Host Architecture for HA


I’m new to OpenNebula and trying to figure out how most suitable way to build it could be. Just have a question about host HA. According to
it can be done by a Host Hook like:
name = “error”,
on = “ERROR”,
command = “ft/host_error.rb”,
arguments = “$ID -m -p 5”,
remote = “no” ]

that makes sense if there’re two hosts. What if there’re more than two hosts. How it’s decided where to migrate VMs on a ERROR state host?

Another one; is it possible to build a shared filesystem cluster for images datastore by using host’s themselves. So there’ll be no need for separate physical cluster&servers for storage.

any recommendations?


The -m flag will perform a vm resched operation. The scheduler will decide the best host using the same mechanism as the first VM deployment from pending.

Yes, sure, you may build images and system datastores using the same or separate hardware. End solution depends on your experience with cluster file systems like CEPH FS or GlusterFS (recommended), your cluster load and of cause your finances. It’s OK to start from combining compute nodes and storage nodes and scale installation depending from your needs in the future.

Yes, used lizardFS and built it like that way, and now working on details…