Storage Node HA Options for a two node cluster. DRBD or something else

Hi! We are currently looking for an Open Nebula ‘TWO’ Node High Availability(HA) Setup for our customer needs. We are fully aware of the disadvantages of a 2 Node HA but we don’t have much of a choice. We’ve managed to have a working HA implementation for most parts(Front End and Host). Thanks largely to the on line documentation and support. The only remaining part is the Storage Node HA. And here’s the thing

  1. We tried the DRBD add-on, but the Open Nebula DRBD Driver seems to only support Image datastore and not the System Datastore feature. The documentation on github also briefly suggests the same. Is it indeed so? Just need a confirmation.

  2. And if not DRBD, how do we go about implementing a Storage Node HA setup with only 2 nodes. Ceph HA also depends on a Paxos consensus algorithm that needs a minimum of 3 nodes. Is it possible? Any ideas?

Edit: We are planning on having the front end, host and storage on each of the two nodes.

Hello, for 2 node cluster I can recommend LSI Synchro instead of DRBD. You can get that cards on eBay for much better price ($1000) that retail ($5000) and create two node HA cluster on HW level.

Well we are primarily looking for a software solution… possibly within the bounds of Open Nebula. A storage controller would be an overkill for our purposes.

You still need some type of better performance raid controllers. You also save on HDD, because no need of replication = half total capacity.

Hello, I have the same scenario and I managed to solve the storage part, use glusterfs with both nodes and it works great, however I have the error of the frontend and hosts, could you tell me how you managed to do it, I share a link in which I I help achieve Storage HA.

Regards.

1 Like

Hello, another cheaper solution is to use 6G HBA like LSI 9207-8e with Dual-IO JBOD enclosure(s).
Connect each server to each enclosure an setup md-cluster raid1 and on top of it stripped clvm (something like raid0), so you get software RAID10 in HA. Them you can decide to use clustered FS like GFS2 or use raw images directly on lvm.