8:50 p.m.
Hi Simon!
I'd given up on ever finding any real person or back-channel on the Oracle side of
oVirt, so you're saying there is actually such a thing!
I'd [have] been more than happy to feed back all those results I was collecting in my
desperate attempts to maintain a HCI infra with all those shifting Enterprise Linux
players doing politics.
Today my enterprise use case is transitioning to the cloud and my private use case to
Proxmox.
The latter has run mostly on Pentium Silver J5005 Atoms during the last couple of years
and I am currently trying to work out the kinks on KVM live-migration between an Orange PI
5+ (32GB) and a Raspberry PI 5 (8GB) under Proxmox (using storage on a Ceph HCI cluster
running on x86), so you may appreciate why an Oracle support contract wasn't in the
picture for an infra I keep under four digits total invests to appease the wife.
(Ok, I justed noticed that you're running OL9 on your OP5+, but I don't see you
trying to port oVirt there...)
Without that contract, it seems, that Oracle keeps very "stumm".
So on HCI:
When I ran across oVirt, that was somewhere when 4.3 was fresh and oVirt was advertised as
"a solution for your entire enterprise" which included HCI, to catch potential
Nutanix and vSphere customers.
It sold me on the idea, that I could take an oVirt node ISO, install that on my hardware
nodes, run a GUI wizard to thurn them into a clustered HCI appliance and be as happy as
the other guy with Nut[ell]anix.
That dream certainly cost me months of my life, not the hours I had imagined, but it paid
a salary, too, when I managed to run it anyway.
It took me long to realize that Oracle had ditched all their Xen stuff and become an oVirt
convert. But even since then, there has been very little details and firm commitments nor
even a branding that doesn't require typing classes to execute, so sorry, if most of
my impressions are simply from informational gaps.
But to my knowledge, Oracle never published node ISO images.
Also to my knowledge, oVirt itself ditched HCI support, Redhat itself made nearly the
entire technology stack oVirt is based on EOL, Gluster, oVirt, VDO and, of course, I had
used all of that.
Except storage tiering, where you'd use SSDs in your VDO/Gluster storage for a caching
layer on top of HDDs: that I never got to work and then SSDs became mainline anyway.
On my first EL8/oVirt 4.4 tests Oracle's Enterprise kernel failed immediately with
VDO, which was missing then. Later even the Redhat kernel sometimes failed with VDO after
kernel upgrades, because evidently nobody at Oracle cares about VDO. Funny, when you
consider those Sun guys used to be very big into something similar...
But also later I got into all kinds of trouble when I was setting up HCI with 4.4 and had
not switched the kernels to the Redhat variant. If I remember correctly, the management
engine never managed to connect to the network after it had been teleported into KVM and
after it had been successfully configured locally on the temporary install node. I could
have been that I tried this on nested virtualization, but it felt more kernel related,
because switching that fixed the issue.
Later I experimented with upgrades from 4.4 to 4.5 and ran into all sorts of trouble when
switching the kernel there. Except that now things started failing with the Redhat kernel.
Generally nobody seems to test switching between UEK and RHK on HCI nodes, which *should*
be totally transparent on the wire and basically to all user space, if I understood Wim
Coekaerts correctly, when we met in 2011.
So my impression was that Oracle supported a subset of what oVirt supported and with HCI
not even giving any search responses anywhere on oracle.com, I didn't see that
remaining.
And perhaps all I missed was to install 'cockpit-ovirt-dashboard'...
So, good to know Oracle hasn't given up, good to know you keep an ear open here and
now if there was a bit more public commitment for oVirt, we'd all be much happier.