[SOLVED] Cluster lacp

[Manutec]

Hi:
Is it posible have a cluster with lacp? I have ping between three nodes in a vlan inside bond with lacp but I´m not able to add the nodes to the cluster. If is not posible can I do with balance-rr? Thanks

 

[ph0x]

Of course you can, lacp is a lower layer and therefore cannot affect clustering. However, a dedicated network for corosync is encouraged.
What errors are you facing when joining a cluster?
Balance-rr is honestly the last bonding mode I would ever use.

 

[Manutec]

I have a dedicated 1G Nic for management and 2 10G Sfps for lacp. I create a bond and bond.xx for the cluster. I create a cluster with link0 vlan.xx and when I add the node2 to the cluster with vlan.xx IP the server hang up after quórum....Ok . Finally I change to balance-rr and make the cluster with management Nics. After that edit corosync.conf with vlan.xx . It's works but I think is not the best way to do that. Thanks

 

[kofik]

Please check out the PVE documentation. While it can be overseen, it is pretty much made clear in 3.3.7. Linux Bond:

If you intend to run your cluster network on the bonding interfaces, then
you have to use active-passive mode on the bonding interfaces, other modes 
are unsupported.

Based on this in our cluster I've made the following setup per Node:

• 2x 1G in active-backup (onboard) for management and primary Corosync link / ring
• 2x 10G in LACP for CEPH, secondary Corosync ring
• 2x 10G in LACP for VMs while all nodes have 1 VLAN is configured as live migration network

This seems to be working pretty solid and I still get recommended / compliant redundancy on the first corosync ring while not wasting a pricier 10G link for cluster synchronization. The secondary link while not compliant would only be used both 1G management links would die out so I guess that this is "good enough" for us. So far, so good.

 

[Manutec]

Ok, so is it correct?
2x 1G active-backup
-vlanXX managment
-vlanXY cluster creation

2x 10G LACP -vlanYY CEPH (public network in ceph configuration?)
2x10 LACP -bridge or whatever for vms outside

 

[kofik]

2x 1G active-backup
-vlanXX managment
-vlanXY cluster creation

In my case I've kept it simple, the 2x 1G is untagged, not VLANs and management is also the cluster network.

 

[Manutec]

Ok. I will try. Thanks a lot

 

[Manutec]

Solved!!. Works like a charm with this config. Finally y separated in two vlans for more isolation managment and cluster.

 

[Pca]

Hello
I wonder what's the reason behind this restriction ?

If you intend to run your cluster network on the bonding interfaces, then you have to use active-passive mode on the bonding interfaces, other modes
are unsupported.

 

[rekordskratch]

Hello
I wonder what's the reason behind this restriction ?

I'd also love to know. I've had a cluster running with 802.3ad LACP across 2 x Cisco Nexus 3064's spanned with VPC in a lab environment for some time now and cannot see any negative effects so far