[ovirt-users] all in one migration

[Simone Tiraboschi]

----- Original Message -----
> From: "Yedidyah Bar David" <didi at redhat.com>
> To: "nicola gentile to" <nicola.gentile.to at gmail.com>
> Cc: "Simone Tiraboschi" <stirabos at redhat.com>, Users at ovirt.org
> Sent: Tuesday, June 23, 2015 3:08:19 PM
> Subject: Re: [ovirt-users] all in one migration
> 
> ----- Original Message -----
> > From: "nicola.gentile.to" <nicola.gentile.to at gmail.com>
> > To: "Simone Tiraboschi" <stirabos at redhat.com>, Users at ovirt.org
> > Sent: Tuesday, June 23, 2015 3:51:18 PM
> > Subject: Re: [ovirt-users] all in one migration
> > 
> > thank you so much
> > 
> > Nicola
> > 
> > Il 23/06/2015 14:38, Simone Tiraboschi ha scritto:
> > >
> > > ----- Original Message -----
> > >> From: "nicola.gentile.to" <nicola.gentile.to at gmail.com>
> > >> To: Users at ovirt.org
> > >> Sent: Tuesday, June 23, 2015 2:10:19 PM
> > >> Subject: [ovirt-users] all in one migration
> > >>
> > >> Good morning,
> > >> I have currently a server with ovirt installed (all-in-one) and I would
> > >> like to extend the structure.
> > >>
> > >> At the end of the work I would like that structures were composed of the
> > >> following:
> > >>
> > >> - 1 manager (virtualized)
> > >> - 2 hosts
> > >> - storage
> > > So hosted-engine is probably the best solution.
> > >
> > > You could follow this guide:
> > > http://www.ovirt.org/Migrate_to_Hosted_Engine
> > >
> > >> Before beginning this complicated process, I would like to know if it is
> > >> possible, and how, to rename the old server.

Sorry, another thing:
all-in-one simply relies on local storage (the engine and the host are the same machine and your VMs are stored on the local disk) while hosted-engine is intended to work with a shared storage (iSCSI and NFS for 3.5, also FC and GlusterFS in the next 3.6 release) so you also have to move all your VMs to the shared storage.
By the way a single datacenter cannot be at the same time a shared and a local one so migrating the old engine it's not really an option there.

On my opinion the simplest solution is
1. deploy hosted-engine on the second (the new) host using the shared storage for the new engine VM skipping all the migration/renaming part.
2. export all the VMs from the all-in-one host
3. import them in the hosted-engine one 
When everything is in the new one and you are sure that it's working you could clean the first host to get rid of all the all-in-one stuff and redeploy it as an hosted-engine additional host.

As Didi suggested you could simply rely on CNAMEs to have your fresh engine VM resolved as the previous all-in-one host; when hosted-engine setup and engine-setup asks you about engine FQDN just use the CNAME cause the communication is over https and so the name should match.  

 
> For this please check:
> 
> http://www.ovirt.org/Changing_Engine_Hostname
> 
> > >>
> > >> I thought to do the following:
> > >> - change the hostname to the old server (all-in-one)
> 
> Before exporting?
> 
> When you write such a plan (either for your self or for posting),
> differentiate between the hostname (output of 'hostname') and resolution
> (i.e. changing /etc/hosts or dns).
> 
> > >> - update the dns server
> 
> If you have to change names etc., I'd suggest to use CNAMEs. E.g.,
> something like:
> 
> host1 IN A addr1
> host2 IN A addr2
> engine-vm IN A addr3
> my-engine IN CNAME engine-vm
> 
> And use 'my-engine' for the official name, so pass this one to
> the rename script, while setting the VM's hostname to 'engine-vm'.
> 
> Pick whatever names you want of course.
> 
> This way, 'my-engine' is a functional, logical name, not tied to
> any specific machine (physical or virtual). If one day you decide to
> do another change, you can just change this CNAME to point to wherever
> new place you need. E.g. if one day you decide that you want a cluster
> of two VMs active/standby and a floating IP address, you'll have:
> 
> engine-vm1 IN A
> engine-vm2 IN A
> clustered-engine IN A (managed by the cluster software)
> my-engine IN CNAME clustered-engine
> 
> > >> - install the new server manager virtualized with old hostname
> > >> - add storage
> > >> - add the new server as a host
> > >> - expoxt the virtual machines from old server
> > >> - import the virtual machines to new server
> > >> - test the virtual machines on new server
> > >> - shutdown old server
> > >> - reinstall old server as a host
> > >>
> > >> What do you think?
> 
> Didn't comment on the rest of your plan, because I am not sure what
> are your uptime requirements, how easy (or not) it is to test old and
> new machines (with same IP?), etc.
> 
> Plan well and if possible test. You can do at least some of the testing
> using nested-kvm (if you do not have enough hardware).
> 
> Feel free to consult us if needed :-)
> 
> Good luck and best regards,
> --
> Didi
>