----- Original Message -----
From: "Itamar Heim" <iheim(a)redhat.com>
To: "Tony James" <tony(a)anthonyjames.org>
Cc: "Nir Soffer" <nsoffer(a)redhat.com>, "Liron Aravot"
<laravot(a)redhat.com>, devel(a)ovirt.org, "Alon Bar-Lev"
Sent: Friday, January 2, 2015 9:44:45 AM
Subject: Re: [ovirt-devel] UI Plugin to Upload ISO Files
On 12/29/2014 06:13 PM, Tony James wrote:
> On Mon, Dec 29, 2014 at 5:26 AM, Itamar Heim <iheim(a)redhat.com> wrote:
>> On 12/29/2014 09:25 AM, Nir Soffer wrote:
>>> ----- Original Message -----
>>>> From: "Tony James" <tony(a)anthonyjames.org>
>>>> To: devel(a)ovirt.org
>>>> Sent: Monday, December 29, 2014 3:30:49 AM
>>>> Subject: [ovirt-devel] UI Plugin to Upload ISO Files
>>>> This message is in response to an earlier thread regarding a UI plugin
>>>> to upload ISO files. Like the original poster, Lucas, I began work on
>>>> a UI plugin to allow uploading ISO files through a UI plugin. After
>>>> reading the previous thread I'm re-thinking the architecture.
>>>> It was suggested that the recommended approach to upload files to a
>>>> storage domain is through the VDSM API . I'm pretty familiar
>>>> the oVirt REST API but have been unable to find documentation
>>>> regarding accessing the VDSM API. Should the VDSM API be accessible
>>>> by a UI plugin? If so, is there documentation available to do so?
>>>>  http://lists.ovirt.org/pipermail/devel/2014-December/009497.html
>>> Basically you have to:
>>> 1. Use the vdsm xmlrpc/jsonrpc to create an image
>>> 2. Use the vdsm http api to upload the data to the image. This will
>>> a task and return a task id.
>>> 3. Use the vdsm xmlrpc/jsonrpc api to check the task status, and clear
>>> the task when done
>>> The xmlrpc/jsonrpc api is documented here:
>>> You can check the code for upload here:
>> I assume the upload will be done via a servlet on the engine, not directly
>> by the ui plugin accessing vdsm.
>> worth discussing your plans here, to make sure architecture/security are
> I was planning on using a python CGI script which would accept the
> upload via POST from the UI plugin. The file would be stored in /tmp
> on the engine host.
> After the file was successfully uploaded, the CGI script would send a
> POST to a python HTTP server (BaseHTTPServer, also running on engine
> host) with the filename and storage domain information. This python
> script would then take care of mounting the storage domain and copying
> the file to the appropriate location.
> This was my initial approach, I plan on checking out the VDSM API as well.
my preference would be to stream via a servlet to the vdsm api, rather
than "store and forward" to avoid potentially exhausting space on engine
or having to deal with two phased task tracking.
the tricky part which requires a review is validating authentication and
authorization by the servlet - to make sure one has the permission to
write to a certain disk (for data domains) / iso domain.
this should be similar to the websocket novnc approach of validating
user has access to relevant VM (but Alon may correct me if its different)
these requirements are for core product feature, while I guess Tony is after a solution to
bridge the gap of the missing basic product feature within his environment. his solution
is the most simple to solve his local problem, please assign core developer(s) to resolve
the generic problem. Expecting a contributer to apply something the product was not
designed for or to fix the entire product is incorrect.
we have long way to enable pluggable server side logic; we have no stable interface for
authentication check (although we plan to have one for 3.6), we have no stable xmlrpc
wrapper, not to mention that the xmlrpc should be obsoleted once the json rpc is out, we
do not have access to dao to obtain information about entities, we do not have stable
interface to engine key and track its usage, we do not have restapi extensions to extend
server side functionality, I guess we do not have many more, as engine was never designed
as modular extensible component.
we should also consider dropping the image upload to vdsm in favor of glance, so that core
product will support glance directly either by embedding it or by delegating into it. this
will enable us to retire the xmlrpc entirely as far as I understand, it will also enable
to implement similar functionality of what initially requested within this thread.