This post is based assumption that KVM as hypervisor, and Openstack is running in Grizzly on top of RHEL6.4.

Openstack VM live migration can have 3 categories:

-Block live migration without shared storage

-Shared storage based live migration

-Volume backed VM live migration

Block live migration

Block live migration does not require shared storage among nova-compute nodes, it uses network(TCP) to copy VM disk from source host to destination host, thus it takes longer time to complete than shared storage based live migration. Also during migration,
host performance will be degraded in network and CPU points of view.

To enable block live migration, we need to change libvirtd and nova.conf configurations on every nova-compute hosts:

->Edit /etc/libvirt/libvirtd.conf

listen_tls = 0

listen_tcp = 1

auth_tcp = "none"

->Edit /etc/sysconfig/libvirtd

LIBVIRTD_ARGS="–listen"

->Restart libvirtd

service libvirtd restart

->Edit /etc/nova/nova.conf, add following line:

live_migration_flag=VIR_MIGRATE_UNDEFINE_SOURCE,VIR_MIGRATE_PEER2PEER,VIR_MIGRATE_LIVE

Adding this line is to ask nova-compute to utilize libvirt's true live migration function.

By default Migrations done by the nova-compute does not use libvirt's live migration functionality,
which means guests are suspended before migration and may therefore
experience several minutes of downtime.

>Restart nova-compute service

service openstack-nova-compute restart

->Check running VMs

->Check which host the VM in running on

->Check all available hosts

>Let's migrate the vm to host compute-2

nova live-migration --block-migrate vm-live-migration compute-2

>Later when we check vm status again, we cloud see the host changes to compute-2

>We can just do a migration without specifying destination host, nova-scheduler will choose a free host for you

nova live-migration --block-migrate vm-live-migration

Block live migration seems a good solution if we don't have/want shared storage , but we still want to do some host level maintenance work without bring downtime to running VMs.

Shared storage based live migration

This example we use GlusterFS as shared storage for nova-compute nodes. Since VM disk is stored on shared storage, hence live migration much much more faster than block live migration.

1.Setup GlusterFS cluster for nova-compute nodes

->Prepare 4 nodes, each node has an extra disk other than OS disk, dedicated for GlusterFS cluster

Node 1: 10.68.124.18 /dev/sdb 1TB

Node 2: 10.68.124.19 /dev/sdb 1TB

Node 3: 10.68.124.20 /dev/sdb 1TB

Node 4: 10.68.124.21 /dev/sdb 1TB

->Configure Node 1-4

yum install -y xfsprogs.x86_64

mkfs.xfs -f -i size=512 /dev/sdb

mkdir /brick1

echo "/dev/sdb /brick1 xfs defaults 1 2" >> /etc/fstab 1014 cat /etc/fstab

mount -a && mount

mkdir /brick1/sdb

yum install glusterfs-fuse glusterfs-server

/etc/init.d/glusterd start

chkconfig glusterd on

->On Node-1, add peers

gluster peer probe 10.68.125.19

gluster peer probe 10.68.125.20

gluster peer probe 10.68.125.21

->On node-1, create a volume for nova-compute

gluster volume create nova-gluster-vol replica 2 10.68.125.18:/brick1/sdb 10.68.125.19:/brick1/sdb 10.68.125.20:/brick1/sdb 10.68.125.21:/brick1/sdb

gluster volume start nova-gluster-vol

->We can check the volume information

gluster volume info

Volume Name: nova-gluster-vol

Type: Distributed-Replicate

Status: Started

Number of Bricks: 2 x 2 = 4

Transport-type: tcp

Bricks:

Brick1: 10.68.125.18:/brick1/sdb

Brick2: 10.68.125.19:/brick1/sdb

Brick3: 10.68.125.20:/brick1/sdb

Brick4: 10.68.125.21:/brick1/sdb

->On each nova-compute node, mount the GluserFS volume to /var/lib/nova/instances

yum install glusterfs-fuse

mount.glusterfs 10.68.125.18:/nova-gluster-vol /var/lib/nova/instances

echo "10.68.125.18:/nova-gluster-vol /var/lib/nova/instances glusterfs defaults 1 2" >> /etc/fstab

chown -R nova:nova /var/lib/nova/instances

2.Configure libvirt and nova.conf on every nova-compute node

->Edit /etc/libvirt/libvirtd.conf

listen_tls = 0

listen_tcp = 1

auth_tcp = "none"

->Edit /etc/sysconfig/libvirtd

LIBVIRTD_ARGS="–listen"

->Restart libvirtd

service libvirtd restart

->Edit /etc/nova/nova.conf, add following line:

live_migration_flag=VIR_MIGRATE_UNDEFINE_SOURCE,VIR_MIGRATE_PEER2PEER,VIR_MIGRATE_LIVE

3.Let's try live migration

->Check running VMs

->Check which host the VM in running on

->Check all available hosts

>Let's migrate the vm to host compute-2

nova live-migration vm-live-migration compute-2

>Migration should be finished in seconds, let's check vm status again, we cloud see the host changes to compute-2

>We can just do a migration without specifying destination host, nova-scheduler will choose a free host for us

nova live-migration vm-live-migration

3.Volume backed VM live migration

VMs booted from volume can be also easily and quickly migrated just like shared storage based live migration since both cases VM disks are on top of shared storages.

