Hi everyone,
I'm designing a new 3-node cluster. It's not the first cluster I make but it's the first with a full-mesh Ceph network (no dedicated switch) . Each node has a double 10G NIC and they're already physically connected each other. So I've been reading the docs and I think the best solution for me could be Routed setup with fallback.
Anyway I was doing some reasoning: what if I create an active-backup bond and two VLANs on top of it? Something like this (let's say this is node 1):
It could handle both connection and node loss easily, IMHO. Do you see any caveats? Too much overhead? Am I missing something? Would you still recommend broadcast bond over this?
Thank you!
I'm designing a new 3-node cluster. It's not the first cluster I make but it's the first with a full-mesh Ceph network (no dedicated switch) . Each node has a double 10G NIC and they're already physically connected each other. So I've been reading the docs and I think the best solution for me could be Routed setup with fallback.
Anyway I was doing some reasoning: what if I create an active-backup bond and two VLANs on top of it? Something like this (let's say this is node 1):
Code:
### /etc/network/interfaces
# (cut)
auto bond1
iface bond1 inet manual
bond-slaves eno49 eno50
bond-miimon 100
bond-mode active-backup
bond-primary eno2
bond-primary-reselect failure
mtu 9000
#bond for Ceph HA
auto bond1.27
iface bond1.27 inet static
address 172.27.0.1/24
#ceph-public
auto bond1.28
iface bond1.28 inet static
address 172.28.0.1/24
#ceph-clusterIt could handle both connection and node loss easily, IMHO. Do you see any caveats? Too much overhead? Am I missing something? Would you still recommend broadcast bond over this?
Thank you!