Announcement on the Cloud Platform Unavailable Problem When Deploying CloudOS in a Virtual Machine Using a Shared File System

2020-07-17 Published

Problem Description

【Problem Description】

When deploying CloudOS in a virtual machine with a shared file system as a system disk, the cloud platform cannot be unavailable.

Cause Analysis

【Cause Analysis】

CloudOS uses K8S as the underlying container management tool, and K8S uses ETCD to synchronize data between clusters. ETCD has high requirements on the IO and latency of the system disk. In case of virtual machines that use the shared file system as the system disk, other virtual machines will also occupy IO resources, resulting in large IO read and write performance or IO network delay. It would cause the time out of ETCD cluster to synchronize CloudOS cluster node management data, triggering unstable K8S cluster operation, then triggers abnormal CloudOS unavailability.  

Circumvention measures/solutions

【Circumvention measures/solutions】

Solution 1 When deploying CloudOS virtual machines, the system disk does not use the shared file system. You can migrate the virtual machine system disk to the local disk of the virtualization platform. This solution is only applicable to version 2.0; versions 3.0 and 5.0 do not support virtual machine deployment.

Solution 2: The CloudOS cluster should be deployed to a server that meets the CloudOS deployment specifications. This solution applies to all versions. For the specifications of the deployed server, refer to Appendix 1.