[ovirt-users] Workflow after restoring engine from backup

[Sven Achtelik]

Hi Didi,

my backups where taken with the end. Backup utility. I have 3 Data centers, two of them with just one host and the third one with 3 hosts running the engine.  The backup three days old, was taken on engine version 4.1 (4.1.7) and the restored engine is running on 4.1.9. I have three HA VMs that would be affected. All others are just normal vms. Sounds like it would be the safest to shut down the HA vm S to make sure that nothing happens ? Or can I disable the HA action in the DB for now ?

Thank you,

Sven



Von meinem Samsung Galaxy Smartphone gesendet.


-------- Ursprüngliche Nachricht --------
Von: Yedidyah Bar David <didi at redhat.com>
Datum: 19.03.18 07:33 (GMT+01:00)
An: Sven Achtelik <Sven.Achtelik at eps.aero>
Cc: users at ovirt.org
Betreff: Re: [ovirt-users] Workflow after restoring engine from backup

On Sun, Mar 18, 2018 at 11:45 PM, Sven Achtelik <Sven.Achtelik at eps.aero> wrote:
> Hi All,
>
>
>
> I had issue with the storage that hosted my engine vm. The disk got
> corrupted and I needed to restore the engine from a backup.

How did you backup, and how did you restore?

Which version was used for each?

> That worked as
> expected, I just didn’t start the engine yet.

OK.

> I know that after the backup
> was taken some machines where migrated around before the engine disks
> failed.

Are these machines HA?

> My question is what will happen once I start the engine service
> which has the restored backup on it ? Will it query the hosts for the
> running VMs

It will, but HA machines are handled differently.

See also:

https://bugzilla.redhat.com/show_bug.cgi?id=1441322
https://bugzilla.redhat.com/show_bug.cgi?id=1446055

> or will it assume that the VMs are still on the hosts as they
> resided at the point of backup ?

It does, initially, but then updates status according to what it
gets from hosts.

But polling the hosts takes time, especially if you have many, and
HA policy might require faster handling. So if it polls first a
host that had a machine on it during backup, and sees that it's
gone, and didn't yet poll the new host, HA handling starts immediately,
which eventually might lead to starting the VM on another host.

To prevent that, the fixes to above bugs make the restore process
mark HA VMs that do not have leases on the storage as "dead".

> Would I need to change the DB manual to let
> the engine know where VMs are up at this point ?

You might need to, if you have HA VMs and a too-old version of restore.

> What will happen to HA VMs
> ? I feel that it might try to start them a second time.  My biggest issue is
> that I can’t get a service Windows to shutdown all VMs and then lat them
> restart by the engine.
>
>
>
> Is there a known workflow for that ?

I am not aware of a tested procedure for handling above if you have
a too-old version, but you can check the patches linked from above bugs
and manually run the SQL command(s) they include. They are essentially
comment 4 of the first bug.

Good luck and best regards,
--
Didi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20180319/d6c1b26b/attachment.html>