[libvirt-users] virsh migrate --copy-storage-inc

[Paul van der Vlis]

Op 11-02-19 om 14:31 schreef Peter Krempa:
> On Sun, Feb 10, 2019 at 12:11:42 +0100, Paul van der Vlis wrote:
>> Hello,
>>
>> I use libvirt on machines without shared storage. My VM's have all one
>> qcow2-disk, with the same name as the VM.
>>
>> When I want to migrate a VM, I check if there is an qcow2 image on the
>> other host with that name. When that's not the case, I copy the image
>> using rsync first. If the image excist, I don't do that, and I think
>> that "--copy-storage-inc" will do it.
> 
> Note that using the 'inc' or incremental mode will copy only the data in
> the top level image rather than a full copy. This means that you need to> pre-create empty qcow2 files which point to the correct backing store.

Hmm, I don't know what's a "top level image", I cannot find how to
create an empty qcow2 file, and I don't know what's a "backing store".

I miss some information, and I don't know where to find this info.

>> But I don't know how intelligent "--copy-storage-inc" is. I use LVM
>> inside the VM's, and it's possible that I have changed the size of a
>> volume for example.
> 
> Non shared storage migration preserves the guest visible state.
> 
>>
>> My questions:
>> Is "--copy-storage-inc" intelligent enough for such tasks?
> 
> If used with --tunelled you can only use it if you wish to migrate ALL
> disks in such a VM and need to pre-create the correct files.
>
> If you wish to migrate only certain disks you need to stop using
> --tunelled.
> 
>> Is there documentation what is done during migration? Is it using rsync?
> 
> Normally (if --tunneled is not used) it's using blockdev-mirror or
> drive-mirror in qemu over an NBD connection to destination.
> 
> In your case it's using the old storage migration protocol.
>
>> Is it possible to make the migration process more verbose?
> 
> What do you mean by this? In case if --tunelled there's nothing we can
> do though. Everything is done inside qemu.

When I understand you well, it would be better to not-use "tunneled".

>> Is what I do a good way?
>>
>> With regards,
>> Paul van der Vlis
>>
>> BTW: this is what I do, sometimes the lines are broken wrong:
>> -----------
>> # Rsync image when not on the other host:
>> if test `ssh $other "if test -e /data/$vm.qcow2; then echo ja; else echo
>> nee; fi"` = "nee"; then
>>   echo "rsync..."
>>   rsync /data/$vm.qcow2 $other://data/$vm.qcow2
>>   if test $? = 0; then echo "gelukt"; else echo "niet gelukt"; exit; fi
>> fi
>>
>> # migrate
>> echo "migrate..."
>> virsh migrate --live --p2p --tunnelled --copy-storage-inc --persistent
> 
> As said above this implies the old storage migration protocol which
> transports the data in the migration stream. This stream is then
> tunelled over the libvirt connection. This is very inefficient and also
> uses the old storage migration protocol which is obsolete and qemu will
> attempt to remove it in the future.

Thanks for your help, I will test how migration work without "tunneled".

I guess I do not need a tunnel, I use allready SSH and a safe network
connection.

With regards,
Paul van der Vlis


>> --undefinesource --verbose $vm qemu+ssh://$other/system
>> if test $? = 0; then echo "gelukt"; else echo "niet gelukt"; exit; fi
>> -----------
>>
>>
>> -- 
>> Paul van der Vlis Linux systeembeheer Groningen
>> https://www.vandervlis.nl/
>>
>> _______________________________________________
>> libvirt-users mailing list
>> libvirt-users at redhat.com
>> https://www.redhat.com/mailman/listinfo/libvirt-users



-- 
Paul van der Vlis Linux systeembeheer Groningen
https://www.vandervlis.nl/