On Wed, Nov 23, 2016 at 1:18 PM, Oscar Segarra <oscar.segarra@gmail.com> wrote:
Hi, 

As on oVirt is it possible to attach local storage I supose it can be used to run virtual machines:

I have drawn a couple of diagrams in order to know if is it possible to set up this configuration:

1.- In on-going scenario:
Every host runs 100 vdi virtual machines whose disks are placed on local storage. There is a common gluster volume shared between all nodes.

Imágenes integradas 1

With local storage you end up losing many of the benefits of shared storage - including migration and HA.
If you do have SSD on your physical hosts, have you considered building gluster volume using these? This could give you improved performance. Regarding performance, I think it is best that you run a test comparing gluster storage performance with local storage and see if this is acceptable to you. Please share the results in case you do.

Yes, but I want to avoid possible corruption problems due to possible unrecoverable gluster bugs. 
We have to make some developement and I don't want to spend money in this process and then discover that the performance is not good enought and have to do a 

"possible unrecoverable gluster bugs" is a sweeping statement. Do you have any particular issue that you can refer us to?
 


In the above diagram each host is in its own cluster - as all hosts in a cluster should have access to the storage domain?

Yes, every host has ho have access to two storage domains: The local one and the shared gluster one.

Is the gluster volume for backup served from a separate set of server?

No, each host will have 2 disks /dev/sdb1 (for runing vm on local storage) and /dev/sdc1 (for shared gluster where store backups)
 

2.- If one node fails:

Imágenes integradas 2

oVirt has to be able to inventory the copy of machines (in our example vdi201 ... vdi300) and start them on remaining nodes.

¿Is it possible to reach this configuration with oVirt? ¿or something similar?

This is the use case for gluster volume shared storage - where volume is a replica 3. If any host goes down, the data is available on the remaining 2 nodes, and the VMs can be migrated to other nodes. 

Yes, I know, but I'm already worried about corruption issues due to possible gluster bugs or performance problems under heavy load.

I don't think what you ask for is possible automatically. If you want local storage to gluster volume backup, you would need 1-1 mapping. i.e each local storage domain has its own gluster volume backup.You could then import the storage domain that's backed up on the gluster volume and start the VMs on the remaining hosts.

I don't want local storage for backup, I prefer gluster shared storage for backup.
 
Making backup with the import-export procedure based on snapshot can take lot of time and resources. Incremental rsync is cheaper in terms of resources.

Geo-replication based backup internally uses rsync, it also takes into account that VM images are consistent on disk before being synced. It however works as a backup option between two gluster volumes.

Do you know if is it possible to have multiple masters geo-replicating against a single slave?

No it is not possible. A master can have multiple slaves not the other way around.
 

Thanks a lot.