GlusterFS NAS Understanding

[bmacadre]

Hi,

I've a 4 nodes promox cluster connected with a 2 nodes Gluster NAS. All works fine, I've connected my NAS under Proxmox with GlusterFS plugin and shared it all over the cluster.

I've just one little question, when I want to migrate a VM from one node to another, all works but I see Proxmox doing a transfer (see output below)... Why ? The VM disk is accessible over all nodes, so there is no need (and nowhere) to transfer it....

See my output :

Jul 18 16:29:39 starting migration of VM 100 to node 'node4' (10.196.170.20)
Jul 18 16:29:39 copying disk images
Jul 18 16:29:39 starting VM 100 on remote node 'node4'
Jul 18 16:29:41 starting ssh migration tunnel
Jul 18 16:29:41 starting online/live migration on localhost:60000
Jul 18 16:29:41 migrate_set_speed: 8589934592
Jul 18 16:29:41 migrate_set_downtime: 0.1
Jul 18 16:29:43 migration status: active (transferred 120296365, remaining 420278272), total 545742848)
Jul 18 16:29:45 migration status: active (transferred 239790246, remaining 300183552), total 545742848)
Jul 18 16:29:47 migration status: active (transferred 359840408, remaining 179339264), total 545742848)
Jul 18 16:29:49 migration status: active (transferred 479939710, remaining 58257408), total 545742848)
Jul 18 16:29:50 migration status: active (transferred 495918494, remaining 39706624), total 545742848)
Jul 18 16:29:50 migration status: active (transferred 514937450, remaining 20942848), total 545742848)
Jul 18 16:29:50 migration speed: 56.89 MB/s - downtime 8 ms
Jul 18 16:29:50 migration status: completed
Jul 18 16:29:53 migration finished successfuly (duration 00:00:14)
TASK OK

As you can see there is a transfer (between what .... I don't know), what is transfered ? and why ? What's the need ?

The worst thing is that any HA managed VM was shutdown instead of migrated when a node was stopped... that's not normal at all.


I'll appreciate if you have some answers or some clues about this behaviour..... I like understand

Regards,
Bruno.

 

[mir]

What you see transferred from source node to destination node is the memory and runtime instance.

 

[bmacadre]

What you see transferred from source node to destination node is the memory and runtime instance.

Ahhh OK.... it's logic.... But since I've activated HA on my whole cluster, all migration are done without any tranfer at all.... I'm a lot confused, but it works

Thanks for your answer.

Regards,
Bruno

 

[acidrop]

HA has nothing to do with that.
As mir said what is transfered is the memory of the vm.
This happens in all kinds of shared storage not only gluster.
What HA refers to mainly is that in case of a node crash all vms that reside on this node will be automatically restarted on the next available node without human intervention.

 

[bmacadre]

I understand well all that mir explained to me, but it's a fact since HA was enabled, when I start an online migration I've got no more transfer at all and the migration log say :

 starting HA migration of VM 100 to node 'node4' (10.196.170.20)

instead of :

starting migration of VM 100 to node 'node4' (10.196.170.20)

like in my first post.

It's just an observation.... All work like a charm.

 

[mir]

When HA is enabled you will see to log instances of your migration one shown as in description 'HA 100 - migrate' as well as another one showing 'VM 100 - migrate'. If you double click the one showing 'VM 100 - migrate' in describtion you will find what you are looking for.