[Users] oVirt-engine and UserPortal "cluster"

Juan Hernandez jhernand at redhat.com
Thu Oct 4 08:20:36 UTC 2012


On 10/04/2012 09:29 AM, Itamar Heim wrote:
> On 10/04/2012 01:34 AM, Brian Vetter wrote:
>> I've been scouring through the install notes and the architecture documents but didn't find my answer. Is there a way to "cluster" or replicate the userportal app or is it strictly a single instance? Any thoughts to the scale of a large VDI system with 10,000 desktops and their VMs and how that impacts the ovirt-engine and the user-portal app?
>>
>> I figure it has been discussed, but using the word cluster in a google search of the wiki results in a lot of hits, none of them that I saw that are to do with clustering the server, just the virtual machine nodes.
> 
> it shouldn't be a huge issue to deploy the user portal on another 
> machine (although current rpm may require the engine).

As of today the RPMs have inter-dependencies, so it is not possible to
deploy only the user portal. We could make the user portal RPM
independent of the rest. As it is just a JavaScript application it could
even be deployed as static content to many web servers. Those servers
would then need to act as proxies for the requests that the user portal
makes to the backend (due to the same origin limitations in GWT), that
is not complicated. However the scalability bottleneck won't be in the
JavaScript application, but in the backend that processes the requests
that it sends, and we don't have any simple way to replicate it at the
moment, that would require many important changes.

Same for webadmin.

Brian, I will try to prepare some instructions on how to deploy the user
portal to a different machine, but it will be a manual process.

> juan - i don't remember if you already looked at this or not yet?

I didn't look at this deeply. It may be worth to try to support this
independent deployment of GUI applications for next release. It would
also simplify GUI installation in Fedora.

But the real scalability bottleneck will still be in the backend, and
suporting an active-active cluster requires major changes.

> Brian - btw, the user portal is pretty lightweight - its mainly java 
> script going to the client, and polling lightly on status changes (which 
> changes if it is in focus on the client (faster polling) or not (slower 
> polling).
> 
> spice also doesn't go through the user portal today.

-- 
Dirección Comercial: C/Jose Bardasano Baos, 9, Edif. Gorbea 3, planta
3ºD, 28016 Madrid, Spain
Inscrita en el Reg. Mercantil de Madrid – C.I.F. B82657941 - Red Hat S.L.



More information about the Users mailing list