2 node High Availability cluster using DRBD and RPi as 3rd node

[topquark]

Soooo, I'm trying to get myself into high availability for a home project. I sort of kind of have a plan, but as my experience is limited I'd like to check here first to make sure I'm not doing something impssible/stupid.

The setup I want to end with is 2 nodes, that both run VM's in HA mode, where if one of the nodes fails the setups are transfered to the the active node. I do have a NAS, but it's by no means redundant enough to act as HA storage.

My (hopely workable) plan is as follows:
Use DRBD to sync the local storage of both Nodes.
Use a Raspberry Pi as a quorum only third node.
Fencing should be accomplished by disabling ports on 2 managed switches.

Not a lot of details yet, but checking if the concept should work before I dive into something that wont work anyway.
Will using DRDB work with HA and the 3rd pi node?
Has anyone done this before/is there a guide somewhere? (I have found the pi as quorum on the wiki, but not sure how HA storage is done).
Can I experiment with 3 VM's running proxmox (on an existing node) or will this not be a good experiment as proxmox normally runs as the os instead of virtualized?

 

[guletz]

Hello,

I do not know if drdb can be used in a such setup. But if it is ok for you let say 5 min data lose, then you can use async replicasion from node1 to/from node 2 for any CT/VM(via zfs)
When let say your node X is broken, then via a custom script you can run all yours VM/CT on the second node. You can define the schedule async zfs replication for each 5 min and so.

And if both nodes are alive you can move any VM/CT from one node to the second node.

Yes Raspberry 3.141 can be used for cluster qourum.

 

[topquark]

Thanks, for your reply, when you say you're not sure if DRDB can be used for this, does that mean that I can try, but the likelyhood of it working is close to 0, or do you just genuinly don't know?

5 min of data loss technically isn't much of a problem. But I technically don't even need HA.
For the sake of exersice I would like to have 0 data loss in such a way that the transition is as seamless as possible. If that would be too hard to accieve then I would settle for async replication.

 

[guletz]

when you say you're not sure if DRDB can be used for this, does that mean that I can try, but the likelyhood of it working is close to 0, or do you just genuinly don't know?

For sure PMX do not support drdb. But I see on the forum some guys who used with pmx. I also use myself many years ago drdb. The setup is let say at a medium level. But in the end it will be a additional software that you must check/administer.
Also note that drdb could go in a split-brain situation(network problems most of the time if I remembered ). In a such case you will need to resync the ALL data from node A to node B. And this operations could take a lot of time (depending of the data size).

Now let compare the drdb landscape with zfs. You will need to install only pve-zsync. Then you will need to create a zfs replication (using pmx web interface) or using pve-zsync command line. In the worst case you will finish in let say 5 min.
In my case I have setup using pve-zsync, and I forget ... so in 6 month I do not have any failures.

In the end , you can test both variants and find what is ok for you.