multiple RHEV-M hosts

Latest response

I may be missing something important, but is there any issue with cloning the RHEV-M machine and running 2 instances of same to provide redundancy?

Would not vdsmd report any status changes to both RHEV-M instances?

Are there are dangers in this situation?


That is an interesting proposal, but... I imagine the answer is a resounding NO.  Not with the way that RHEV (oVirt) is currently implemented.  This is just my opinion, by the way - I do not know for certain.

That said... I believe RHEVM is solely responsible for any mutexing of resources of the environment (i.e. a guest should not start in 2 places at once).  It is not simply a destination for the Hypervisors to report back to.  If there was a way to keep the 2 RHEV Managers in sync, it would seem more likely that they could be paired to manage the same resources.

Another reason that I believe this would not currently be technically possible is the method(s) that RHEVM communicates with RHEV Hypervisors.  I think there is a ssl/certificate component involved when you establish a partnership between the Manager and Hypervisor.  I believe you would not be able to communicate from the Hypervisor back to more than one Manager due to that - since you would have to give each machine a different IP.

I have setup the RHEV Manager to be a clustered Application, but that is not the same as what you are proposing.  I don't know whether other Hypervisor managers are capable of what you are proposing - especially in an Active/Active configuration.   However, I think IBM has clustered VIO systems.

I look forward to hearing other people's perspective as well.  Great question.

Let me explain how far I got with this, before stress made me stop and ask for opinions. 

My installation of RHEV-M is running as a KVM guest on a RHEL6 host (not RHEV).   RHEV-M manages 2 remote datacenters over the WAN link.

My plan was to clone the RHEV-M guest to another RHEL6 host, that resides at one of the remote locations, the DR site.   The thought was to send copies of the postgres database over nightly, and keep the second database nearly in sync.  Any changes like adding a guest or storage domain would get there with the new database copy.

So today, I cloned the RHEV-M guest, started it to runlevel 1, setup new mac and IP addresses, then let the machine fully boot.   At that point, I had 2 RHEV-M machines running on different IP addresses.    I connected to the webportal on the cloned machine.   After a minute or two, the datacenters showed as started, with storage domains and VMs reporting in.   Obviously, vdsmd was communicating back to the new machine.

Maybe the ssl certificate issue did not happen because I did not change the hostname.   SSL, as fas as I understand, are certificates on the hostname.domainname, with DNS translating the IP address to get you there.    I set /etc/hosts for resolution, and do not use DNS yet for the cloned machine.

I wasn't sure this was the calm before the storm, so I shutdown the 2nd RHEV-M, and await the opinions of the group.