Separate cluster network and i spossible also for ceph?

[cheissol]

Hi,
I followed the wiki https://pve.proxmox.com/wiki/Separate_Cluster_Network and with this I can configure a master/slave network for corosync. Since in my 3 node cluster I also have a ceph network, is it possible to configure a master/slave separate network for this storage?
In my ceph initialization I used

pveceph init -network 192.168.99.0/24

So how can I configure a slave network using RRP?

this could be the configuration of master/salve networks:

root@vs1:~# cat /etc/hosts
127.0.0.1 localhost.localdomain localhost

# corosync master
192.168.89.1 vs1.domain.bl vs1 pvelocalhost
192.168.89.2 vs2.domain.bl vs2
192.168.89.3 vs3.domain.bl vs3
# corosync slave
192.168.88.1 vs1-slave.domain.bl vs1-slave pvelocalhost
192.168.88.2 vs2-slave.domain.bl vs2-slave
192.168.88.3 vs3-slave.domain.bl vs3-slave

#ceph master
192.168.99.1 one-ceph0.domain.bl one-ceph0
192.168.99.2 two-ceph0.domain.bl two-ceph0
192.168.99.3 three-ceph0.domain.bl three-ceph0
#ceph slave
192.168.98.1 one-ceph1.domain.bl one-ceph1
192.168.98.2 two-ceph1.domain.bl two-ceph1
192.168.98.3 three-ceph1.domain.bl three-ceph1

Thanks.

 

[wolfgang]

Hi,

AFAIK there are no fail-over functionality in ceph hammer.
We normally do this with a linux bond.

 

[cheissol]

Hi,
so can I use linux bond also for inter cluster communication without RRP (used in corosync.conf)? I mean I can always use linux bond, am I right? I read that using bond for replication switch can cause loss/retransmission packets but only in critical environment that isn't my case.

 

[wolfgang]

It is possible to use bond for the cluster communication but I would use a second ring.
It is more stable and faster.

see
https://pve.proxmox.com/wiki/Separate_Cluster_Network