Disk move from one node to another via migration net

[chrwa]

Hi,

in our last 3 cluster, I have installed 3 networks - migration net / cluster net / public net.
With nload I could see, that the move of a disk is transferred over the cluster net, what seems wrong for me.
How can I decide the network for moving disks, Iwould prefer the migration net

Thanks in advance

 

[dcsapak]

but the network for the disk move depends on where you move to and from. e.g depending on my network config it can happen that it uses neither the migration nor cluster net....
where did you move the disk to and from ? and how does your network config look like?

 

[chrwa]

but the network for the disk move depends on where you move to and from. e.g depending on my network config it can happen that it uses neither the migration nor cluster net....
where did you move the disk to and from ? and how does your network config look like?

auto enp19s0f0
iface enp19s0f0 inet manual

auto enp19s0f1
iface enp19s0f1 inet manual

auto bond0
iface bond0 inet manual
bond-slaves enp19s0f0 enp19s0f1
bond-miimon 100
bond-mode 802.3ad
mtu 9000
#migration-net-bond

auto vmbr0
iface vmbr0 inet static
address 192.xxx.xxx.227/24
gateway 192.xxx.xxx.1
bridge-ports eno1
bridge-stp off
bridge-fd 0
bridge-vlan-aware yes
bridge-vids 2-4094
#public-net

auto vmbr1
iface vmbr1 inet static
address 192.xxx.yyy.222/24
bridge-ports eno2
bridge-stp off
bridge-fd 0
#cluster-net

auto vmbr2
iface vmbr2 inet static
address 192.xxx.zzz.222/24
bridge-ports bond0
bridge-stp off
bridge-fd 0
mtu 9000
#migration-net

I tested the move from node1-zfs1 to node2-zfs2. Both zfs are shared via nfs, in the case I want to migrate a guest from node1 to node2.
By the way, migrating a machine works over the migration-net.
Afaik the cluster-net should have no latency, therefor the use of it for a move job looks like not the best way.

 

[dcsapak]

Both zfs are shared via nfs

and how did you configure those on pve ? (/etc/pve/storage.cfg)

if the nfs has an ip in 192.xxx.yyy.0/24 it will always go via the cluster nic since that is the route

 

[chrwa]

Thats why ...
Thanks a lot