[ovirt-users] Upgrade path from Fedora 20 oVirt 3.5 to Fedora 22 oVirt 3.6

[Blaster]

On 12/26/2015 2:22 PM, Blaster wrote:
> On 11/10/2015 1:10 AM, Sandro Bonazzola wrote:
>> Hi, upgrade from FC20 to FC22 has not been tested and you're welcome 
>> to help testing it.
>> What I would suggest is:
>> - make an image of your existing engine (I suggest clonezilla for 
>> doing it)
>> - install ovirt-release36 from 
>> http://resources.ovirt.org/pub/yum-repo/ovirt-release36.rpm
>> - fedup from fc20 to fc22 directly, without the fc21 step
>> - run rpmconf -a and update relevant config files (be sure to not 
>> just overwrite them!)
>> - run postgresql-setup --upgrade
>> - run engine-setup
>>
>> I suggest to update to fc22 directly because there aren't FC21 builds 
>> of ovirt engine and the fedup process may not work as expected with 
>> missing repos.
>>
>> If the upgrade fails, you can restore the disk image created with 
>> clonezilla and retry with a different procedure.
>>
>> If you hit any issue or if you succeed let us know!
>
> So I did a fedup from 20 to 22, which went fairly well. I did have 
> some issues though:
>
> named was missing some libraries after the upgrade.  Removing named 
> and reinstalling has fixed that up.
>
> rpmconf -a is giving me the following error:
> # rpmconf -a
> /usr/libexec/pk-command-not-found: error while loading shared 
> libraries: libpackagekit-glib2.so.16: cannot open shared object file: 
> No such file or directory
>  dnf provides "*/libpackagekit-glib2.so.16"
> Last metadata expiration check performed 0:59:14 ago on Sat Dec 26 
> 13:19:20 2015.
> Error: No Matches found
>
> So not sure what's going on there.  I do have PackageKit-glib-devel 
> installed.
>
>  postgresql-setup upgrade
>
> Cannot upgrade because the database in /var/lib/pgsql/data is not of
> compatible previous version 9.2.
>
> What do I do about this?
>
> I did engine-setup afterwards and everything seems to be running fine.
>
> I did have to set my host CPU type back to Sandybridge as it wasn't 
> recognizing my I7 4790K.  Seems to be a known issue.
>
> I also have explanation points next to all my VMs stating that my time 
> zone configuration differs between the VM config and the guest config, 
> which they do not.
>

Finally having a chance to get back at this.   This ended up not going 
anywhere near as well as I had originally thought.

The reason why the Postgress upgrade was failing was because I was still 
running 9.3 for some reason after the fedup upgrade.  As I started 
diagnosing the known NFS server problem, I discovered after the fed22 
update, it apparently only updates you to the base Fedora 22 packages. 
You then need to do a dnf update to get the latest. Which then fixed my 
NFS problem, but then vdsm refused to connect to the database.  It was 
at that point I discovered my installed version of Postgress was now @ 
9.4, instead of 9.3.  Not sure how I got to 9.3, as I thought Fedora 20, 
which is where I came from has 9.2, and Fedora 22 had 9.4.

So now I did the Postgress upgrade, which worked as expected above. But 
vdsm still would not connect to the database.  After googling that 
issue, the only suggestion I saw was to remove the password from the 
vdsm conf file and re-run engine setup.  That really upset Ovirt and it 
wiped out everything and proceeded to start with a fresh configuration, 
wiping out my datacenter.

I do run a nightly ovirt backup, which I attempted to restore from, but 
it wouldn't let me do that as my backup which was from the previous 
night was for 3.5, and I was now at 3.6.1.  So much for backups.

It was at that point I just decided to investigate Hosted Engine, seeing 
as I was needing to start from scratch.  After spending several hours 
researching that, it appears HE is know where near ready for any sort of 
daily driver.  The WIKI how to seemed severely out of date and I 
couldn't find a usable engine appliance ISO that seemed to be ready to 
with out hours or days of additional hacking.

So in the end I just re-ran the All-In-One setup and spent a couple 
hours rebuilding my datacenter.

So to the above steps, a dnf update needs to added as a step after the 
fedup reboot.