Design And Implementation Of High Availability Programme On OpenStack Cluster

With the continuous development of cloud computing technology,users can use computing resource like water,electric and so on.To make an easier management of the resources in the resource pool which contains lots of computing,storage and networking resource,the cloud operating system OpenStack was born.In the field of finance,politics,servers have to process computing and storage business with huge important data.If the server crash down,it will lead to disaster and make huge losses.When some situation like server device breakdown,system crash or unexpected power-off happened,the less time it spend to recover from failure the better for the users.That means we should make the availability of the cluster as high as possible.But OpenStack do not have high availability with itself.So if we want to take use of the convenience of OpenStack,we should develop high availability on it.By analyzing the structure of high availability cluster,we make some improvements on the very popular method that use corosync and pacemaker to build a high availability cluster.To solve the problem that when there are lots of nodes in cluster corosync will spend pretty long time to restrain to a ring after node join or leave the cluster.And to decrease the probability of erroneous judgment,heartbeat detection was based on both manage network and storage network.At the same time,because of the bad performance of pacemaker's resource agent in the situation that lots of nodes are in the cluster,we create resource agent to achieve the following functions: uploading information of physical machine failure,uploading information of virtual machine failure or closure,sending warning of manage network of storage network failure,detecting node changing in detection field,when uploading message failed finding connected node by using totem to uploading message,affirming heartbeat of isolate machine or isolate group and getting hold of field lock,answering the virtual machine failure message sending by pacemaker.Because when migrating virtual machine,it is needed to decide the destination of the virtual,dynamic resource scheduler service was implied as well.After functional testing and availability testing,we found that this cluster can accomplish starting and stopping virtual machine.When physical machine or virtual has encountered a failure,virtual machine can be migrated and the migrated virtual machine can continuously run the task like before.The time it spends to recover taskrunning is around 20 seconds,which meets the standard of high availability.