Cluster node sizing advice: 2 Qdevices?

[iccbroadcast]

Hi forum...
We're starting to have trouble with our loaned hardware earlier than expected, so, although we've been following and testing carefully the 6.x releases, now we're in the real one, as we need to migrate from our 5 nodes 5.x cluster to a new one.

The question is, regarding cluster size, available hardware, datacenters and, of course very important, costs, that we're having figures of a 4 node cluster as optimal for our needs, being 5 too expensive, and 3 kinda scary. We got two Debian auxiliary servers running happily as cluster backup, auxiliary storage and tasks comapnions, and we're thinking of using them as Qdevices... so here's the question, since we got some options on the table:

- Do Not overthink: Run just one Qdevice, as typically depicted and documented.
- Explore the possibility of running two Qdevices... but, again, this is 4 nodes + 2 Ddevs ... even number... maybe bad idea!?!
- Use keepalived (VRRP) Debian package to make both Debian servers share a common 'floating' IP for QDevice daemon. We got very much positive experience with keepalived in router appliances and load-balancers, but... has anyone ever tried that?!?!

Hope some experienced members could share some advice!
Thank you very much anyways and best regards.

 

[aaron]

I would go with "Do not overthink".

All other options (2x QDevices, floating IP) are both setups that are not tested nor supported AFAIK.

2 QDevices will not gain you much anyway as you want to have an odd number of votes in the cluster.

I would try to make sure that the machine on which the qnetd service is running, does not share failure domains with the cluster (network, power, ...) .

 

[iccbroadcast]

Hi! thank you very much for your advice.

The thing is that all machines have only two network connection types which are the single, common, point of failure.
We carefully select all machines in different datacenters, so we isolate datanceter-wide problems, and, regarding hardware, we have no complain.. it is good quality, with redundant power supply and redundant storage (our guess, by experience, is that RAM and/or NVME are the failing/weak points after some 2-3 years of extensive usage)
The problem is that we got machines inter-connected via a kind of 'LAN' layer2 / broadcast domain (My guess it is xVLAN L2 over L3, VLAN-able, inter-datacenter connection), and, every machine has its WAN facing connection.

Is it possible in 6.x to inter connect cluster/Qdevices from more than one interface simultaneosly? I'm affraid, again, on this being 'Overthinking' the whole thing , causing more problems than solving ones, as cluster traffic on WAN interfaces would have to share bandwidth with actual services payload...

EDIT: now I'm also thinking on security concerns on using WAN connections as a second/redundant cluster connection... If it is SSL/TLS secured traffic I guess it is no problem