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)

巴特西

《转》OpenStack Live Migration