the test result conforms to the expected result.Therefore,the solution is feasible.