Re: [Opensg-users] Misbehaving clustering with OpenSG 2

[Patrick H. <pa...@in...>]

Andreas Zieringer wrote:
> Hi Patrick,
>=20
> looks like a bug.

In OpenSG or in my code?

> Running the cluster servers all the time is not a good=20
> idea.

Why is that?

> Better run a simple program (we use a minimal webserver service)=20
> which starts the cluster server. On disconnection exit the cluster serv=
er.

I could do that, but I would like to understand why adding that complexit=
y
is necessary first. Is it a resource usage issue, a design limitation of
OpenSG, or something else?

 -Patrick

>> We at Infiscape are in the process of developing an open source 3D
>> presentation tool that will be based on OpenSG 2 (see the in-developme=
nt
>> site https://realityforge.vrsource.org/trac/vpresent/ for information)=
, and
>> I have discovered that the newly ported clustering capabilities are no=
t
>> quite fully functional yet.
>>
>> Before I explain the problems, I'll provide a little background. What =
I plan
>> to do with this software is have OpenSG cluster servers running at all=
 times
>> on the render nodes (as daemons/services) so that a user of the client=
 GUI
>> can launch their presentation at any time on a cluster just as they wo=
uld do
>> with PowerPoint, OO.o Impress, or Keynote on a single machine. Early t=
esting
>> shows that this idea will work out fine--almost.
>>
>> The problem arises if I connect the client to the servers, disconnect,=
 and
>> then reconnect. The geometry shows up on the first connection, and I h=
ave
>> the windows closing correctly upon client disconnection. When I reconn=
ect
>> the client to the servers using different geometry, however, the geome=
try
>> that is displayed is the same as what was shown the first time, though=
 there
>> appear to be some adjustments to the viewport (?) that seem to account=
 for
>> changes to the volume of what is to be displayed. It seems that some
>> communication is failing to happen or that I have not migrated working=

>> OpenSG 1.8 code to OpenSG 2 code correctly.
>>
>> The code that I am testing is written in Python using PyOpenSG, and wh=
at I
>> have so far is a Python translation/adaptation of the 12ClusterServer =
and
>> 13ClusterClient tutorial applications. The ports of these that I poste=
d to
>> this list last week are what I used as a starting point, though there =
are
>> changes to my Python applications intended to target the specific need=
s that
>> I described above. The same behavior happens with the 12ClusterServer =
and
>> 13ClusterClient applications as I posted them here, though, so it isn'=
t
>> something about PyOpenSG or what I am doing with window management.
>>
>> Is there something extra that needs to be done on the server side in O=
penSG
>> 2 in order to clean up the aspect after the client disconnects? I trie=
d
>> adding a call to OSG::commitChanges() in 12ClusterServer.cpp's display=
()
>> function, but that didn't help. If there is a bug in OpenSG or a missi=
ng
>> feature, what should I be looking for in the code to track that down? =
I am
>> ready and willing to fix this and submit a patch, but I could use a ti=
p or
>> two about how to get started so that I can fix things that much faster=
=2E
>>
>> Thanks.
>>
>>  -Patrick


--=20
Patrick L. Hartling                    | VP Engineering, Infiscape Corp.
PGP: http://tinyurl.com/2msw3          | http://www.infiscape.com/