Suggestion : Offline migration of VM without shared storage

[kYann]

It would be nice to migrate VM offline without shared storage.

This is a very common scenario when you have a pool of server at OVH or Online.net and you can't have shared storage.

[primalmotion]

This is now kind of straightforward with the new libvirt flags. Moving this to milestone 1.0

[CyrilPeponnet]

libvirt can handle this but we need a precreated storage on the destination with qemu-img or a lvm block device (same vg name and lv name).

Options :

--copy-storage-all indicates
migration with non-shared storage with full disk copy,

--copy-storage-inc indicates
migration with non-shared storage with
incremental copy (same base image shared between source and
destination).

And for the record, the % when migrating is the % of RAM transferred it does'nt include the disk transfert.

[Nowaker]

This is what I did manually:

source# qemu-img info --output=json bbb.img | jq '.["virtual-size"]' 
88081432576
source# qemu-img info --output=json bbb.img | jq '.["format"]'
raw
target# qemu-img create -f raw bbb.img 88081432576
source# virsh migrate bbb qemu+ssh://utgard.cumulushost.eu/system --live --copy-storage-all --persistent --verbose

Looks like this shouldn't be hard. Target hypervisor, once migrate is called, needs first to ask the VM on source hypervisor for disks list (archipel:vm:disk/get). This returns everything we need.

Then new Archipel agent thread is started. Before the actual libvirt migration is started, disks need to be created (archipel:vm:disk/create) with the remembered details.

@CyrilPeponnet Does it make sense?

[Nowaker]

Worth noting the migration process may be stuck at 99% when CPU usage is constantly high. Then this needs to be called during the migration to make it working. Won't work before migration process is started.

virsh migrate-setmaxdowntime --domain bbb --downtime 10000