->Create a bootable volume from an existing image

cinder create --image-id --display-name bootable-vol-rhel6.3 5

->Launch a VM from the bootable volume

nova boot --image --flavor m1.medium --block_device_mapping vda=:::0 vm-boot-from-vol

->Check which host the VM is running on

->Do live migration

nova live-migration vm-boot-from-vol

->Check VM status again

VM recovery from failed compute node

If shared storage used, or for VMs booted from volume, we can recovery them if their compute host is failed.

>Launch 1 normal VM and 1 volume backed VM, make they are all running on compute1 node. ( Use live migration if they are not on compute-1)

->Shutdown compute-1 node

[root@compute-1 ~]# shutdown now

->Check VM status

Both VMs get into "SHUTOFF" status.

>Recovery them to compute-2 node

nova evacuate vm-boot-from-vol compute-2 --on-shared-storage

nova evacuate vm-live-migration compute-2 --on-shared-storage

->Check VM status

Both VMs are managed by compute-2 now

-> Reboot 2 VMs

nova reboot vm-boot-from-vol

nova reboot vm-live-migration

-> Check VMs list

Both VMs are back to ACTIVE.

Notes

Currently Openstack has no mechanism to detect host and VM failure and automatically to recover/migrate VM to health host, this reply on 3rd party means to do so

There's also an interesting project named openstack-neat, which is intended to provide an extension to OpenStack implementing dynamic consolidation of Virtual Machines
(VMs) using live migration. The major objective of dynamic VM consolidation is to improve the utilization of physical resources and reduce energy consumption by re-allocating VMs using live migration according to their real-time resource demand and switching
idle hosts to the sleep mode. This really like vSphere DRS function.

Project website: http://openstack-neat.org

http://www.mirantis.com/blog/tutorial-openstack-live-migration-with-kvm-hypervisor-and-nfs-shared-storage/

(http://blog.sina.com.cn/s/blog_6330de3f01019trq.html)

《转》OpenStack Live Migration的更多相关文章

openstack live migration性能分析
http://blog.zhaw.ch/icclab/an-analysis-of-the-performance-of-live-migration-in-openstack/ http://blo ...
OpenStack Live Migration
About live migration of KVM virtual machines with NFS storage, from Mirantis blog: click this link w ...
OpenStack nova VM migration (live and cold) call flow
OpenStack nova compute supports two flavors of Virtual Machine (VM) migration: Cold migration -- mig ...
Migration workstation vms to openstack kvm
Migration workstation vms to openstack kvm 分别分成(磁盘是否拆分,vms 是否有多个磁盘驱动器) 1, linux迁移 vmware workstatio ...
Openstack虚拟机在线迁移(Live Migration)
Openstack VM live migration can have 3 categories: -Block live migration without shared storage -Sha ...
Setting up multi nodes live migration in Openstack Juno with devstack
Setting up multi nodes live migration in Openstack Juno with devstack Summary Live migration overvie ...
migration vmware vms to openstack kvm 修改vmware windows scsi to ide
vmware vms migration to openstack
Converting a VMware Workstation virtual machine to KVM Leave a commentPosted by rbgeek on August 13, ...
VMware migration to openstack kvm

随机推荐

shell中判断用法
测试结构: 测试命令可用于测试表达式条件的真假,true,则返回0,false,则返回非0:这一点c/c++有区别: 格式: test expression #expression是一个 ...
探究绑定事件的this指向以及event传参的小问题
this指向直接上代码,凑合着看 <input type="button" onclick="page()" value="点我"& ...
IntelliJ IDEA 开发swing（二）
原文:idea开发swing(二) 闲话少说,书接idea开发swing(一). 程序编译完成后,需要打包发布,如果有fat_jar的同学可以通过该插件打包,这里是使用ant来打包,步骤如下: 一.编 ...
MvvmLight Messenger（信使）
MvvmLight信使需要三个部分: 1.自定义信件类,普通的Model,供在发布者和订阅者之间传递信息用. 2.发布,通常是在某一事件函数中进行发布,Messenger.Default.Send 3 ...
Android核心基础(十一)
1.Android的状态栏通知(Notification) 通知用于在状态栏显示消息,消息到来时以图标方式表示,如下: //获取通知管理器 NotificationManager mNotificat ...
BZOJ 1264: [AHOI2006]基因匹配Match( LCS )
序列最大长度2w * 5 = 10w, O(n²)的LCS会T.. LCS 只有当a[i] == b[j]时, 才能更新答案, 我们可以记录n个数在第一个序列中出现的5个位置, 然后从左往右扫第二个序 ...
NVelocity 实例
using System; using System.IO; using System.Collections; using System.Collections.Generic; using Sys ...
C++内存管理（超长）
[导语] 内存管理是C++最令人切齿痛恨的问题,也是C++最有争议的问题,C++高手从中获得了更好的性能,更大的自由,C++菜鸟的收获则是一遍一遍的检查代码和对C++的痛恨,但内存管理在C++中无处不 ...
Windows 7 taskbar and startmenu pin
原文 Windows 7 taskbar and startmenu pin 在Windows 7上,用户可以将自己喜欢的软件“钉”在开始菜单或任务栏,使用起来更加方便.但有时候我们也需要用程序来将这 ...
hdu4708 Rotation Lock Puzzle
Rotation Lock Puzzle Time Limit: 2000/1000 MS (Java/Others) Memory Limit: 32768/32768 K (Java/Others ...

《转》OpenStack Live Migration