On 20/08/2020 20:55, Michael Jones wrote:
On 19/08/2020 14:48, Michael Jones wrote:
> On 19/08/2020 12:12, Michael Jones wrote:
>> On 19/08/2020 10:41, Nir Soffer wrote:
>>>> There is no warning the method was deprecated and will be missing
functionality.
>>>>
>>>> The steps detailed on the alt install page are for the all-in-one running
engine-setup.
>>>>
>>>> It's also worth noting this works fine in;
>>>>
>>>> Version 4.3.1.1-1.el7
>>>>
>>>> but not in;
>>>>
>>>> Version 4.4.1.10-1.el8
>>>>
>>>> (el8 has the change in imageio daemons)
>>>>
>>>> The alternate install method is still useful to have, but i think a red
warning about all-in-one on el8 on that page would be good.
>>>>
>>>> Kind Regards,
>>>> Michael Jones
>>> Micheal, can you file a bug for this?
>>>
>>> If you have a good use case for all-in-one deployment (not using
>>> hosted engine), please explain
>>> it in the bug.
>>>
>>> Personally I think simple all-in-one deployment without the complexity
>>> of hosted engine is better,
>>> and we should keep it, but for this we need to teach engine to handle
>>> the case when the proxy
>>> and the daemon are the same server.
>>>
>>> In this case engine will not try to setup a proxy ticket, and image
>>> transfer would work directly
>>> with the host daemon.
>>>
>>> I'm not very optimistic that we will support this again, since this
>>> feature is not needed for RHV
>>> customers, but for oVirt this makes sense.
>>>
>>> Nir
>>>
>> Yes, I can file a bug,
>>
>> The main usage / setup's I have are;
>>
>> on-prem installs:
>>
>> - hosted engine
>> - gluster
>> - high availiblity
>> - internal ip address
>> - easy great...
>>
>> dedicated host provider for example OVH single machine:
>>
>> - alternate install
>> - all-in-one
>>
>> The main reason for the separation is that using the cockpit install
>> / hosted engine install causes problems with ip allocations;
>>
>> cockpit method requires 1x ip for host, and 1x ip for engine vm, and
>> both ip must be in the same subnet...
>>
>> applying internal ip would cut off access, and to make it even
>> harder, getting public ip blocks didn't work as the box main ip
>> wouldn't be in the same subnet, adding nic alias ip doesn't work
>> either (fail on install due to failing to setup ovirtmgmt network).
>>
>> atm, i'll struggle with changing the machine's main ip to be one of
>> the same subnet with the engine one... (currently causes host to be
>> taken offline due to hosting provider, health checks)
>>
>> provided i can change the host primary ip to be one of the OVH
>> failover ip allocated in a block... i will be able to install using
>> the cockpit.
>>
>> and after the install i can setup internal ip with the network tag.
>>
>> Kind Regards,
>>
>> Mike
>>
> despite managing to get OVH to disable monitoring (ping to the main
> ip, and rebooting host) and getting the host in the same ip range as
> the engine vm...
>
> ie:
>
> host ip: 158.x.x.13/32 = not used anymore
>
> new subnet: 54.x.x.x/28
>
> and reserving;
> host = 54.x.x.16
> engine = 54.x.x.17
>
> [ ERROR ] The Engine VM (54.x.x.17/28) and the default gateway
> (158.x.x.254) will not be in the same IP subnet.
>
> the hosted engine installer crashes due to the gw being in a
> different subnet, so all three;
>
> - host
> - engine
> - gateway
>
> must be in the same subnet...
>
> this rules out an install on ovh dedicated server.
>
> unless... I can install the all-in-one again (this bit works), and
> then install the engine vm in an existing all-in-one setup...
>
> essentially the cockpit installation is not compatible with this
> infra setup.
>
After going through the documentation again, I understand the best way
to approach this would be to have a remote manager, ie;
self hosted engine (on-prem) > host/2nd DC/Cluster (remote/ovh)
standalone manager (on-prem) > host/2nd DC/Cluster (remote/ovh)
That way resolves the ip issues (only need host ip, just don't install
the manager on the remote server)
outstanding... i need to workout the security implications of this.
shame all-in-one is gone, but the above does work, and even means the
remote host can again use local storage.
I'll raise the bug report now i've finished testing, as I think stand
alone, all-in-one, dedicated hosts are affordable and open ovirt to a
wider user base (keeping hardware requirements minimal).
Thanks again,
Mike
Bug raised:
https://bugzilla.redhat.com/show_bug.cgi?id=1871348
There was also a comment that vprotect shouldn't be using the proxy,
I'll check that and raise a separate bug with them.
Kind Regards,
Mike