O04三、计算节点宕机了怎么办

参考 http://www.javashuo.com/article/p-pllqwvvg-cz.html

 

Rebuild 能够恢复损坏的instance 。那若是是宿主机坏了怎么办呢？好比硬件故障或者断电形成整台计算机节点没法工做，该节点上运行的instance如何恢复呢？

 

用shelve或者Migrate可不能够？很不幸，这两个操做都要求instance 所在的计算节点的 nova-compute服务正常运行。幸运的是，还有 Evacuate

 

Evacuate 可在 nova-compute 没法工做的状况下将节点上的instance 迁移到其余计算节点上，但有个前提：instance的镜像文件必须放在共享存储上

 

咱们能够经过断电模拟计算节点故障，而后执行Evacuate 操做恢复 instance。目前Evacuate只能经过CLI 执行。

 

stack@DevStack-Controller:~$ nova help evacuate

usage: nova evacuate [--password <password>] [--force] <server> [<host>]

 

Evacuate server from failed host.

 

Positional arguments:

  <server>               Name or ID of server.

  <host>                 Name or ID of the target host. If no host is

                         specified, the scheduler will choose one.

 

Optional arguments:

  --password <password>  Set the provided admin password on the evacuated

                         server. Not applicable if the server is on shared

                         storage.

  --force                Force to not verify the scheduler if a host is

                         provided. (Supported by API versions '2.29' -

                         '2.latest')

 

其实 Evacuate 是经过rebuild 操做实现的，由于Evacuate 是用共享存储上的instance的镜像文件从新建立虚机

 

一、先是 nova-scheduler 选择新的计算节点

二、nova-compute 在计算节点上执行rebuild操做